emnlp emnlp2012 emnlp2012-63 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Xin Zhao ; Baihan Shu ; Jing Jiang ; Yang Song ; Hongfei Yan ; Xiaoming Li
Abstract: Activities on social media increase at a dramatic rate. When an external event happens, there is a surge in the degree of activities related to the event. These activities may be temporally correlated with one another, but they may also capture different aspects of an event and therefore exhibit different bursty patterns. In this paper, we propose to identify event-related bursts via social media activities. We study how to correlate multiple types of activities to derive a global bursty pattern. To model smoothness of one state sequence, we propose a novel function which can capture the state context. The experiments on a large Twitter dataset shows our methods are very effective.
Reference: text
sentIndex sentText sentNum sentScore
1 gcn@ Abstract Activities on social media increase at a dramatic rate. [sent-11, score-0.359]
2 When an external event happens, there is a surge in the degree of activities related to the event. [sent-12, score-0.478]
3 These activities may be temporally correlated with one another, but they may also capture different aspects of an event and therefore exhibit different bursty patterns. [sent-13, score-0.838]
4 In this paper, we propose to identify event-related bursts via social media activities. [sent-14, score-0.654]
5 We study how to correlate multiple types of activities to derive a global bursty pattern. [sent-15, score-0.778]
6 To model smoothness of one state sequence, we propose a novel function which can capture the state context. [sent-16, score-0.411]
7 Millions of users engage in a diverse range of routine activities on social media such as posting blog messages, images, videos or status messages, as well as interacting with items generated by others such as forwarding messages. [sent-22, score-0.772]
8 When an event interesting to a certain group of individuals takes place, there is usually a surge in the degree of activities related to the event (e. [sent-23, score-0.52]
9 Since social media activities may indicate the happenings of external events, can we leverage on the rich social media activities to help identify meaningful external events? [sent-26, score-1.489]
10 Figure 1: The amount of activities within a 10-hour window for two queries. [sent-34, score-0.367]
11 Three types of activities are considered: (1) posting a tweet (upward triangle), (2) retweet (downward triangle), (3) posting a URLembedded tweet (excluding retweet) (filled circle). [sent-35, score-0.562]
12 Mining events from text streams is usually achieved by detecting bursty patterns (Swan and Allan, 2000; Kleinberg, 2003; Fung et al. [sent-37, score-0.715]
13 There is still a lack of systematic investigations into the problem of identifying eventrelated bursty patterns via social media activities. [sent-40, score-0.814]
14 First, social media involve various types of activities taking place in real time. [sent-42, score-0.677]
15 These activities may be temporally correlated with one another, but they may also capture different aspects of an event and therefore exhibit different bursty patterns. [sent-43, score-0.838]
16 The first two bursts are judged as noise since they do not correspond to any meaningful external events. [sent-63, score-0.367]
17 In fact, the reasons why a burst appears in social media can be quite diverse. [sent-64, score-0.703]
18 Figure 1 shows the change of the amount of activities of three different types over a 10-hour time window for two queries. [sent-69, score-0.367]
19 The first burst was caused by the broadcast of an advertisement from several Twitter bots, and the second burst was caused by numerous retweets of a status update of a movie star1 . [sent-72, score-0.688]
20 On the other hand, interestingly, we can see that not all the activity streams display noisy bursty patterns at the same time. [sent-74, score-0.947]
21 The intuition is that using multiple types of activities may help learn a better global picture of event-related bursty patterns. [sent-76, score-0.742]
22 Second, in social media, burst detection is challenged by irregular, unpredictable and spurious noisy bursts. [sent-78, score-0.627]
23 To overcome this challenge, a reasonable assumption is that a burst corresponding to a real event should not fluctuate too much within a relatively short time window. [sent-79, score-0.424]
24 To illustrate it, we present an example in Figure 2, in which we first use a simple threshold method to detect bursts and then analyze the effect of local smoothness. [sent-80, score-0.365]
25 In particular, if the amount of activities at a certain time is above a pre-defined threshold, we set its state to 1, which indicates a bursty state. [sent-81, score-0.808]
26 It shows two examples of threshold methods for burst detection in a 10-hour window. [sent-88, score-0.438]
27 If the number of activities is above the threshold in one time interval, the state of this time interval is judge as bursty. [sent-90, score-0.506]
28 In contrast, in Figure 2(b), the state sequence for the query “Nobel” is “000001 1111,” in which the longer and smoother burst corresponds to a true event. [sent-94, score-0.53]
29 A good function for evaluating the smoothness of a state sequence should be able to discriminate these cases and model the context of state sequences effectively. [sent-95, score-0.439]
30 With its unique characteristics and challenges, there is an emergent need to deeply study the problem of event-related burst detection via social media activities. [sent-96, score-0.772]
31 We formulate this problem as burst detection from time series of social media activities. [sent-98, score-0.772]
32 We develop an optimization model to learn bursty patterns based on multiple types of activities. [sent-99, score-0.418]
33 We propose to detect bursts by considering both local state smoothness and correlation across multiple streams. [sent-100, score-0.691]
34 We systematically evaluate three types of activities for burst detection on a large Twitter dataset and analyze different properties of these three streams for burst detection. [sent-102, score-1.34]
35 Activity Stream: An activity stream of length N and type m is a sequence of numbers (n1m, n2m, . [sent-105, score-0.436]
36 , nNm), where each nim denotes the amount of activities of type m that occur during the ith time interval. [sent-108, score-0.38]
37 ” Event-related Burst: Given a query Q, an eventrelated burst is defined as a period [ts, te] in which some event related with Q takes place, where ts and te are the start timestamp and end timestamp of the event period respectively. [sent-114, score-0.765]
38 During the event period the amount of activities is significantly higher than average. [sent-115, score-0.432]
39 Based on these definitions, our task is to try to identify event-related bursts via multiple social media activity streams. [sent-116, score-0.892]
40 3 Identifying Event-related Bursts from Social Media In this section, we discuss how to identify eventrelated bursts via social media activities. [sent-117, score-0.691]
41 Formally, given a query Q, we first build M activity streams related with Q on T timestamps: {(n1m, . [sent-118, score-0.561]
42 The definition of activity in our {m(enthods is very general; it includes various types of social media activities, including textual and nontextual activities, e. [sent-122, score-0.597]
43 We use the Poisson distribution to study the probability ×× of observing the number of social media activities, and we treat one hour as one time interval in this paper. [sent-134, score-0.447]
44 Homogeneous Poisson Distribution The generative probability of the ith number in one activity stream of type m is defined as f(nim, i,zim) = (λzmi)nimneimxp! [sent-135, score-0.408]
45 If one state is bursty, it would emit activities with a faster rate and result in a larger expectation λ1. [sent-137, score-0.418]
46 two-state machine in (Kleinberg, 2003) used two global references for all the time intervals: one for bursty and the other for non-bursty. [sent-140, score-0.424]
47 3 Gseitv oefn h ao tuirm-sep iencdifeicx hra,t ewse { λset λ0,h toa n bed {thλe expectation of the number of activities in hth time interval every day, then we have λ1,h = λ0,h ρ. [sent-147, score-0.381]
48 2 Smoothness of a State Sequence For burst detection, the major aim is to identify steady and meaningful bursts and to discard transient and spurious bursts. [sent-152, score-0.673]
49 − Kleinberg (2003) uses state transition probabilities to model the smoothness of state sequences. [sent-177, score-0.411]
50 In social media, very short noisy bursts like “0000100000” are very frequent. [sent-179, score-0.509]
51 To better measure the smoothness of a state sequence , we propose a novel context-sensitive function, which sums the square ofthe length ofthe maximum subsequences in which all states are the same. [sent-183, score-0.394]
52 Instead of examining the smoothness ofthe whole state sequence, we propose to measure the smoothness of all the Llength subsequences, so called “local smoothness”. [sent-193, score-0.522]
53 3 Correlating Multiple Activity Streams 1 19023end endbm0[i=n](sd b[ i]0<[s←<][b−10]|,sd)&[i(−1 <][s Lb]−+c1)[;i]+c2[i]); In this section, we discuss how to correlate multiple activity streams to learn a global bursty patterns. [sent-219, score-0.927]
54 The hidden state sequences corresponding to these activity streams are not fully independent. [sent-220, score-0.603]
55 An external event may intricate surges in multiple activity streams simultaneously. [sent-221, score-0.621]
56 We propose to correlate multiple activity streams in an optimization model. [sent-222, score-0.503]
57 The idea is that activity streams related with one query might be dependent, i. [sent-223, score-0.561]
58 , the states of multiple activity streams on the same timestamp tend to be the same6; if not, it would incur a cost. [sent-225, score-0.607]
59 For convenience, we call the states of each activity stream as “local × states” while the overall states learnt from multiple activity streams as “global states”. [sent-227, score-1.021]
60 The idea is that although various activity streams are different in the scale of frequencies, they tend to share similar trend patterns. [sent-228, score-0.503]
61 Here we zmT)}mM=1 (zG1, zTG) give a few options: CONJUNCT: we set a global state zi as bursty if all local states are bursty, i. [sent-267, score-0.696]
62 DISJUNCT: we set a global st=ate ∩ zi as bursty if one of the local states is bursty, i. [sent-270, score-0.596]
63 For each query, we invite two senior graduate students to manually identify their golden bursty intervals, and each bursty interval is represented as a pair of timestamps in terms of hours. [sent-287, score-0.959]
64 2 Evaluation Metrics Before introducing our evaluation metrics, we first define the Bursty Interval Overlap Ratio (BIOR) BIOR(f,X) = Pf0∈XL∆(fl)(f,f0), f is a bursty interval, ∆l(f, f0) is the length of overlap between f0 and f, L(f) is the length of 7We refer to some gold news resources, e. [sent-295, score-0.39]
65 X0, X1 and X2 are three sets of bursty intervals. [sent-302, score-0.39]
66 X is a set of bursty intervals, BbuIrOsRty m peeariosudre osf ft fhe. [sent-308, score-0.39]
67 p Xrop isor taio sne to fo fth beu rtismtyes intatmerpvas lisn, f which are covered by one of bursty intervals in X. [sent-310, score-0.431]
68 s W, beec uasues Be aO system may en pota rrtietaulr mn aatcl h hth oef einxtearc-t bursty intervals9. [sent-312, score-0.39]
69 Given two sets of bursty intervals which have the same value of BIOR, we prefer the one with fewer intervals. [sent-321, score-0.431]
70 In Figure 3, we can easily derive X1 and X2 have the same value 9A simple evaluation method is that we label each one hour time slot as being part of a burst or not and compare with the gold standard. [sent-322, score-0.405]
71 However, in our experiments, we find that some methods tend to break one meaningful burst into small parts and easier to be affected by small fluctuations although they may have a good coverage of bursty points. [sent-323, score-0.768]
72 We compute the average cross correlation between different activity streams for these 17 queries in our test collection, and we summarize the results in Table 3. [sent-338, score-0.584]
73 SimpleDisjunct: we first find the optimal state sequences for each single activity stream, and then we derive a global state sequence by take the disjunction of all local states. [sent-358, score-0.618]
74 Note that, our method in Equation 4 has two merits: 1) the length of local window can be tuned on different datasets; 2) a novel state smoothness function is adopted. [sent-364, score-0.405]
75 These preliminary findings indicate that state smoothness is very important for burst detection, and the length of state context window will affect the performance significantly. [sent-375, score-0.833]
76 Since for each query, we can obtain multiple results in different activity streams, we further categorize the 17 annotated queries to the stream which leads to the optimal performance on that query. [sent-377, score-0.449]
77 Combining previous analysis of Table 5, overall we find the retweet stream is more capable to identify bursts which correspond to significant events. [sent-382, score-0.507]
78 3 1370 4 251 Preliminary results on multiple streams After examining the basic results on a single stream, we continue to evaluate the performance of our proposed models on multiple activity streams. [sent-410, score-0.503]
79 We can see that MBurst has four candidate methods to derive global states from local states; for L2G, we use the states of St as the final states, and we empwieric uaselly th fein sdta attheast oitf pSerforms best compared with the other two streams in L2G. [sent-413, score-0.49]
80 For SimpleConjunct and SimpleDisj,uLnc =t, 4w)e fnirs St find the optimal state sequences for each single activity stream using our proposed method in Equation 4 (Φ = g2, L = 4), and then we derive a global state sequence by take the conjunction or disjunction of all local states respectively. [sent-415, score-0.843]
81 For different methods to derive global bursty patterns, we can see that MBurst+DISJUNCT performs best while MBurst+CONJUNCT performs worst. [sent-436, score-0.46]
82 Interestingly, however, SimpleConjunct is better than SimpleDisjunct, the major reason is that MBurst performs a local-state correlation of multiple activity streams to correct possible noisy fluctuations from single streams before the conjunction or disjunction of local states. [sent-437, score-0.916]
83 After such correlation, the performance of each activity stream should improve. [sent-438, score-0.408]
84 Indeed, we find that the step of local-state correlation is more important for our multiple stream algorithm than the step of how to derive global states based on local states. [sent-441, score-0.38]
85 5 Related Work Our work is related to burst detection from text streams. [sent-462, score-0.413]
86 Pioneered by the automaton model proposed in (Kleinberg, 2003), many techniques have been proposed for burst detection such as the χ2test based method (Swan and Allan, 2000), the parameter-free method (Fung et al. [sent-463, score-0.413]
87 Our work is related to the applications of these burst detection algorithms for event detection (He et al. [sent-466, score-0.562]
88 However, few of these methods consider modeling the local smoothness of one state sequence in a systematic way and often use a fixed window length of 2. [sent-480, score-0.433]
89 Little work considers making use of different types of social media activities for burst detection. [sent-481, score-1.021]
90 , 2009) conducted some preliminary studies of mining correlated bursty patterns from multiple sources. [sent-486, score-0.504]
91 Online social activities are dynamic, with a large number of new items generated continuously. [sent-488, score-0.506]
92 In such a dynamic setting, burst detection algorithms should effectively collect evidence, efficiently adjust prediction models and respond to the users as social media activities evolve. [sent-489, score-1.128]
93 6 Conclusion In this paper, we propose to identify event-related bursts via social media activities. [sent-502, score-0.654]
94 We propose one optimization model to correlate multiple activity streams to learn the bursty patterns. [sent-503, score-0.893]
95 To better measure local smoothness of the state sequence, we propose a novel state cost function. [sent-504, score-0.491]
96 Our work can provide a preliminary understanding of the correlation between the happenings of events and the degree of online social media activities. [sent-507, score-0.485]
97 Earthquake shakes twitter users: real-time event detection by social sensors. [sent-619, score-0.431]
98 Mining correlated bursty topic patterns from coordinated text streams. [sent-641, score-0.443]
99 Temporal and social context based burst detection from folksonomies. [sent-649, score-0.601]
100 Context modeling for ranking and tagging bursty features in text Wayne streams. [sent-653, score-0.39]
wordName wordTfidf (topN-words)
[('bursty', 0.39), ('burst', 0.344), ('activities', 0.318), ('bursts', 0.295), ('streams', 0.265), ('activity', 0.238), ('smoothness', 0.211), ('social', 0.188), ('mburst', 0.172), ('media', 0.171), ('stream', 0.17), ('bior', 0.123), ('state', 0.1), ('kleinberg', 0.097), ('twitter', 0.094), ('event', 0.08), ('zim', 0.074), ('zi', 0.072), ('detection', 0.069), ('fung', 0.066), ('tweets', 0.064), ('interval', 0.063), ('query', 0.058), ('poisson', 0.057), ('posting', 0.057), ('sr', 0.057), ('states', 0.055), ('mathioudakis', 0.049), ('simpleconjunct', 0.049), ('timestamp', 0.049), ('timestamps', 0.049), ('window', 0.049), ('xiaoming', 0.048), ('local', 0.045), ('tweet', 0.044), ('xin', 0.043), ('golden', 0.042), ('swan', 0.042), ('retweet', 0.042), ('surge', 0.042), ('ix', 0.041), ('intervals', 0.041), ('judges', 0.041), ('queries', 0.041), ('correlation', 0.04), ('equation', 0.04), ('hongfei', 0.038), ('external', 0.038), ('users', 0.038), ('cheong', 0.037), ('disjunct', 0.037), ('disjunction', 0.037), ('eclipse', 0.037), ('eventrelated', 0.037), ('nim', 0.037), ('nobel', 0.037), ('pui', 0.037), ('simpledisjunct', 0.037), ('statemachine', 0.037), ('zig', 0.037), ('derive', 0.036), ('cost', 0.035), ('st', 0.034), ('meaningful', 0.034), ('period', 0.034), ('global', 0.034), ('wayne', 0.033), ('binomial', 0.033), ('events', 0.032), ('mining', 0.032), ('sakaki', 0.032), ('shan', 0.032), ('mm', 0.031), ('yan', 0.031), ('xt', 0.03), ('preliminary', 0.029), ('su', 0.029), ('gabriel', 0.029), ('zhao', 0.029), ('sequence', 0.028), ('patterns', 0.028), ('wsdm', 0.027), ('noisy', 0.026), ('yao', 0.026), ('correlated', 0.025), ('threshold', 0.025), ('temporally', 0.025), ('denotes', 0.025), ('aramki', 0.025), ('aristides', 0.025), ('baihan', 0.025), ('budak', 0.025), ('conjunct', 0.025), ('gionis', 0.025), ('happenings', 0.025), ('hour', 0.025), ('huan', 0.025), ('ihler', 0.025), ('invite', 0.025), ('junjie', 0.025)]
simIndex simValue paperId paperTitle
same-paper 1 1.000001 63 emnlp-2012-Identifying Event-related Bursts via Social Media Activities
Author: Xin Zhao ; Baihan Shu ; Jing Jiang ; Yang Song ; Hongfei Yan ; Xiaoming Li
Abstract: Activities on social media increase at a dramatic rate. When an external event happens, there is a surge in the degree of activities related to the event. These activities may be temporally correlated with one another, but they may also capture different aspects of an event and therefore exhibit different bursty patterns. In this paper, we propose to identify event-related bursts via social media activities. We study how to correlate multiple types of activities to derive a global bursty pattern. To model smoothness of one state sequence, we propose a novel function which can capture the state context. The experiments on a large Twitter dataset shows our methods are very effective.
2 0.12718569 120 emnlp-2012-Streaming Analysis of Discourse Participants
Author: Benjamin Van Durme
Abstract: Inferring attributes of discourse participants has been treated as a batch-processing task: data such as all tweets from a given author are gathered in bulk, processed, analyzed for a particular feature, then reported as a result of academic interest. Given the sources and scale of material used in these efforts, along with potential use cases of such analytic tools, discourse analysis should be reconsidered as a streaming challenge. We show that under certain common formulations, the batchprocessing analytic framework can be decomposed into a sequential series of updates, using as an example the task of gender classification. Once in a streaming framework, and motivated by large data sets generated by social media services, we present novel results in approximate counting, showing its applicability to space efficient streaming classification.
3 0.092849538 72 emnlp-2012-Joint Inference for Event Timeline Construction
Author: Quang Do ; Wei Lu ; Dan Roth
Abstract: This paper addresses the task of constructing a timeline of events mentioned in a given text. To accomplish that, we present a novel representation of the temporal structure of a news article based on time intervals. We then present an algorithmic approach that jointly optimizes the temporal structure by coupling local classifiers that predict associations and temporal relations between pairs of temporal entities with global constraints. Moreover, we present ways to leverage knowledge provided by event coreference to further improve the system performance. Overall, our experiments show that the joint inference model significantly outperformed the local classifiers by 9.2% of relative improvement in F1. The experiments also suggest that good event coreference could make remarkable contribution to a robust event timeline construction system.
4 0.077861428 32 emnlp-2012-Detecting Subgroups in Online Discussions by Modeling Positive and Negative Relations among Participants
Author: Ahmed Hassan ; Amjad Abu-Jbara ; Dragomir Radev
Abstract: A mixture of positive (friendly) and negative (antagonistic) relations exist among users in most social media applications. However, many such applications do not allow users to explicitly express the polarity of their interactions. As a result most research has either ignored negative links or was limited to the few domains where such relations are explicitly expressed (e.g. Epinions trust/distrust). We study text exchanged between users in online communities. We find that the polarity of the links between users can be predicted with high accuracy given the text they exchange. This allows us to build a signed network representation of discussions; where every edge has a sign: positive to denote a friendly relation, or negative to denote an antagonistic relation. We also connect our analysis to social psychology theories of balance. We show that the automatically predicted networks are consistent with those theories. Inspired by that, we present a technique for identifying subgroups in discussions by partitioning singed networks representing them.
5 0.064480633 134 emnlp-2012-User Demographics and Language in an Implicit Social Network
Author: Katja Filippova
Abstract: We consider the task of predicting the gender of the YouTube1 users and contrast two information sources: the comments they leave and the social environment induced from the affiliation graph of users and videos. We propagate gender information through the videos and show that a user’s gender can be predicted from her social environment with the accuracy above 90%. We also show that the gender can be predicted from language alone (89%). A surprising result of our study is that the latter predictions correlate more strongly with the gender predominant in the user’s environment than with the sex of the person as reported in the profile. We also investigate how the two views (linguistic and social) can be combined and analyse how prediction accuracy changes over different age groups.
6 0.052656744 121 emnlp-2012-Supervised Text-based Geolocation Using Language Models on an Adaptive Grid
7 0.046356518 89 emnlp-2012-Mixed Membership Markov Models for Unsupervised Conversation Modeling
8 0.044388607 5 emnlp-2012-A Discriminative Model for Query Spelling Correction with Latent Structural SVM
9 0.042067789 22 emnlp-2012-Automatically Constructing a Normalisation Dictionary for Microblogs
10 0.041579276 130 emnlp-2012-Unambiguity Regularization for Unsupervised Learning of Probabilistic Grammars
11 0.040367927 71 emnlp-2012-Joint Entity and Event Coreference Resolution across Documents
12 0.039543003 35 emnlp-2012-Document-Wide Decoding for Phrase-Based Statistical Machine Translation
13 0.036697224 78 emnlp-2012-Learning Lexicon Models from Search Logs for Query Expansion
14 0.033946604 97 emnlp-2012-Natural Language Questions for the Web of Data
15 0.03385932 114 emnlp-2012-Revisiting the Predictability of Language: Response Completion in Social Media
16 0.033392727 8 emnlp-2012-A Phrase-Discovering Topic Model Using Hierarchical Pitman-Yor Processes
17 0.032226365 93 emnlp-2012-Multi-instance Multi-label Learning for Relation Extraction
18 0.032122344 33 emnlp-2012-Discovering Diverse and Salient Threads in Document Collections
19 0.03148336 37 emnlp-2012-Dynamic Programming for Higher Order Parsing of Gap-Minding Trees
20 0.031407822 86 emnlp-2012-Locally Training the Log-Linear Model for SMT
topicId topicWeight
[(0, 0.127), (1, 0.053), (2, -0.002), (3, 0.024), (4, -0.043), (5, -0.028), (6, 0.006), (7, -0.052), (8, 0.062), (9, 0.062), (10, 0.04), (11, -0.158), (12, -0.094), (13, -0.041), (14, -0.074), (15, 0.045), (16, -0.064), (17, -0.041), (18, 0.032), (19, -0.123), (20, -0.212), (21, -0.049), (22, 0.124), (23, -0.039), (24, 0.141), (25, 0.139), (26, -0.134), (27, 0.163), (28, -0.054), (29, 0.136), (30, 0.111), (31, 0.139), (32, 0.126), (33, 0.005), (34, 0.057), (35, 0.04), (36, -0.126), (37, -0.002), (38, -0.01), (39, -0.122), (40, -0.072), (41, 0.041), (42, 0.101), (43, -0.169), (44, 0.126), (45, -0.084), (46, -0.072), (47, 0.219), (48, 0.007), (49, -0.144)]
simIndex simValue paperId paperTitle
same-paper 1 0.95994437 63 emnlp-2012-Identifying Event-related Bursts via Social Media Activities
Author: Xin Zhao ; Baihan Shu ; Jing Jiang ; Yang Song ; Hongfei Yan ; Xiaoming Li
Abstract: Activities on social media increase at a dramatic rate. When an external event happens, there is a surge in the degree of activities related to the event. These activities may be temporally correlated with one another, but they may also capture different aspects of an event and therefore exhibit different bursty patterns. In this paper, we propose to identify event-related bursts via social media activities. We study how to correlate multiple types of activities to derive a global bursty pattern. To model smoothness of one state sequence, we propose a novel function which can capture the state context. The experiments on a large Twitter dataset shows our methods are very effective.
2 0.60529733 120 emnlp-2012-Streaming Analysis of Discourse Participants
Author: Benjamin Van Durme
Abstract: Inferring attributes of discourse participants has been treated as a batch-processing task: data such as all tweets from a given author are gathered in bulk, processed, analyzed for a particular feature, then reported as a result of academic interest. Given the sources and scale of material used in these efforts, along with potential use cases of such analytic tools, discourse analysis should be reconsidered as a streaming challenge. We show that under certain common formulations, the batchprocessing analytic framework can be decomposed into a sequential series of updates, using as an example the task of gender classification. Once in a streaming framework, and motivated by large data sets generated by social media services, we present novel results in approximate counting, showing its applicability to space efficient streaming classification.
3 0.46833196 134 emnlp-2012-User Demographics and Language in an Implicit Social Network
Author: Katja Filippova
Abstract: We consider the task of predicting the gender of the YouTube1 users and contrast two information sources: the comments they leave and the social environment induced from the affiliation graph of users and videos. We propagate gender information through the videos and show that a user’s gender can be predicted from her social environment with the accuracy above 90%. We also show that the gender can be predicted from language alone (89%). A surprising result of our study is that the latter predictions correlate more strongly with the gender predominant in the user’s environment than with the sex of the person as reported in the profile. We also investigate how the two views (linguistic and social) can be combined and analyse how prediction accuracy changes over different age groups.
4 0.42660716 22 emnlp-2012-Automatically Constructing a Normalisation Dictionary for Microblogs
Author: Bo Han ; Paul Cook ; Timothy Baldwin
Abstract: Microblog normalisation methods often utilise complex models and struggle to differentiate between correctly-spelled unknown words and lexical variants of known words. In this paper, we propose a method for constructing a dictionary of lexical variants of known words that facilitates lexical normalisation via simple string substitution (e.g. tomorrow for tmrw). We use context information to generate possible variant and normalisation pairs and then rank these by string similarity. Highlyranked pairs are selected to populate the dictionary. We show that a dictionary-based approach achieves state-of-the-art performance for both F-score and word error rate on a standard dataset. Compared with other methods, this approach offers a fast, lightweight and easy-to-use solution, and is thus suitable for high-volume microblog pre-processing. 1 Lexical Normalisation A staggering number of short text “microblog” messages are produced every day through social media such as Twitter (Twitter, 2011). The immense volume of real-time, user-generated microblogs that flows through sites has been shown to have utility in applications such as disaster detection (Sakaki et al., 2010), sentiment analysis (Jiang et al., 2011; Gonz a´lez-Ib ´a n˜ez et al., 2011), and event discovery (Weng and Lee, 2011; Benson et al., 2011). However, due to the spontaneous nature of the posts, microblogs are notoriously noisy, containing many non-standard forms e.g., tmrw “tomorrow” and 2day “today” which degrade the performance of — — 421 natural language processing (NLP) tools (Ritter et al., 2010; Han and Baldwin, 2011). To reduce this effect, attempts have been made to adapt NLP tools to microblog data (Gimpel et al., 2011; Foster et al., 2011; Liu et al., 2011b; Ritter et al., 2011). An alternative approach is to pre-normalise non-standard lexical variants to their standard orthography (Liu et al., 2011a; Han and Baldwin, 2011; Xue et al., 2011; Gouws et al., 2011). For example, se u 2morw!!! would be normalised to see you tomorrow! The normalisation approach is especially attractive as a preprocessing step for applications which rely on keyword match or word frequency statistics. For example, earthqu, eathquake, and earthquakeee all attested in a Twitter corpus have the standard form earthquake; by normalising these types to their standard form, better coverage can be achieved for keyword-based methods, and better word frequency estimates can be obtained. In this paper, we focus on the task of lexical normalisation of English Twitter messages, in which out-of-vocabulary (OOV) tokens are normalised to their in-vocabulary (IV) standard form, i.e., a standard form that is in a dictionary. Following other recent work on lexical normalisation (Liu et al., 2011a; Han and Baldwin, 2011; Gouws et al., 2011; Liu et al., 2012), we specifically focus on one-to-one normalisation in which one OOV token is normalised to one IV word. Naturally, not all OOV words in microblogs are lexical variants of IV words: named entities, e.g., — — are prevalent in microblogs, but not all named entities are included in our dictionary. One challenge for lexical normalisation is therefore to disPLraoncge uadgineg Lse oafr tnhineg 2,0 p1a2g Jeosin 42t C1–o4n3f2e,re Jnecjue Iosnla Enmd,p Kiroicraela, M 1e2t–h1o4ds Ju ilny N 20a1tu2r.a ?lc L2a0n1g2ua Agseso Pcrioactieosnsi fnogr a Cnodm Cpoumtaptiuotna tilo Lnianlg Nuaist uircasl tinguish those OOV tokens that require normalisation from those that are well-formed. Recent unsupervised approaches have not attempted to distinguish such tokens from other types of OOV tokens (Cook and Stevenson, 2009; Liu et al., 2011a), limiting their applicability to real-world normalisation tasks. Other approaches (Han and Baldwin, 2011; Gouws et al., 2011) have followed a cascaded approach in which lexical variants are first identified, and then normalised. However, such two-step approaches suffer from poor lexical variant identification performance, which is propagated to the normalisation step. Motivated by the observation that most lexical variants have an unambiguous standard form (especially for longer tokens), and that a lexical variant and its standard form typically occur in similar contexts, in this paper we propose methods for automatically constructing a lexical normalisation dictionary a dictionary whose entries consist — of (lexical variant, standard form) pairs that enables type-based normalisation. Despite the simplicity of this dictionary-based normalisation method, we show it to outperform previously-proposed approaches. This very fast, lightweight solution is suitable for real-time processing of the large volume of streaming microblog data available from Twitter, and offers a simple solution to the lexical variant detection problem that hinders other normalisation methods. Furthermore, this dictionary-based method can be easily integrated with other more-complex normalisation approaches (Liu et al., 2011a; Han and Baldwin, 2011; Gouws et al., 2011) to produce hybrid systems. After discussing related work in Section 2, we present an overview of our dictionary-based approach to normalisation in Section 3. In Sections 4 and 5 we experimentally select the optimised context similarity parameters and string similarity reranking method. We present experimental results on the unseen test data in Section 6, and offer some concluding remarks in Section 7. — 2 Related Work Given a token t, lexical normalisation is the task of finding arg max P(s|t) ∝ arg max P(t| s)P(s), wofh efinred s igs tahreg smtaanxdaPrd(s form, i.e., an aIVx Pw(otr|sd). PSt(asn)-, dardly in lexical normalisation, t is assumed to be an 422 OOV token, relative to a fixed dictionary. In practice, not all OOV tokens should be normalised; i.e., only lexical variants (e.g., tmrw “tomorrow”) should be normalised and tokens that are OOV but otherwise not lexical variants (e.g., iPad “iPad”) should be unchanged. Most work in this area focuses only on the normalisation task itself, oftentimes assuming that the task of lexical variant detection has already been completed. Various approaches have been proposed to estimate the error model, P(t|s). For example, in work on spell-checking, eBl,ril Pl (atn|ds) M. Fooorre e (2000) improve on a standard edit-distance approach by considering multi-character edit operations; Toutanova and Moore (2002) build on this by incorporating phonological information. Li et al. (2006) utilise distributional similarity (Lin, 1998) to correct misspelled search queries. In text message normalisation, Choudhury et al. (2007) model the letter transformations and emissions using a hidden Markov model (Rabiner, 1989). Cook and Stevenson (2009) and Xue et al. (201 1) propose multiple simple error models, each of which captures a particular way in which lexical variants are formed, such as phonetic spelling (e.g., epik “epic”) or clipping (e.g., walkin “walking”). Nevertheless, optimally weighting the various error models in these approaches is challenging. Without pre-categorising lexical variants into different types, Liu et al. (201 1a) collect Google search snippets from carefully-designed queries from which they then extract noisy lexical variant– standard form pairs. These pairs are used to train a conditional random field (Lafferty et al., 2001) to estimate P(t|s) at the character level. One shortcoming eo fP querying a ese cahracrha engine teol. .o Obtanein strhaoirnt-ing pairs is it tends to be costly in terms of time and bandwidth. Here we exploit microblog data directly to derive (lexical variant, standard form) pairs, instead of relying on external resources. In morerecent work, Liu et al. (2012) endeavour to improve the accuracy of top-n normalisation candidates by integrating human cognitive inference, characterlevel transformations and spell checking in their normalisation model. The encouraging results shift the focus to reranking and promoting the correct normalisation to the top-1 position. However, like much previous work on lexical normalisation, this work assumes perfect lexical variant detection. Aw et al. (2006) and Kaufmann and Kalita (2010) consider normalisation as a machine translation task from lexical variants to standard forms using off-theshelf tools. These methods do not assume that lexical variants have been pre-identified; however, these methods do rely on large quantities of labelled training data, which is not available for microblogs. Recently, Han and Baldwin (201 1) and Gouws et al. (201 1) propose two-step unsupervised approaches to normalisation, in which lexical variants are first identified, and then normalised. They approach lexical variant detection by using a context fitness classifier (Han and Baldwin, 2011) or through dictionary lookup (Gouws et al., 2011). However, the lexical variant detection of both meth- ods is rather unreliable, indicating the challenge of this aspect of normalisation. Both of these approaches incorporate a relatively small normalisation dictionary to capture frequent lexical variants with high precision. In particular, Gouws et al. (201 1) produce a small normalisation lexicon based on distributional similarity and string similarity (Lodhi et al., 2002). Our method adopts a similar strategy using distributional/string similarity, but instead of constructing a small lexicon for preprocessing, we build a much wider-coverage normalisation dictionary and opt for a fully lexiconbased end-to-end normalisation approach. In contrast to the normalisation dictionaries of Han and Baldwin (201 1) and Gouws et al. (201 1) which focus on very frequent lexical variants, we focus on moderate frequency lexical variants of a minimum character length, which tend to have unambiguous standard forms; our intention is to produce normalisation lexicons that are complementary to those currently available. Furthermore, we investigate the impact of a variety of contextual and string similarity measures on the quality of the resulting lexicons. In summary, our dictionary-based normalisation ap- proach is a lightweight end-to-end method which performs both lexical variant detection and normalisation, and thus is suitable for practical online preprocessing, despite its simplicity. 423 3 A Lexical Normalisation Dictionary Before discussing our method for creating a normalisation dictionary, we first discuss the feasibility of such an approach. 3.1 Feasibility Dictionary lookup approaches to normalisation have been shown to have high precision but low recall (Han and Baldwin, 2011; Gouws et al., 2011). Frequent (lexical variant, standard form) pairs such as (u, you) are typically included in the dictionaries used by such methods, while less-frequent items such as (g0tta, gotta) are generally omitted. Because of the degree of lexical creativity and large number of non-standard forms observed on Twitter, a wide-coverage normalisation dictionary would be expensive to construct manually. Based on the assumption that lexical variants occur in similar con- texts to their standard forms, however, it should be possible to automatically construct a normalisation dictionary with wider coverage than is currently available. Dictionary lookup is a type-based approach to normalisation, i.e., every token instance of a given type will always be normalised in the same way. However, lexical variants can be ambiguous, e.g., y corresponds to “you” in yeah, y r right! LOL but “why” in AM CONFUSED!!! y you did that? Nevertheless, the relative occurrence of ambiguous lexical variants is small (Liu et al., 2011a), and it has been observed that while shorter variants such as y are often ambiguous, longer variants tend to be unambiguous. For example bthday and 4eva are unlikely to have standard forms other than “birthday” and “forever”, respectively. Therefore, the normalisation lexicons we produce will only contain entries for OOVs with character length greater than a specified threshold, which are likely to have an unambiguous standard form. 3.2 Overview of approach Our method for constructing a normalisation dictio- nary is as follows: Input: Tokenised English tweets 1. Extract (OOV, IV) pairs based on distributional similarity. 2. Re-rank the extracted pairs by string similarity. Output: A list of (OOV, IV) pairs ordered by string similarity; select the top-n pairs for inclusion in the normalisation lexicon. In Step 1, we leverage large volumes of Twitter data to identify the most distributionally-similar IV type for each OOV type. The result of this process is a set of (OOV, IV) pairs, ranked by distributional similarity. The extracted pairs will include (lexical variant, standard form) pairs, such as (tmrw, tomorrow), but will also contain false positives such as (Tusday, Sunday) Tusday is a lexical variant, but its standard form is not “Sunday” and (Youtube, web) Youtube is an OOV named entity, not a lexical variant. Nevertheless, lexical variants are typically formed from their standard forms through regular processes (Thurlow, 2003) e.g., the omission of characters and from this perspective Sunday and web are not plausible standard — — — — — forms for Tusday and Youtube, respectively. In Step 2, we therefore capture this intuition to re-rank the extracted pairs by string similarity. The top-n items in this re-ranked list then form the normalisation lexicon, which is based only on development data. Although computationally-expensive to build, this dictionary can be created offline. Once built, it then offers a very fast approach to normalisation. We can only reliably compute distributional similarity for types that are moderately frequent in a corpus. Nevertheless, many lexical variants are sufficiently frequent to be able to compute distributional similarity, and can potentially make their way into our normalisation lexicon. This approach is not suitable for normalising low-frequency lexical variants, nor is it suitable for shorter lexical variant types which as discussed in Section 3.1 are more likely to have an ambiguous standard form. Nevertheless, previously-proposed normalisation methods that can handle such phenomena also rely in part on a normalisation lexicon. The normalisation lexicons we create can therefore be easily integrated with previous approaches to form hybrid normalisation systems. — — 4 Contextually-similar Pair Generation Our objective is to extract contextually-similar (OOV, IV) pairs from a large-scale collection of mi424 croblog data. Fundamentally, the surrounding words define the primary context, but there are different ways of representing context and different similarity measures we can use, which may influence the quality of generated normalisation pairs. In representing the context, we experimentally explore the following factors: (1) context window size (from 1 to 3 tokens on both sides); (2) n-gram order ofthe context tokens (unigram, bigram, trigram); (3) whether context words are indexed for relative position or not; and (4) whether we use all context tokens, or only IV words. Because high-accuracy linguistic processing tools for Twitter are still under exploration (Liu et al., 2011b; Gimpel et al., 2011; Ritter et al., 2011; Foster et al., 2011), we do not consider richer representations of context, for example, incorporating information about part-of-speech tags or syntax. We also experiment with a number of simple but widely-used geometric and information theoretic distance/similarity measures. In particular, we use Kullback–Leibler (KL) divergence (Kullback and Leibler, 195 1), Jensen–Shannon (JS) divergence (Lin, 1991), Euclidean distance and Cosine distance. We use a corpus of 10 million English tweets to do parameter tuning over, and a larger corpus of tweets in the final candidate ranking. All tweets were collected from September 2010 to January 2011 via the Twitter API.1 From the raw data we extract English tweets using a language identification tool (Lui and Baldwin, 2011), and then apply a simplified Twitter tokeniser (adapted from O’Connor et al. (2010)). We use the Aspell dictionary (v6.06)2 to determine whether a word is IV, and only include in our normalisation dictionary OOV tokens with at least 64 occurrences in the corpus and character length ≥ 4, both of which were determined through empirical 4o,b bsoetrhva otifo wnh. Fcohr w weearceh d OetOeVrm winoedrd t type ginh the corpus, we select the most similar IV type to form (OOV, IV) pairs. To further narrow the search space, we only consider IV words which are morphophonemically similar to the OOV type, follow- ing settings in Han and Baldwin (201 1).3 1http s : / / dev .twitter . com/ docs / st reaming-api /methods 2http : / / aspe l .net / l 3We only consider IV words within an edit distance of 2 or a phonemic edit distance of 1from the OOV type, and we further In order to evaluate the generated pairs, we randomly selected 1000 OOV words from the 10 million tweet corpus. We set up an annotation task on Amazon Mechanical Turk,4 presenting five independent annotators with each word type (with no context) and asking for corrections where appropriate. For instance, given tmrw, the annotators would likely identify it as a non-standard variant of “tomorrow”. For correct OOV words like iPad, on the other hand, we would expect them to leave the word unchanged. If 3 or more of the 5 annotators make the same suggestion (in the form of either a canonical spelling or leaving the word unchanged), we include this in our gold standard for evaluation. In total, this resulted in 351 lexical variants and 282 correct OOV words, accounting for 63.3% of the 1000 OOV words. These 633 OOV words were used as (OOV, IV) pairs for parameter tuning. The remainder of the 1000 OOV words were ignored on the grounds that there was not sufficient consensus amongst the annotators.5 Contextually-similar pair generation aims to include as many correct normalisation pairs as possible. We evaluate the quality of the normalisation pairs using “Cumulative Gain” (CG): XN0 CG = Xreli0 Xi=1 Suppose there are N0 correct generated pairs (oovi, ivi), each of which is weighted by reli0, the frequency of oovi to indicate its relative importance; for example, (thinkin, thinking) has a higher weight than (g0tta, gotta) because thinkin is more frequent than g0tta in our corpus. In this evaluation we don’t consider the position of normalisation pairs, and nor do we penalise incorrect pairs. Instead, we push distinguishing between correct and incorrect pairs into the downstream re-ranking step in which we incorporate string similarity information. Given the development data and CG, we run an exhaustive search of parameter combinations over only consider the top 30% most-frequent of these IV words. 4https : / /www .mturk .com/mturk/welcome 5Note that the objective of this annotation task is to identify lexical variants that have agreed-upon standard forms irrespective of context, as a special case of the more general task of lexical normalisation (where context may or may not play a significant role in the determination of the normalisation). 425 our development corpus. The five best parameter combinations are shown in Table 1. We notice the CG is almost identical for the top combinations. As a context window size of 3 incurs a heavy processing and memory overhead over a size of 2, we use the 3rd-best parameter combination for subsequent experiments, namely: context window of ±2 tokens, teoxkpeenr bigrams, positional index, nadnodw wK oLf divergence as our distance measure. To better understand the sensitivity of the method to each parameter, we perform a post-hoc parameter analysis relative to a default setting (as underlined in Table 2), altering one parameter at a time. The results in Table 2 show that bigrams outperform other n-gram orders by a large margin (note that the evaluation is based on a log scale), and information-theoretic measures are superior to the geometric measures. Furthermore, it also indicates using the positional indexing better captures context. However, there is little to distinguish context modelling with just IV words or all tokens. Similarly, the context window size has relatively little impact on the overall performance, supporting our earlier observation from Table 1. 5 Pair Re-ranking by String Similarity Once the contextually-similar (OOV, IV) pairs are generated using the selected parameters in Section 4, we further re-rank this set of pairs in an attempt to boost morphophonemically-similar pairs like (bananaz, bananas), and penalise noisy pairs like (paninis, beans). Instead of using the small 10 million tweet corpus, from this step onwards, we use a larger corpus of 80 million English tweets (collected over the same period as the development corpus) to develop a larger-scale normalisation dictionary. This is because once pairs are generated, re-ranking based on string comparison is much faster. We only include in the dictionary OOV words with a token frequency > 15 to include more OOV types than in Section 4, and again apply a minimum length cutoff of 4 char- acters. To measure how well our re-ranking method promotes correct pairs and demotes incorrect pairs (including both OOV words that should not be normalised, e.g. (Youtube, web), and incorrect normalRankWindow sizen-gramPositional index?Lex. choiceSim/distance measurelog(CG) 1±32YesAllKL divergence19.571 2 ±±33 2 No All KL divergence 19.562 3 ±±23 2 Yes All KL divergence 19.562 4 ±±32 2 Yes IVs KL divergence 19.561 5 ±±23 2 Yes IVs JS divergence 19.554 ±2 Table 1: The five best parameter combinations in the exhaustive search of parameter combinations Window sizen-gramPositional index?Lexical choiceSimilarity/distance measure ±1 19.3251 19.328Yes 19.328IVs 19.335KL divergence 19.328 ±±21 1199..332275 2 19.571 No 19.263 All 19.328 Euclidean 19.227 ±±32 1199..332287 3 19.324 JS divergence 19.31 1 Cosine 19.170 Table 2: Parameter sensitivity analysis measured as log(CG) for correctly-generated pairs. We tune one parameter at a time, using the default (underlined) setting for other parameters; the non-exhaustive best-performing setting in each case is indicated in bold. isations for lexical variants, e.g. (bcuz, cause)), we modify our evaluation metric from Section 4 to evaluate the ranking at different points, using Discounted Cumulative Gain (DCG@N: Jarvelin and Kekalainen (2002)): DCG@N = rel1+XiN=2logr2el(i ) where reli again represents the frequency of the OOV, but it can be gain (a positive number) or loss (a negative number), depending on whether the ith pair is correct or incorrect. Because we also expect correct pairs to be ranked higher than incorrect pairs, DCG@N takes both factors into account. Given the generated pairs and the evaluation metric, we first consider three baselines: no re-ranking (i.e., the final ranking is that of the contextual similarity scores), and re-rankings of the pairs based on the frequencies of the OOVs in the Twitter corpus, and the IV unigram frequencies in the Google Web 1T corpus (Brants and Franz, 2006) to get less-noisy frequency estimates. We also compared a variety of re-rankings based on a number of string similarity measures that have been previously considered in normalisation work (reviewed in Section 2). We experiment with standard edit distance (Levenshtein, 1966), edit distance over double metaphone codes (phonetic edit distance: (Philips, 2000)), longest common subsequence ratio over the consonant edit distance of the paired words (hereafter, denoted as 426 consonant edit distance: (Contractor et al., 2010)), and a string subsequence kernel (Lodhi et al., 2002). In Figure 1, we present the DCG@N results for each of our ranking methods at different rank cutoffs. Ranking by OOV frequency is motivated by the assumption that lexical variants are frequently used by social media users. This is confirmed by our findings that lexical pairs like (goin, going) and (nite, night) are at the top of the ranking. However, many proper nouns and named entities are also used frequently and ranked at the top, mixed with lexical variants like (Facebook, speech) and (Youtube, web). In ranking by IV word frequency, we assume the lexical variants are usually derived from frequently-used IV equivalents, e.g. (abou, about). However, many less-frequent lexical variant types have high-frequency (IV) normalisations. For instance, the highest-frequency IV word the has more than 40 OOV lexical variants, such as tthe and thhe. These less-frequent types occupy the top positions, reducing the cumulative gain. Compared with these two baselines, ranking by default contextual similarity scores delivers promising results. It successfully ranks many more intuitive normalisation pairs at the top, such as (2day, today) and (wknd, weekend), but also ranks some incorrect pairs highly, such as (needa, gotta). The string similarity-based methods perform better than our baselines in general. Through manual analysis, we found that standard edit distance ranking is fairly accurate for lexical variants with low edit distance to their standard forms, but fails to identify heavily-altered variants like (tmrw, tomorrow). Consonant edit distance is similar to standard edit distance, but places many longer words at the top of the ranking. Edit distance over double metaphone codes (phonetic edit distance) performs particularly well for lexical variants that include character repetitions commonly used for emphasis on Twitter because such repetitions do not typically alter the phonetic codes. Compared with the other methods, the string subsequence kernel delivers encouraging results. It measures common character subsequences of length n between (OOV, IV) pairs. Because it is computationally expensive to calculate similarity for larger n, we choose n=2, following Gouws et al. (201 1). As N (the lexicon size cut-off) increases, the performance drops more slowly than the other meth— — ods. Although this method fails to rank heavilyaltered variants such as (4get,forget) highly, it typically works well for longer words. Given that we focus on longer OOVs (specifically those longer than 4 characters), this ultimately isn’t a great handicap. 6 Evaluation Given the re-ranked pairs from Section 5, here we apply them to a token-level normalisation task using the normalisation dataset of Han and Baldwin (201 1). 6.1 Metrics We evaluate using the standard evaluation metrics of precision (P), recall (R) and F-score (F) as detailed below. We also consider the false alarm rate (FA) and word error rate (WER), also as shown below. FA measures the negative effects of applying normalisation; a good approach to normalisation should not (incorrectly) normalise tokens that are already in their standard form and do not require normalisation.6 WER, like F-score, shows the overall benefits of normalisation, but unlike F-score, measures how many token-level edits are required for the output to be the same as the ground truth data. In general, dictionaries with a high F-score/low WER and low FA 6FA + P ≤ 1because some lexical variants might be incorrectly Ano +rm Pa ≤lise 1d b. 427 are preferable. P = R= F = FA = WER = # cor#re nctolrym naolrismedal tioskeden toskens # to ckoernresc rtelyqu niori nmga nloisremda tloiskaetniosn P2P +R R # inco#rr encotrlmya nliosremda tloikseedns tokens # token edits n#ee adlletd o akfetnesr normalisation 6.2 Results We select the three best re-ranking methods, and best cut-off N for each method, based on the highest DCG@N value for a given method over the development data, as presented in Figure 1. Namely, they are string subsequence kernel (S-dict, N=40,000), double metaphone edit distance (DMdict, N=10,000) and default contextual similarity without re-ranking (C-dict, N=10,000).7 We evaluate each of the learned dictionaries in Table 3. We also compare each dictionary with the performance of the manually-constructed Internet slang dictionary (HB-dict) used by Han and Baldwin (201 1), the small automatically-derived dictionary of Gouws et al. (201 1) (GHM-dict), and combinations of the different dictionaries. In addition, the contribution of these dictionaries in hybrid normalisation approaches is also presented, in which we first normalise OOVs using a given dictionary (combined or otherwise), and then apply the normalisation method of Gouws et al. (201 1) based on consonant edit distance (GHM-norm), or the approach of Han and Baldwin (201 1) based on the summation of many unsupervised approaches (HB-norm), to the remaining OOVs. Results are shown in Table 3, and discussed below. 6.2.1 Individual Dictionaries Overall, the individual dictionaries derived by the re-ranking methods (DM-dict, S-dict) perform bet- 7We also experimented with combining ranks using Mean Reciprocal Rank. However, the combined rank didn’t improve performance on the development data. We plan to explore other ranking aggregation methods in future work. 1 3 5 7 9 11 31 51 71 91 N cut−offs Figure 1: Re-ranking based on different string similarity methods. ter than that based on contextual similarity (C-dict) in terms of precision and false alarm rate, indicating the importance of re-ranking. Even though C-dict delivers higher recall indicating that many lexical variants are correctly normalised this is offset by its high false alarm rate, which is particularly undesirable in normalisation. Because S-dict has better performance than DM-dict in terms of both F-score and WER, and a much lower false alarm rate than C-dict, subsequent results are presented using S-dict only. — — Both HB-dict and GHM-dict achieve better than 90% precision with moderate recall. Compared to these methods, S-dict is not competitive in terms of either precision or recall. This result seems rather discouraging. However, considering that S-dict is an automatically-constructed dictionary targeting lexical variants of varying frequency, it is not surprising that the precision is worse than that of HB-dict which is manually-constructed and GHM-dict which includes entries only for more-frequent OOVs for which distributional similarity is more accurate. Additionally, the recall of S-dict is hampered by the — — — 428 restriction on lexical variant token length of 4 characters. 6.2.2 Combined Dictionaries Next we look to combining HB-dict, GHM-dict and S-dict. In combining the dictionaries, a given OOV word can be listed with different standard forms in different dictionaries. In such cases we use the following preferences for dictionaries motivated by our confidence in the normalisation pairs — of the dictionaries to resolve conflicts: HB-dict > GHM-dict > S-dict. When we combine dictionaries in the second section of Table 3, we find that they contain complementary information: in each case the recall and F-score are higher for the combined dictionary than any of the individual dictionaries. The combination of HB-dict+GHM-dict produces only a small improvement in terms of F-score over HBdict (the better-performing dictionary) suggesting that, as claimed, HB-dict and GHM-dict share many frequent normalisation pairs. HB-dict+S-dict and GHM-dict+S-dict, on the other hand, improve sub— MethodPrecisionRecallF-ScoreFalse AlarmWord Error Rate C-dict0.4740.2180.2990.2980.103 DM-dict S-dict HB-dict GHM-dict 0.727 0.700 0.915 0.982 0.106 0.179 0.435 0.319 0.185 0.285 0.590 0.482 0.145 0.162 0.048 0.000 0.102 0.097 0.066 0.076 HB-dict+S-dict0.8400.6010.7010.0900.052 GHM-dict+S-dict HB-dict+GHM-dict HB-dict+GHM-dict+S-dict 0.863 0.920 0.847 0.498 0.465 0.630 0.632 0.618 0.723 0.072 0.045 0.086 0.061 0.063 0.049 GHM-dict+GHM-norm0.3380.5780.4270.4580.135 HB-dict+GHM-dict+S-dict+GHM-norm HB-dict+HB-norm HB-dict+GHM-dict+S-dict+HB-norm 0.406 0.515 0.527 0.715 0.771 0.789 0.518 0.618 0.632 0.468 0.332 0.332 0.124 0.081 0.079 Table 3: Normalisation results using our derived dictionaries (contextual similarity (C-dict); double metaphone rendering (DM-dict); string subsequence kernel scores (S-dict)), the dictionary of Gouws et al. (201 1) (GHM-dict), the Internet slang dictionary (HB-dict) from Han and Baldwin (201 1), and combinations of these dictionaries. In addition, we combine the dictionaries with the normalisation method of Gouws et al. (201 1) (GHM-norm) and the combined unsupervised approach of Han and Baldwin (201 1) (HB-norm). stantially over HB-dict and GHM-dict, respectively, indicating that S-dict contains markedly different entries to both HB-dict and GHM-dict. The best Fscore and WER are obtained using the combination of all three dictionaries, HB-dict+GHM-dict+S-dict. Furthermore, the difference between the results using HB-dict+GHM-dict+S-dict and HB-dict+GHMdict is statistically significant (p < 0.01), based on the computationally-intensive Monte Carlo method of Yeh (2000), demonstrating the contribution of Sdict. 6.2.3 Hybrid Approaches The methods of Gouws et al. (201 1) (i.e. GHM-dict+GHM-norm) and Han and Baldwin (201 1) (i.e. HB-dict+HB-norm) have lower precision and higher false alarm rates than the dictionarybased approaches; this is largely caused by lexical variant detection errors.8 Using all dictionaries in combination with these methods HB-dict+GHM-dict+S-dict+GHM-norm and HBdict+GHM-dict+S-dict+HB-norm gives some improvements, but the false alarm rates remain high. Despite the limitations of a pure dictionary-based approach to normalisation discussed in Section 3.1 the current best practical approach to normal— — — — 8Here we report results that do not assume perfect detection of lexical variants, unlike the original published results in each case. 429 Error typeOOVDSitcat.ndard fGoromld (a) pluralsplayeplayersplayer (b) negation unlike like dislike (c) possessives anyones anyone anyone ’s (d) correct OOVs iphone phone iphone (e) test data errors durin during durin (f) ambiguity siging signing singing Table 4: Error types in the combined dictionary (HBdict+GHM-dict+S-dict) isation is to use a lexicon, combining hand-built and automatically-learned normalisation dictionaries. 6.3 Discussion and Error Analysis We first manually analyse the errors in the combined dictionary (HB-dict+GHM-dict+S-dict) and give examples of each error type in Table 4. The most frequent word errors are caused by slight morphologi- cal variations, including plural forms (a), negations (b), possessive cases (c), and OOVs that are correct and do not require normalisation (d). In addition, we also notice some missing annotations where lexical variants are skipped by human annotations but captured by our method (e). Ambiguity (f) definitely exists in longer OOVs, however, these cases do not appear to have a strong negative impact on the normalisation performance. An example of a remainLength cut-off (N)#VariantsPrecisionRecall (≥ N)Recall (all)False Alarm ≥45560.700Rec0al.l3 8(≥1 N)0.1790.162 ≥≥54 382 0.814 0.471 0.152 0.122 ≥≥65 254 0.804 0.484 0.104 0.131 ≥≥76 138 0.793 0.471 0.055 0.122 ≥71380.7930.4710.0550.122 Table 5: S-dict normalisation results broken down according to OOV token length. Recall is presented both over the subset of instances of length ≥ N in the data (“Recall (≥ N)”), and over the entirety of the dataset (“Recall (all)”); “su#bVsaertia onftis n” sitsa tnhcee snu omfb leenrg othf t≥ok Nen iinns tthaenc deast ao f( “tRhee cinadllic (a≥ted N length idn o othveer rt tehset d eanttaisreetty. ing miscellaneous error is bday “birthday”, which is mis-normalised as day. To further study the influence of OOV word length relative to the normalisation performance, we conduct a fine-grained analysis of the performance of the derived dictionary (S-dict) in Table 5, broken down across different OOV word lengths. The results generally support our hypothesis that our method works better for longer OOV words. The derived dictionary is much more reliable for longer tokens (length 5, 6, and 7 characters) in terms of precision and false alarm. Although the recall is relatively modest, in the future we intend to improve recall by mining more normalisation pairs from larger collections of microblog data. 7 Conclusions and Future Work In this paper, we describe a method for automatically constructing a normalisation dictionary that supports normalisation of microblog text through direct substitution of lexical variants with their standard forms. After investigating the impact of different distributional and string similarity methods on the quality of the dictionary, we present experimental results on a standard dataset showing that our proposed methods acquire high quality (lexical variant, standard form) pairs, with reasonable coverage, and achieve state-of-the-art end-toend lexical normalisation performance on a realworld token-level task. Furthermore, this dictionarylookup method combines the detection and normalisation of lexical variants into a simple, lightweight solution which is suitable for processing of highvolume microblog feeds. In the future, we intend to improve our dictionary by leveraging the constantly-growing volume of microblog data, and considering alternative ways to combine distributional and string similarity. In addi430 tion to direct evaluation, we also want to explore the benefits of applying normalisation for downstream social media text processing applications, e.g. event detection. Acknowledgements We would like to thank the three anonymous reviewers for their insightful comments, and Stephan Gouws for kindly sharing his data and discussing his work. NICTA is funded by the Australian government as represented by Department of Broadband, Communication and Digital Economy, and the Australian Research Council through the ICT centre of Excellence programme. References AiTi Aw, Min Zhang, Juan Xiao, and Jian Su. 2006. A phrase-based statistical model for SMS text normalization. In Proceedings of COLING/ACL 2006, pages 33–40, Sydney, Australia. Edward Benson, Aria Haghighi, and Regina Barzilay. 2011. Event discovery in social media feeds. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 389–398, Portland, Oregon, USA. Thorsten Brants and Alex Franz. 2006. Web 1T 5-gram Version 1. Eric Brill and Robert C. Moore. 2000. An improved error model for noisy channel spelling correction. In Proceedings of the 38th Annual Meeting of the Association for Computational Linguistics, pages 286–293, Hong Kong. Monojit Choudhury, Rahul Saraf, Vijit Jain, Animesh Mukherjee, Sudeshna Sarkar, and Anupam Basu. 2007. Investigation and modeling of the structure of texting language. International Journal on Document Analysis and Recognition, 10: 157–174. Danish Contractor, Tanveer A. Faruquie, and L. Venkata Subramaniam. 2010. Unsupervised cleansing of noisy text. In Proceedings of the 23rd International Conference on Computational Linguistics (COLING 2010), pages 189–196, Beijing, China. Paul Cook and Suzanne Stevenson. 2009. An unsu- pervised model for text message normalization. In CALC ’09: Proceedings of the Workshop on Computational Approaches to Linguistic Creativity, pages 71– 78, Boulder, USA. Jennifer Foster, O¨zlem C ¸etinoglu, Joachim Wagner, Joseph L. Roux, Stephen Hogan, Joakim Nivre, Deirdre Hogan, and Josef van Genabith. 2011. #hardtoparse: POS Tagging and Parsing the Twitterverse. In Analyzing Microtext: Papers from the 2011 AAAI Workshop, volume WS-1 1-05 of AAAI Workshops, pages 20–25, San Francisco, CA, USA. Kevin Gimpel, Nathan Schneider, Brendan O’Connor, Dipanjan Das, Daniel Mills, Jacob Eisenstein, Michael Heilman, Dani Yogatama, Jeffrey Flanigan, and Noah A. Smith. 2011. Part-of-speech tagging for Twitter: Annotation, features, and experiments. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 42–47, Portland, Oregon, USA. Roberto Gonz a´lez-Ib ´a n˜ez, Smaranda Muresan, and Nina Wacholder. 2011. Identifying sarcasm in Twitter: a closer look. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 581–586, Portland, Oregon, USA. Stephan Gouws, Dirk Hovy, and Donald Metzler. 2011. Unsupervised mining of lexical variants from noisy text. In Proceedings of the First workshop on Unsupervised Learning in NLP, pages 82–90, Edinburgh, Scotland, UK. Bo Han and Timothy Baldwin. 2011. Lexical normalisation of short text messages: Makn sens a #twitter. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 368–378, Portland, Oregon, USA. K. Jarvelin and J. Kekalainen. 2002. Cumulated gainbased evaluation of IR techniques. ACM Transactions on Information Systems, 20(4). Long Jiang, Mo Yu, Ming Zhou, Xiaohua Liu, and Tiejun Zhao. 2011. Target-dependent Twitter sentiment classification. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 151–160, Portland, Oregon, USA. Joseph Kaufmann and Jugal Kalita. 2010. Syntactic normalization of Twitter messages. In International Con431 ference on Natural Language Processing, Kharagpur, India. S. Kullback and R. A. Leibler. 1951. On information and sufficiency. Annals of Mathematical Statistics, 22:49– 86. John D. Lafferty, Andrew McCallum, and Fernando C. N. Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. In Proceedings ofthe Eighteenth International Conference on Machine Learning, pages 282–289, San Francisco, CA, USA. Vladimir I. Levenshtein. 1966. Binary codes capable of correcting deletions, insertions, and reversals. Soviet Physics Doklady, 10:707–710. Mu Li, Yang Zhang, Muhua Zhu, and Ming Zhou. 2006. Exploring distributional similarity based models for query spelling correction. In Proceedings of COLING/ACL 2006, pages 1025–1032, Sydney, Australia. Jianhua Lin. 1991. Divergence measures based on the shannon entropy. IEEE Transactions on Information Theory, 37(1): 145–151. Dekang Lin. 1998. Automatic retrieval and clustering of similar words. In Proceedings of the 36th Annual Meeting of the ACL and 1 International Con7th ference on Computational Linguistics (COLING/ACL98), pages 768–774, Montreal, Quebec, Canada. Fei Liu, Fuliang Weng, Bingqing Wang, and Yang Liu. 2011a. Insertion, deletion, or substitution? normalizing text messages without pre-categorization nor supervision. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 71–76, Portland, Oregon, USA. Xiaohua Liu, Shaodian Zhang, Furu Wei, and Ming Zhou. 2011b. Recognizing named entities in tweets. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 359–367, Portland, Oregon, USA. Fei Liu, Fuliang Weng, and Xiao Jiang. 2012. A broadcoverage normalization system for social media language. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (ACL 2012), Jeju, Republic of Korea. Huma Lodhi, Craig Saunders, John Shawe-Taylor, Nello Cristianini, and Chris Watkins. 2002. Text classification using string kernels. J. Mach. Learn. Res., 2:419– 444. Marco Lui and Timothy Baldwin. 2011. Cross-domain feature selection for language identification. In Proceedings of the 5th International Joint Conference on Natural Language Processing (IJCNLP 2011), pages 553–561, Chiang Mai, Thailand. Brendan O’Connor, Michel Krieger, and David Ahn. 2010. TweetMotif: Exploratory search and topic summarization for Twitter. In Proceedings of the 4th International Conference on Weblogs and Social Media (ICWSM 2010), pages 384–385, Washington, USA. Lawrence Philips. 2000. The double metaphone search algorithm. C/C++ Users Journal, 18:38–43. Lawrence R. Rabiner. 1989. A tutorial on hidden Markov models and selected applications in speech recognition. Proceedings of the IEEE, 77(2):257–286. Alan Ritter, Colin Cherry, and Bill Dolan. 2010. Unsupervised modeling of Twitter conversations. In Proceedings of Human Language Technologies: The 11th Annual Conference of the North American Chap- ter of the Association for Computational Linguistics (NAACL-HLT 2010), pages 172–180, Los Angeles, USA. Alan Ritter, Sam Clark, Mausam, and Oren Etzioni. 2011. Named entity recognition in tweets: An experimental study. In Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing (EMNLP 2011), pages 1524–1534, Edinburgh, Scotland, UK. Takeshi Sakaki, Makoto Okazaki, and Yutaka Matsuo. 2010. Earthquake shakes Twitter users: real-time event detection by social sensors. In Proceedings of the 19th International Conference on the World Wide Web (WWW 2010), pages 851–860, Raleigh, North Carolina, USA. Crispin Thurlow. 2003. Generation txt? The sociolinguistics of young people’s text-messaging. Discourse Analysis Online, 1(1). Kristina Toutanova and Robert C. Moore. 2002. Pronunciation modeling for improved spelling correction. In Proceedings of the 40th Annual Meeting of the ACL and 3rd Annual Meeting of the NAACL (ACL-02), pages 144–15 1, Philadelphia, USA. Official Blog Twitter. 2011. 200 million tweets per day. Retrived at August 17th, 2011. Jianshu Weng and Bu-Sung Lee. 2011. Event detection in Twitter. In Proceedings of the 5th International Conference on Weblogs and Social Media (ICWSM 2011), Barcelona, Spain. Zhenzhen Xue, Dawei Yin, and Brian D. Davison. 2011. Normalizing microtext. In Proceedings of the AAAI11 Workshop on Analyzing Microtext, pages 74–79, San Francisco, USA. Alexander Yeh. 2000. More accurate tests for the statistical significance of result differences. In Proceedings of the 23rd International Conference on Computational Linguistics (COLING 2010), pages 947–953, Saarbr¨ ucken, Germany. 432
5 0.37020662 121 emnlp-2012-Supervised Text-based Geolocation Using Language Models on an Adaptive Grid
Author: Stephen Roller ; Michael Speriosu ; Sarat Rallapalli ; Benjamin Wing ; Jason Baldridge
Abstract: The geographical properties of words have recently begun to be exploited for geolocating documents based solely on their text, often in the context of social media and online content. One common approach for geolocating texts is rooted in information retrieval. Given training documents labeled with latitude/longitude coordinates, a grid is overlaid on the Earth and pseudo-documents constructed by concatenating the documents within a given grid cell; then a location for a test document is chosen based on the most similar pseudo-document. Uniform grids are normally used, but they are sensitive to the dispersion of documents over the earth. We define an alternative grid construction using k-d trees that more robustly adapts to data, especially with larger training sets. We also provide a better way of choosing the locations for pseudo-documents. We evaluate these strategies on existing Wikipedia and Twitter corpora, as well as a new, larger Twitter corpus. The adaptive grid achieves competitive results with a uniform grid on small training sets and outperforms it on the large Twitter corpus. The two grid constructions can also be combined to produce consistently strong results across all training sets.
6 0.29843661 72 emnlp-2012-Joint Inference for Event Timeline Construction
7 0.27343762 38 emnlp-2012-Employing Compositional Semantics and Discourse Consistency in Chinese Event Extraction
9 0.18987614 35 emnlp-2012-Document-Wide Decoding for Phrase-Based Statistical Machine Translation
10 0.18916887 78 emnlp-2012-Learning Lexicon Models from Search Logs for Query Expansion
11 0.16541739 89 emnlp-2012-Mixed Membership Markov Models for Unsupervised Conversation Modeling
12 0.1590106 43 emnlp-2012-Exact Sampling and Decoding in High-Order Hidden Markov Models
13 0.15192363 37 emnlp-2012-Dynamic Programming for Higher Order Parsing of Gap-Minding Trees
14 0.1511687 56 emnlp-2012-Framework of Automatic Text Summarization Using Reinforcement Learning
15 0.1511171 77 emnlp-2012-Learning Constraints for Consistent Timeline Extraction
16 0.15030195 130 emnlp-2012-Unambiguity Regularization for Unsupervised Learning of Probabilistic Grammars
17 0.14591981 33 emnlp-2012-Discovering Diverse and Salient Threads in Document Collections
18 0.13274445 123 emnlp-2012-Syntactic Transfer Using a Bilingual Lexicon
19 0.13249889 5 emnlp-2012-A Discriminative Model for Query Spelling Correction with Latent Structural SVM
20 0.13108388 97 emnlp-2012-Natural Language Questions for the Web of Data
topicId topicWeight
[(2, 0.021), (11, 0.01), (16, 0.047), (25, 0.011), (34, 0.065), (60, 0.069), (63, 0.034), (64, 0.017), (65, 0.018), (68, 0.405), (70, 0.028), (74, 0.031), (76, 0.048), (80, 0.016), (81, 0.01), (86, 0.032), (95, 0.042)]
simIndex simValue paperId paperTitle
same-paper 1 0.75839239 63 emnlp-2012-Identifying Event-related Bursts via Social Media Activities
Author: Xin Zhao ; Baihan Shu ; Jing Jiang ; Yang Song ; Hongfei Yan ; Xiaoming Li
Abstract: Activities on social media increase at a dramatic rate. When an external event happens, there is a surge in the degree of activities related to the event. These activities may be temporally correlated with one another, but they may also capture different aspects of an event and therefore exhibit different bursty patterns. In this paper, we propose to identify event-related bursts via social media activities. We study how to correlate multiple types of activities to derive a global bursty pattern. To model smoothness of one state sequence, we propose a novel function which can capture the state context. The experiments on a large Twitter dataset shows our methods are very effective.
2 0.62319571 77 emnlp-2012-Learning Constraints for Consistent Timeline Extraction
Author: David McClosky ; Christopher D. Manning
Abstract: We present a distantly supervised system for extracting the temporal bounds of fluents (relations which only hold during certain times, such as attends school). Unlike previous pipelined approaches, our model does not assume independence between each fluent or even between named entities with known connections (parent, spouse, employer, etc.). Instead, we model what makes timelines of fluents consistent by learning cross-fluent constraints, potentially spanning entities as well. For example, our model learns that someone is unlikely to start a job at age two or to marry someone who hasn’t been born yet. Our system achieves a 36% error reduction over a pipelined baseline.
3 0.32988316 120 emnlp-2012-Streaming Analysis of Discourse Participants
Author: Benjamin Van Durme
Abstract: Inferring attributes of discourse participants has been treated as a batch-processing task: data such as all tweets from a given author are gathered in bulk, processed, analyzed for a particular feature, then reported as a result of academic interest. Given the sources and scale of material used in these efforts, along with potential use cases of such analytic tools, discourse analysis should be reconsidered as a streaming challenge. We show that under certain common formulations, the batchprocessing analytic framework can be decomposed into a sequential series of updates, using as an example the task of gender classification. Once in a streaming framework, and motivated by large data sets generated by social media services, we present novel results in approximate counting, showing its applicability to space efficient streaming classification.
4 0.32815671 22 emnlp-2012-Automatically Constructing a Normalisation Dictionary for Microblogs
Author: Bo Han ; Paul Cook ; Timothy Baldwin
Abstract: Microblog normalisation methods often utilise complex models and struggle to differentiate between correctly-spelled unknown words and lexical variants of known words. In this paper, we propose a method for constructing a dictionary of lexical variants of known words that facilitates lexical normalisation via simple string substitution (e.g. tomorrow for tmrw). We use context information to generate possible variant and normalisation pairs and then rank these by string similarity. Highlyranked pairs are selected to populate the dictionary. We show that a dictionary-based approach achieves state-of-the-art performance for both F-score and word error rate on a standard dataset. Compared with other methods, this approach offers a fast, lightweight and easy-to-use solution, and is thus suitable for high-volume microblog pre-processing. 1 Lexical Normalisation A staggering number of short text “microblog” messages are produced every day through social media such as Twitter (Twitter, 2011). The immense volume of real-time, user-generated microblogs that flows through sites has been shown to have utility in applications such as disaster detection (Sakaki et al., 2010), sentiment analysis (Jiang et al., 2011; Gonz a´lez-Ib ´a n˜ez et al., 2011), and event discovery (Weng and Lee, 2011; Benson et al., 2011). However, due to the spontaneous nature of the posts, microblogs are notoriously noisy, containing many non-standard forms e.g., tmrw “tomorrow” and 2day “today” which degrade the performance of — — 421 natural language processing (NLP) tools (Ritter et al., 2010; Han and Baldwin, 2011). To reduce this effect, attempts have been made to adapt NLP tools to microblog data (Gimpel et al., 2011; Foster et al., 2011; Liu et al., 2011b; Ritter et al., 2011). An alternative approach is to pre-normalise non-standard lexical variants to their standard orthography (Liu et al., 2011a; Han and Baldwin, 2011; Xue et al., 2011; Gouws et al., 2011). For example, se u 2morw!!! would be normalised to see you tomorrow! The normalisation approach is especially attractive as a preprocessing step for applications which rely on keyword match or word frequency statistics. For example, earthqu, eathquake, and earthquakeee all attested in a Twitter corpus have the standard form earthquake; by normalising these types to their standard form, better coverage can be achieved for keyword-based methods, and better word frequency estimates can be obtained. In this paper, we focus on the task of lexical normalisation of English Twitter messages, in which out-of-vocabulary (OOV) tokens are normalised to their in-vocabulary (IV) standard form, i.e., a standard form that is in a dictionary. Following other recent work on lexical normalisation (Liu et al., 2011a; Han and Baldwin, 2011; Gouws et al., 2011; Liu et al., 2012), we specifically focus on one-to-one normalisation in which one OOV token is normalised to one IV word. Naturally, not all OOV words in microblogs are lexical variants of IV words: named entities, e.g., — — are prevalent in microblogs, but not all named entities are included in our dictionary. One challenge for lexical normalisation is therefore to disPLraoncge uadgineg Lse oafr tnhineg 2,0 p1a2g Jeosin 42t C1–o4n3f2e,re Jnecjue Iosnla Enmd,p Kiroicraela, M 1e2t–h1o4ds Ju ilny N 20a1tu2r.a ?lc L2a0n1g2ua Agseso Pcrioactieosnsi fnogr a Cnodm Cpoumtaptiuotna tilo Lnianlg Nuaist uircasl tinguish those OOV tokens that require normalisation from those that are well-formed. Recent unsupervised approaches have not attempted to distinguish such tokens from other types of OOV tokens (Cook and Stevenson, 2009; Liu et al., 2011a), limiting their applicability to real-world normalisation tasks. Other approaches (Han and Baldwin, 2011; Gouws et al., 2011) have followed a cascaded approach in which lexical variants are first identified, and then normalised. However, such two-step approaches suffer from poor lexical variant identification performance, which is propagated to the normalisation step. Motivated by the observation that most lexical variants have an unambiguous standard form (especially for longer tokens), and that a lexical variant and its standard form typically occur in similar contexts, in this paper we propose methods for automatically constructing a lexical normalisation dictionary a dictionary whose entries consist — of (lexical variant, standard form) pairs that enables type-based normalisation. Despite the simplicity of this dictionary-based normalisation method, we show it to outperform previously-proposed approaches. This very fast, lightweight solution is suitable for real-time processing of the large volume of streaming microblog data available from Twitter, and offers a simple solution to the lexical variant detection problem that hinders other normalisation methods. Furthermore, this dictionary-based method can be easily integrated with other more-complex normalisation approaches (Liu et al., 2011a; Han and Baldwin, 2011; Gouws et al., 2011) to produce hybrid systems. After discussing related work in Section 2, we present an overview of our dictionary-based approach to normalisation in Section 3. In Sections 4 and 5 we experimentally select the optimised context similarity parameters and string similarity reranking method. We present experimental results on the unseen test data in Section 6, and offer some concluding remarks in Section 7. — 2 Related Work Given a token t, lexical normalisation is the task of finding arg max P(s|t) ∝ arg max P(t| s)P(s), wofh efinred s igs tahreg smtaanxdaPrd(s form, i.e., an aIVx Pw(otr|sd). PSt(asn)-, dardly in lexical normalisation, t is assumed to be an 422 OOV token, relative to a fixed dictionary. In practice, not all OOV tokens should be normalised; i.e., only lexical variants (e.g., tmrw “tomorrow”) should be normalised and tokens that are OOV but otherwise not lexical variants (e.g., iPad “iPad”) should be unchanged. Most work in this area focuses only on the normalisation task itself, oftentimes assuming that the task of lexical variant detection has already been completed. Various approaches have been proposed to estimate the error model, P(t|s). For example, in work on spell-checking, eBl,ril Pl (atn|ds) M. Fooorre e (2000) improve on a standard edit-distance approach by considering multi-character edit operations; Toutanova and Moore (2002) build on this by incorporating phonological information. Li et al. (2006) utilise distributional similarity (Lin, 1998) to correct misspelled search queries. In text message normalisation, Choudhury et al. (2007) model the letter transformations and emissions using a hidden Markov model (Rabiner, 1989). Cook and Stevenson (2009) and Xue et al. (201 1) propose multiple simple error models, each of which captures a particular way in which lexical variants are formed, such as phonetic spelling (e.g., epik “epic”) or clipping (e.g., walkin “walking”). Nevertheless, optimally weighting the various error models in these approaches is challenging. Without pre-categorising lexical variants into different types, Liu et al. (201 1a) collect Google search snippets from carefully-designed queries from which they then extract noisy lexical variant– standard form pairs. These pairs are used to train a conditional random field (Lafferty et al., 2001) to estimate P(t|s) at the character level. One shortcoming eo fP querying a ese cahracrha engine teol. .o Obtanein strhaoirnt-ing pairs is it tends to be costly in terms of time and bandwidth. Here we exploit microblog data directly to derive (lexical variant, standard form) pairs, instead of relying on external resources. In morerecent work, Liu et al. (2012) endeavour to improve the accuracy of top-n normalisation candidates by integrating human cognitive inference, characterlevel transformations and spell checking in their normalisation model. The encouraging results shift the focus to reranking and promoting the correct normalisation to the top-1 position. However, like much previous work on lexical normalisation, this work assumes perfect lexical variant detection. Aw et al. (2006) and Kaufmann and Kalita (2010) consider normalisation as a machine translation task from lexical variants to standard forms using off-theshelf tools. These methods do not assume that lexical variants have been pre-identified; however, these methods do rely on large quantities of labelled training data, which is not available for microblogs. Recently, Han and Baldwin (201 1) and Gouws et al. (201 1) propose two-step unsupervised approaches to normalisation, in which lexical variants are first identified, and then normalised. They approach lexical variant detection by using a context fitness classifier (Han and Baldwin, 2011) or through dictionary lookup (Gouws et al., 2011). However, the lexical variant detection of both meth- ods is rather unreliable, indicating the challenge of this aspect of normalisation. Both of these approaches incorporate a relatively small normalisation dictionary to capture frequent lexical variants with high precision. In particular, Gouws et al. (201 1) produce a small normalisation lexicon based on distributional similarity and string similarity (Lodhi et al., 2002). Our method adopts a similar strategy using distributional/string similarity, but instead of constructing a small lexicon for preprocessing, we build a much wider-coverage normalisation dictionary and opt for a fully lexiconbased end-to-end normalisation approach. In contrast to the normalisation dictionaries of Han and Baldwin (201 1) and Gouws et al. (201 1) which focus on very frequent lexical variants, we focus on moderate frequency lexical variants of a minimum character length, which tend to have unambiguous standard forms; our intention is to produce normalisation lexicons that are complementary to those currently available. Furthermore, we investigate the impact of a variety of contextual and string similarity measures on the quality of the resulting lexicons. In summary, our dictionary-based normalisation ap- proach is a lightweight end-to-end method which performs both lexical variant detection and normalisation, and thus is suitable for practical online preprocessing, despite its simplicity. 423 3 A Lexical Normalisation Dictionary Before discussing our method for creating a normalisation dictionary, we first discuss the feasibility of such an approach. 3.1 Feasibility Dictionary lookup approaches to normalisation have been shown to have high precision but low recall (Han and Baldwin, 2011; Gouws et al., 2011). Frequent (lexical variant, standard form) pairs such as (u, you) are typically included in the dictionaries used by such methods, while less-frequent items such as (g0tta, gotta) are generally omitted. Because of the degree of lexical creativity and large number of non-standard forms observed on Twitter, a wide-coverage normalisation dictionary would be expensive to construct manually. Based on the assumption that lexical variants occur in similar con- texts to their standard forms, however, it should be possible to automatically construct a normalisation dictionary with wider coverage than is currently available. Dictionary lookup is a type-based approach to normalisation, i.e., every token instance of a given type will always be normalised in the same way. However, lexical variants can be ambiguous, e.g., y corresponds to “you” in yeah, y r right! LOL but “why” in AM CONFUSED!!! y you did that? Nevertheless, the relative occurrence of ambiguous lexical variants is small (Liu et al., 2011a), and it has been observed that while shorter variants such as y are often ambiguous, longer variants tend to be unambiguous. For example bthday and 4eva are unlikely to have standard forms other than “birthday” and “forever”, respectively. Therefore, the normalisation lexicons we produce will only contain entries for OOVs with character length greater than a specified threshold, which are likely to have an unambiguous standard form. 3.2 Overview of approach Our method for constructing a normalisation dictio- nary is as follows: Input: Tokenised English tweets 1. Extract (OOV, IV) pairs based on distributional similarity. 2. Re-rank the extracted pairs by string similarity. Output: A list of (OOV, IV) pairs ordered by string similarity; select the top-n pairs for inclusion in the normalisation lexicon. In Step 1, we leverage large volumes of Twitter data to identify the most distributionally-similar IV type for each OOV type. The result of this process is a set of (OOV, IV) pairs, ranked by distributional similarity. The extracted pairs will include (lexical variant, standard form) pairs, such as (tmrw, tomorrow), but will also contain false positives such as (Tusday, Sunday) Tusday is a lexical variant, but its standard form is not “Sunday” and (Youtube, web) Youtube is an OOV named entity, not a lexical variant. Nevertheless, lexical variants are typically formed from their standard forms through regular processes (Thurlow, 2003) e.g., the omission of characters and from this perspective Sunday and web are not plausible standard — — — — — forms for Tusday and Youtube, respectively. In Step 2, we therefore capture this intuition to re-rank the extracted pairs by string similarity. The top-n items in this re-ranked list then form the normalisation lexicon, which is based only on development data. Although computationally-expensive to build, this dictionary can be created offline. Once built, it then offers a very fast approach to normalisation. We can only reliably compute distributional similarity for types that are moderately frequent in a corpus. Nevertheless, many lexical variants are sufficiently frequent to be able to compute distributional similarity, and can potentially make their way into our normalisation lexicon. This approach is not suitable for normalising low-frequency lexical variants, nor is it suitable for shorter lexical variant types which as discussed in Section 3.1 are more likely to have an ambiguous standard form. Nevertheless, previously-proposed normalisation methods that can handle such phenomena also rely in part on a normalisation lexicon. The normalisation lexicons we create can therefore be easily integrated with previous approaches to form hybrid normalisation systems. — — 4 Contextually-similar Pair Generation Our objective is to extract contextually-similar (OOV, IV) pairs from a large-scale collection of mi424 croblog data. Fundamentally, the surrounding words define the primary context, but there are different ways of representing context and different similarity measures we can use, which may influence the quality of generated normalisation pairs. In representing the context, we experimentally explore the following factors: (1) context window size (from 1 to 3 tokens on both sides); (2) n-gram order ofthe context tokens (unigram, bigram, trigram); (3) whether context words are indexed for relative position or not; and (4) whether we use all context tokens, or only IV words. Because high-accuracy linguistic processing tools for Twitter are still under exploration (Liu et al., 2011b; Gimpel et al., 2011; Ritter et al., 2011; Foster et al., 2011), we do not consider richer representations of context, for example, incorporating information about part-of-speech tags or syntax. We also experiment with a number of simple but widely-used geometric and information theoretic distance/similarity measures. In particular, we use Kullback–Leibler (KL) divergence (Kullback and Leibler, 195 1), Jensen–Shannon (JS) divergence (Lin, 1991), Euclidean distance and Cosine distance. We use a corpus of 10 million English tweets to do parameter tuning over, and a larger corpus of tweets in the final candidate ranking. All tweets were collected from September 2010 to January 2011 via the Twitter API.1 From the raw data we extract English tweets using a language identification tool (Lui and Baldwin, 2011), and then apply a simplified Twitter tokeniser (adapted from O’Connor et al. (2010)). We use the Aspell dictionary (v6.06)2 to determine whether a word is IV, and only include in our normalisation dictionary OOV tokens with at least 64 occurrences in the corpus and character length ≥ 4, both of which were determined through empirical 4o,b bsoetrhva otifo wnh. Fcohr w weearceh d OetOeVrm winoedrd t type ginh the corpus, we select the most similar IV type to form (OOV, IV) pairs. To further narrow the search space, we only consider IV words which are morphophonemically similar to the OOV type, follow- ing settings in Han and Baldwin (201 1).3 1http s : / / dev .twitter . com/ docs / st reaming-api /methods 2http : / / aspe l .net / l 3We only consider IV words within an edit distance of 2 or a phonemic edit distance of 1from the OOV type, and we further In order to evaluate the generated pairs, we randomly selected 1000 OOV words from the 10 million tweet corpus. We set up an annotation task on Amazon Mechanical Turk,4 presenting five independent annotators with each word type (with no context) and asking for corrections where appropriate. For instance, given tmrw, the annotators would likely identify it as a non-standard variant of “tomorrow”. For correct OOV words like iPad, on the other hand, we would expect them to leave the word unchanged. If 3 or more of the 5 annotators make the same suggestion (in the form of either a canonical spelling or leaving the word unchanged), we include this in our gold standard for evaluation. In total, this resulted in 351 lexical variants and 282 correct OOV words, accounting for 63.3% of the 1000 OOV words. These 633 OOV words were used as (OOV, IV) pairs for parameter tuning. The remainder of the 1000 OOV words were ignored on the grounds that there was not sufficient consensus amongst the annotators.5 Contextually-similar pair generation aims to include as many correct normalisation pairs as possible. We evaluate the quality of the normalisation pairs using “Cumulative Gain” (CG): XN0 CG = Xreli0 Xi=1 Suppose there are N0 correct generated pairs (oovi, ivi), each of which is weighted by reli0, the frequency of oovi to indicate its relative importance; for example, (thinkin, thinking) has a higher weight than (g0tta, gotta) because thinkin is more frequent than g0tta in our corpus. In this evaluation we don’t consider the position of normalisation pairs, and nor do we penalise incorrect pairs. Instead, we push distinguishing between correct and incorrect pairs into the downstream re-ranking step in which we incorporate string similarity information. Given the development data and CG, we run an exhaustive search of parameter combinations over only consider the top 30% most-frequent of these IV words. 4https : / /www .mturk .com/mturk/welcome 5Note that the objective of this annotation task is to identify lexical variants that have agreed-upon standard forms irrespective of context, as a special case of the more general task of lexical normalisation (where context may or may not play a significant role in the determination of the normalisation). 425 our development corpus. The five best parameter combinations are shown in Table 1. We notice the CG is almost identical for the top combinations. As a context window size of 3 incurs a heavy processing and memory overhead over a size of 2, we use the 3rd-best parameter combination for subsequent experiments, namely: context window of ±2 tokens, teoxkpeenr bigrams, positional index, nadnodw wK oLf divergence as our distance measure. To better understand the sensitivity of the method to each parameter, we perform a post-hoc parameter analysis relative to a default setting (as underlined in Table 2), altering one parameter at a time. The results in Table 2 show that bigrams outperform other n-gram orders by a large margin (note that the evaluation is based on a log scale), and information-theoretic measures are superior to the geometric measures. Furthermore, it also indicates using the positional indexing better captures context. However, there is little to distinguish context modelling with just IV words or all tokens. Similarly, the context window size has relatively little impact on the overall performance, supporting our earlier observation from Table 1. 5 Pair Re-ranking by String Similarity Once the contextually-similar (OOV, IV) pairs are generated using the selected parameters in Section 4, we further re-rank this set of pairs in an attempt to boost morphophonemically-similar pairs like (bananaz, bananas), and penalise noisy pairs like (paninis, beans). Instead of using the small 10 million tweet corpus, from this step onwards, we use a larger corpus of 80 million English tweets (collected over the same period as the development corpus) to develop a larger-scale normalisation dictionary. This is because once pairs are generated, re-ranking based on string comparison is much faster. We only include in the dictionary OOV words with a token frequency > 15 to include more OOV types than in Section 4, and again apply a minimum length cutoff of 4 char- acters. To measure how well our re-ranking method promotes correct pairs and demotes incorrect pairs (including both OOV words that should not be normalised, e.g. (Youtube, web), and incorrect normalRankWindow sizen-gramPositional index?Lex. choiceSim/distance measurelog(CG) 1±32YesAllKL divergence19.571 2 ±±33 2 No All KL divergence 19.562 3 ±±23 2 Yes All KL divergence 19.562 4 ±±32 2 Yes IVs KL divergence 19.561 5 ±±23 2 Yes IVs JS divergence 19.554 ±2 Table 1: The five best parameter combinations in the exhaustive search of parameter combinations Window sizen-gramPositional index?Lexical choiceSimilarity/distance measure ±1 19.3251 19.328Yes 19.328IVs 19.335KL divergence 19.328 ±±21 1199..332275 2 19.571 No 19.263 All 19.328 Euclidean 19.227 ±±32 1199..332287 3 19.324 JS divergence 19.31 1 Cosine 19.170 Table 2: Parameter sensitivity analysis measured as log(CG) for correctly-generated pairs. We tune one parameter at a time, using the default (underlined) setting for other parameters; the non-exhaustive best-performing setting in each case is indicated in bold. isations for lexical variants, e.g. (bcuz, cause)), we modify our evaluation metric from Section 4 to evaluate the ranking at different points, using Discounted Cumulative Gain (DCG@N: Jarvelin and Kekalainen (2002)): DCG@N = rel1+XiN=2logr2el(i ) where reli again represents the frequency of the OOV, but it can be gain (a positive number) or loss (a negative number), depending on whether the ith pair is correct or incorrect. Because we also expect correct pairs to be ranked higher than incorrect pairs, DCG@N takes both factors into account. Given the generated pairs and the evaluation metric, we first consider three baselines: no re-ranking (i.e., the final ranking is that of the contextual similarity scores), and re-rankings of the pairs based on the frequencies of the OOVs in the Twitter corpus, and the IV unigram frequencies in the Google Web 1T corpus (Brants and Franz, 2006) to get less-noisy frequency estimates. We also compared a variety of re-rankings based on a number of string similarity measures that have been previously considered in normalisation work (reviewed in Section 2). We experiment with standard edit distance (Levenshtein, 1966), edit distance over double metaphone codes (phonetic edit distance: (Philips, 2000)), longest common subsequence ratio over the consonant edit distance of the paired words (hereafter, denoted as 426 consonant edit distance: (Contractor et al., 2010)), and a string subsequence kernel (Lodhi et al., 2002). In Figure 1, we present the DCG@N results for each of our ranking methods at different rank cutoffs. Ranking by OOV frequency is motivated by the assumption that lexical variants are frequently used by social media users. This is confirmed by our findings that lexical pairs like (goin, going) and (nite, night) are at the top of the ranking. However, many proper nouns and named entities are also used frequently and ranked at the top, mixed with lexical variants like (Facebook, speech) and (Youtube, web). In ranking by IV word frequency, we assume the lexical variants are usually derived from frequently-used IV equivalents, e.g. (abou, about). However, many less-frequent lexical variant types have high-frequency (IV) normalisations. For instance, the highest-frequency IV word the has more than 40 OOV lexical variants, such as tthe and thhe. These less-frequent types occupy the top positions, reducing the cumulative gain. Compared with these two baselines, ranking by default contextual similarity scores delivers promising results. It successfully ranks many more intuitive normalisation pairs at the top, such as (2day, today) and (wknd, weekend), but also ranks some incorrect pairs highly, such as (needa, gotta). The string similarity-based methods perform better than our baselines in general. Through manual analysis, we found that standard edit distance ranking is fairly accurate for lexical variants with low edit distance to their standard forms, but fails to identify heavily-altered variants like (tmrw, tomorrow). Consonant edit distance is similar to standard edit distance, but places many longer words at the top of the ranking. Edit distance over double metaphone codes (phonetic edit distance) performs particularly well for lexical variants that include character repetitions commonly used for emphasis on Twitter because such repetitions do not typically alter the phonetic codes. Compared with the other methods, the string subsequence kernel delivers encouraging results. It measures common character subsequences of length n between (OOV, IV) pairs. Because it is computationally expensive to calculate similarity for larger n, we choose n=2, following Gouws et al. (201 1). As N (the lexicon size cut-off) increases, the performance drops more slowly than the other meth— — ods. Although this method fails to rank heavilyaltered variants such as (4get,forget) highly, it typically works well for longer words. Given that we focus on longer OOVs (specifically those longer than 4 characters), this ultimately isn’t a great handicap. 6 Evaluation Given the re-ranked pairs from Section 5, here we apply them to a token-level normalisation task using the normalisation dataset of Han and Baldwin (201 1). 6.1 Metrics We evaluate using the standard evaluation metrics of precision (P), recall (R) and F-score (F) as detailed below. We also consider the false alarm rate (FA) and word error rate (WER), also as shown below. FA measures the negative effects of applying normalisation; a good approach to normalisation should not (incorrectly) normalise tokens that are already in their standard form and do not require normalisation.6 WER, like F-score, shows the overall benefits of normalisation, but unlike F-score, measures how many token-level edits are required for the output to be the same as the ground truth data. In general, dictionaries with a high F-score/low WER and low FA 6FA + P ≤ 1because some lexical variants might be incorrectly Ano +rm Pa ≤lise 1d b. 427 are preferable. P = R= F = FA = WER = # cor#re nctolrym naolrismedal tioskeden toskens # to ckoernresc rtelyqu niori nmga nloisremda tloiskaetniosn P2P +R R # inco#rr encotrlmya nliosremda tloikseedns tokens # token edits n#ee adlletd o akfetnesr normalisation 6.2 Results We select the three best re-ranking methods, and best cut-off N for each method, based on the highest DCG@N value for a given method over the development data, as presented in Figure 1. Namely, they are string subsequence kernel (S-dict, N=40,000), double metaphone edit distance (DMdict, N=10,000) and default contextual similarity without re-ranking (C-dict, N=10,000).7 We evaluate each of the learned dictionaries in Table 3. We also compare each dictionary with the performance of the manually-constructed Internet slang dictionary (HB-dict) used by Han and Baldwin (201 1), the small automatically-derived dictionary of Gouws et al. (201 1) (GHM-dict), and combinations of the different dictionaries. In addition, the contribution of these dictionaries in hybrid normalisation approaches is also presented, in which we first normalise OOVs using a given dictionary (combined or otherwise), and then apply the normalisation method of Gouws et al. (201 1) based on consonant edit distance (GHM-norm), or the approach of Han and Baldwin (201 1) based on the summation of many unsupervised approaches (HB-norm), to the remaining OOVs. Results are shown in Table 3, and discussed below. 6.2.1 Individual Dictionaries Overall, the individual dictionaries derived by the re-ranking methods (DM-dict, S-dict) perform bet- 7We also experimented with combining ranks using Mean Reciprocal Rank. However, the combined rank didn’t improve performance on the development data. We plan to explore other ranking aggregation methods in future work. 1 3 5 7 9 11 31 51 71 91 N cut−offs Figure 1: Re-ranking based on different string similarity methods. ter than that based on contextual similarity (C-dict) in terms of precision and false alarm rate, indicating the importance of re-ranking. Even though C-dict delivers higher recall indicating that many lexical variants are correctly normalised this is offset by its high false alarm rate, which is particularly undesirable in normalisation. Because S-dict has better performance than DM-dict in terms of both F-score and WER, and a much lower false alarm rate than C-dict, subsequent results are presented using S-dict only. — — Both HB-dict and GHM-dict achieve better than 90% precision with moderate recall. Compared to these methods, S-dict is not competitive in terms of either precision or recall. This result seems rather discouraging. However, considering that S-dict is an automatically-constructed dictionary targeting lexical variants of varying frequency, it is not surprising that the precision is worse than that of HB-dict which is manually-constructed and GHM-dict which includes entries only for more-frequent OOVs for which distributional similarity is more accurate. Additionally, the recall of S-dict is hampered by the — — — 428 restriction on lexical variant token length of 4 characters. 6.2.2 Combined Dictionaries Next we look to combining HB-dict, GHM-dict and S-dict. In combining the dictionaries, a given OOV word can be listed with different standard forms in different dictionaries. In such cases we use the following preferences for dictionaries motivated by our confidence in the normalisation pairs — of the dictionaries to resolve conflicts: HB-dict > GHM-dict > S-dict. When we combine dictionaries in the second section of Table 3, we find that they contain complementary information: in each case the recall and F-score are higher for the combined dictionary than any of the individual dictionaries. The combination of HB-dict+GHM-dict produces only a small improvement in terms of F-score over HBdict (the better-performing dictionary) suggesting that, as claimed, HB-dict and GHM-dict share many frequent normalisation pairs. HB-dict+S-dict and GHM-dict+S-dict, on the other hand, improve sub— MethodPrecisionRecallF-ScoreFalse AlarmWord Error Rate C-dict0.4740.2180.2990.2980.103 DM-dict S-dict HB-dict GHM-dict 0.727 0.700 0.915 0.982 0.106 0.179 0.435 0.319 0.185 0.285 0.590 0.482 0.145 0.162 0.048 0.000 0.102 0.097 0.066 0.076 HB-dict+S-dict0.8400.6010.7010.0900.052 GHM-dict+S-dict HB-dict+GHM-dict HB-dict+GHM-dict+S-dict 0.863 0.920 0.847 0.498 0.465 0.630 0.632 0.618 0.723 0.072 0.045 0.086 0.061 0.063 0.049 GHM-dict+GHM-norm0.3380.5780.4270.4580.135 HB-dict+GHM-dict+S-dict+GHM-norm HB-dict+HB-norm HB-dict+GHM-dict+S-dict+HB-norm 0.406 0.515 0.527 0.715 0.771 0.789 0.518 0.618 0.632 0.468 0.332 0.332 0.124 0.081 0.079 Table 3: Normalisation results using our derived dictionaries (contextual similarity (C-dict); double metaphone rendering (DM-dict); string subsequence kernel scores (S-dict)), the dictionary of Gouws et al. (201 1) (GHM-dict), the Internet slang dictionary (HB-dict) from Han and Baldwin (201 1), and combinations of these dictionaries. In addition, we combine the dictionaries with the normalisation method of Gouws et al. (201 1) (GHM-norm) and the combined unsupervised approach of Han and Baldwin (201 1) (HB-norm). stantially over HB-dict and GHM-dict, respectively, indicating that S-dict contains markedly different entries to both HB-dict and GHM-dict. The best Fscore and WER are obtained using the combination of all three dictionaries, HB-dict+GHM-dict+S-dict. Furthermore, the difference between the results using HB-dict+GHM-dict+S-dict and HB-dict+GHMdict is statistically significant (p < 0.01), based on the computationally-intensive Monte Carlo method of Yeh (2000), demonstrating the contribution of Sdict. 6.2.3 Hybrid Approaches The methods of Gouws et al. (201 1) (i.e. GHM-dict+GHM-norm) and Han and Baldwin (201 1) (i.e. HB-dict+HB-norm) have lower precision and higher false alarm rates than the dictionarybased approaches; this is largely caused by lexical variant detection errors.8 Using all dictionaries in combination with these methods HB-dict+GHM-dict+S-dict+GHM-norm and HBdict+GHM-dict+S-dict+HB-norm gives some improvements, but the false alarm rates remain high. Despite the limitations of a pure dictionary-based approach to normalisation discussed in Section 3.1 the current best practical approach to normal— — — — 8Here we report results that do not assume perfect detection of lexical variants, unlike the original published results in each case. 429 Error typeOOVDSitcat.ndard fGoromld (a) pluralsplayeplayersplayer (b) negation unlike like dislike (c) possessives anyones anyone anyone ’s (d) correct OOVs iphone phone iphone (e) test data errors durin during durin (f) ambiguity siging signing singing Table 4: Error types in the combined dictionary (HBdict+GHM-dict+S-dict) isation is to use a lexicon, combining hand-built and automatically-learned normalisation dictionaries. 6.3 Discussion and Error Analysis We first manually analyse the errors in the combined dictionary (HB-dict+GHM-dict+S-dict) and give examples of each error type in Table 4. The most frequent word errors are caused by slight morphologi- cal variations, including plural forms (a), negations (b), possessive cases (c), and OOVs that are correct and do not require normalisation (d). In addition, we also notice some missing annotations where lexical variants are skipped by human annotations but captured by our method (e). Ambiguity (f) definitely exists in longer OOVs, however, these cases do not appear to have a strong negative impact on the normalisation performance. An example of a remainLength cut-off (N)#VariantsPrecisionRecall (≥ N)Recall (all)False Alarm ≥45560.700Rec0al.l3 8(≥1 N)0.1790.162 ≥≥54 382 0.814 0.471 0.152 0.122 ≥≥65 254 0.804 0.484 0.104 0.131 ≥≥76 138 0.793 0.471 0.055 0.122 ≥71380.7930.4710.0550.122 Table 5: S-dict normalisation results broken down according to OOV token length. Recall is presented both over the subset of instances of length ≥ N in the data (“Recall (≥ N)”), and over the entirety of the dataset (“Recall (all)”); “su#bVsaertia onftis n” sitsa tnhcee snu omfb leenrg othf t≥ok Nen iinns tthaenc deast ao f( “tRhee cinadllic (a≥ted N length idn o othveer rt tehset d eanttaisreetty. ing miscellaneous error is bday “birthday”, which is mis-normalised as day. To further study the influence of OOV word length relative to the normalisation performance, we conduct a fine-grained analysis of the performance of the derived dictionary (S-dict) in Table 5, broken down across different OOV word lengths. The results generally support our hypothesis that our method works better for longer OOV words. The derived dictionary is much more reliable for longer tokens (length 5, 6, and 7 characters) in terms of precision and false alarm. Although the recall is relatively modest, in the future we intend to improve recall by mining more normalisation pairs from larger collections of microblog data. 7 Conclusions and Future Work In this paper, we describe a method for automatically constructing a normalisation dictionary that supports normalisation of microblog text through direct substitution of lexical variants with their standard forms. After investigating the impact of different distributional and string similarity methods on the quality of the dictionary, we present experimental results on a standard dataset showing that our proposed methods acquire high quality (lexical variant, standard form) pairs, with reasonable coverage, and achieve state-of-the-art end-toend lexical normalisation performance on a realworld token-level task. Furthermore, this dictionarylookup method combines the detection and normalisation of lexical variants into a simple, lightweight solution which is suitable for processing of highvolume microblog feeds. In the future, we intend to improve our dictionary by leveraging the constantly-growing volume of microblog data, and considering alternative ways to combine distributional and string similarity. In addi430 tion to direct evaluation, we also want to explore the benefits of applying normalisation for downstream social media text processing applications, e.g. event detection. Acknowledgements We would like to thank the three anonymous reviewers for their insightful comments, and Stephan Gouws for kindly sharing his data and discussing his work. NICTA is funded by the Australian government as represented by Department of Broadband, Communication and Digital Economy, and the Australian Research Council through the ICT centre of Excellence programme. References AiTi Aw, Min Zhang, Juan Xiao, and Jian Su. 2006. A phrase-based statistical model for SMS text normalization. In Proceedings of COLING/ACL 2006, pages 33–40, Sydney, Australia. Edward Benson, Aria Haghighi, and Regina Barzilay. 2011. Event discovery in social media feeds. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 389–398, Portland, Oregon, USA. Thorsten Brants and Alex Franz. 2006. Web 1T 5-gram Version 1. Eric Brill and Robert C. Moore. 2000. An improved error model for noisy channel spelling correction. In Proceedings of the 38th Annual Meeting of the Association for Computational Linguistics, pages 286–293, Hong Kong. Monojit Choudhury, Rahul Saraf, Vijit Jain, Animesh Mukherjee, Sudeshna Sarkar, and Anupam Basu. 2007. Investigation and modeling of the structure of texting language. International Journal on Document Analysis and Recognition, 10: 157–174. Danish Contractor, Tanveer A. Faruquie, and L. Venkata Subramaniam. 2010. Unsupervised cleansing of noisy text. In Proceedings of the 23rd International Conference on Computational Linguistics (COLING 2010), pages 189–196, Beijing, China. Paul Cook and Suzanne Stevenson. 2009. An unsu- pervised model for text message normalization. In CALC ’09: Proceedings of the Workshop on Computational Approaches to Linguistic Creativity, pages 71– 78, Boulder, USA. Jennifer Foster, O¨zlem C ¸etinoglu, Joachim Wagner, Joseph L. Roux, Stephen Hogan, Joakim Nivre, Deirdre Hogan, and Josef van Genabith. 2011. #hardtoparse: POS Tagging and Parsing the Twitterverse. In Analyzing Microtext: Papers from the 2011 AAAI Workshop, volume WS-1 1-05 of AAAI Workshops, pages 20–25, San Francisco, CA, USA. Kevin Gimpel, Nathan Schneider, Brendan O’Connor, Dipanjan Das, Daniel Mills, Jacob Eisenstein, Michael Heilman, Dani Yogatama, Jeffrey Flanigan, and Noah A. Smith. 2011. Part-of-speech tagging for Twitter: Annotation, features, and experiments. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 42–47, Portland, Oregon, USA. Roberto Gonz a´lez-Ib ´a n˜ez, Smaranda Muresan, and Nina Wacholder. 2011. Identifying sarcasm in Twitter: a closer look. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 581–586, Portland, Oregon, USA. Stephan Gouws, Dirk Hovy, and Donald Metzler. 2011. Unsupervised mining of lexical variants from noisy text. In Proceedings of the First workshop on Unsupervised Learning in NLP, pages 82–90, Edinburgh, Scotland, UK. Bo Han and Timothy Baldwin. 2011. Lexical normalisation of short text messages: Makn sens a #twitter. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 368–378, Portland, Oregon, USA. K. Jarvelin and J. Kekalainen. 2002. Cumulated gainbased evaluation of IR techniques. ACM Transactions on Information Systems, 20(4). Long Jiang, Mo Yu, Ming Zhou, Xiaohua Liu, and Tiejun Zhao. 2011. Target-dependent Twitter sentiment classification. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 151–160, Portland, Oregon, USA. Joseph Kaufmann and Jugal Kalita. 2010. Syntactic normalization of Twitter messages. In International Con431 ference on Natural Language Processing, Kharagpur, India. S. Kullback and R. A. Leibler. 1951. On information and sufficiency. Annals of Mathematical Statistics, 22:49– 86. John D. Lafferty, Andrew McCallum, and Fernando C. N. Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. In Proceedings ofthe Eighteenth International Conference on Machine Learning, pages 282–289, San Francisco, CA, USA. Vladimir I. Levenshtein. 1966. Binary codes capable of correcting deletions, insertions, and reversals. Soviet Physics Doklady, 10:707–710. Mu Li, Yang Zhang, Muhua Zhu, and Ming Zhou. 2006. Exploring distributional similarity based models for query spelling correction. In Proceedings of COLING/ACL 2006, pages 1025–1032, Sydney, Australia. Jianhua Lin. 1991. Divergence measures based on the shannon entropy. IEEE Transactions on Information Theory, 37(1): 145–151. Dekang Lin. 1998. Automatic retrieval and clustering of similar words. In Proceedings of the 36th Annual Meeting of the ACL and 1 International Con7th ference on Computational Linguistics (COLING/ACL98), pages 768–774, Montreal, Quebec, Canada. Fei Liu, Fuliang Weng, Bingqing Wang, and Yang Liu. 2011a. Insertion, deletion, or substitution? normalizing text messages without pre-categorization nor supervision. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 71–76, Portland, Oregon, USA. Xiaohua Liu, Shaodian Zhang, Furu Wei, and Ming Zhou. 2011b. Recognizing named entities in tweets. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies (ACL-HLT 2011), pages 359–367, Portland, Oregon, USA. Fei Liu, Fuliang Weng, and Xiao Jiang. 2012. A broadcoverage normalization system for social media language. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (ACL 2012), Jeju, Republic of Korea. Huma Lodhi, Craig Saunders, John Shawe-Taylor, Nello Cristianini, and Chris Watkins. 2002. Text classification using string kernels. J. Mach. Learn. Res., 2:419– 444. Marco Lui and Timothy Baldwin. 2011. Cross-domain feature selection for language identification. In Proceedings of the 5th International Joint Conference on Natural Language Processing (IJCNLP 2011), pages 553–561, Chiang Mai, Thailand. Brendan O’Connor, Michel Krieger, and David Ahn. 2010. TweetMotif: Exploratory search and topic summarization for Twitter. In Proceedings of the 4th International Conference on Weblogs and Social Media (ICWSM 2010), pages 384–385, Washington, USA. Lawrence Philips. 2000. The double metaphone search algorithm. C/C++ Users Journal, 18:38–43. Lawrence R. Rabiner. 1989. A tutorial on hidden Markov models and selected applications in speech recognition. Proceedings of the IEEE, 77(2):257–286. Alan Ritter, Colin Cherry, and Bill Dolan. 2010. Unsupervised modeling of Twitter conversations. In Proceedings of Human Language Technologies: The 11th Annual Conference of the North American Chap- ter of the Association for Computational Linguistics (NAACL-HLT 2010), pages 172–180, Los Angeles, USA. Alan Ritter, Sam Clark, Mausam, and Oren Etzioni. 2011. Named entity recognition in tweets: An experimental study. In Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing (EMNLP 2011), pages 1524–1534, Edinburgh, Scotland, UK. Takeshi Sakaki, Makoto Okazaki, and Yutaka Matsuo. 2010. Earthquake shakes Twitter users: real-time event detection by social sensors. In Proceedings of the 19th International Conference on the World Wide Web (WWW 2010), pages 851–860, Raleigh, North Carolina, USA. Crispin Thurlow. 2003. Generation txt? The sociolinguistics of young people’s text-messaging. Discourse Analysis Online, 1(1). Kristina Toutanova and Robert C. Moore. 2002. Pronunciation modeling for improved spelling correction. In Proceedings of the 40th Annual Meeting of the ACL and 3rd Annual Meeting of the NAACL (ACL-02), pages 144–15 1, Philadelphia, USA. Official Blog Twitter. 2011. 200 million tweets per day. Retrived at August 17th, 2011. Jianshu Weng and Bu-Sung Lee. 2011. Event detection in Twitter. In Proceedings of the 5th International Conference on Weblogs and Social Media (ICWSM 2011), Barcelona, Spain. Zhenzhen Xue, Dawei Yin, and Brian D. Davison. 2011. Normalizing microtext. In Proceedings of the AAAI11 Workshop on Analyzing Microtext, pages 74–79, San Francisco, USA. Alexander Yeh. 2000. More accurate tests for the statistical significance of result differences. In Proceedings of the 23rd International Conference on Computational Linguistics (COLING 2010), pages 947–953, Saarbr¨ ucken, Germany. 432
5 0.30806351 5 emnlp-2012-A Discriminative Model for Query Spelling Correction with Latent Structural SVM
Author: Huizhong Duan ; Yanen Li ; ChengXiang Zhai ; Dan Roth
Abstract: Discriminative training in query spelling correction is difficult due to the complex internal structures of the data. Recent work on query spelling correction suggests a two stage approach a noisy channel model that is used to retrieve a number of candidate corrections, followed by discriminatively trained ranker applied to these candidates. The ranker, however, suffers from the fact the low recall of the first, suboptimal, search stage. This paper proposes to directly optimize the search stage with a discriminative model based on latent structural SVM. In this model, we treat query spelling correction as a multiclass classification problem with structured input and output. The latent structural information is used to model the alignment of words in the spelling correction process. Experiment results show that as a standalone speller, our model outperforms all the baseline systems. It also attains a higher recall compared with the noisy channel model, and can therefore serve as a better filtering stage when combined with a ranker.
6 0.3037768 123 emnlp-2012-Syntactic Transfer Using a Bilingual Lexicon
7 0.30309543 14 emnlp-2012-A Weakly Supervised Model for Sentence-Level Semantic Orientation Analysis with Multiple Experts
8 0.30184376 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers
10 0.2989085 71 emnlp-2012-Joint Entity and Event Coreference Resolution across Documents
11 0.29883215 52 emnlp-2012-Fast Large-Scale Approximate Graph Construction for NLP
12 0.29854837 82 emnlp-2012-Left-to-Right Tree-to-String Decoding with Prediction
13 0.29732144 23 emnlp-2012-Besting the Quiz Master: Crowdsourcing Incremental Classification Games
14 0.29646394 107 emnlp-2012-Polarity Inducing Latent Semantic Analysis
15 0.29615483 64 emnlp-2012-Improved Parsing and POS Tagging Using Inter-Sentence Consistency Constraints
16 0.29577246 45 emnlp-2012-Exploiting Chunk-level Features to Improve Phrase Chunking
17 0.29492438 70 emnlp-2012-Joint Chinese Word Segmentation, POS Tagging and Parsing
18 0.29488075 124 emnlp-2012-Three Dependency-and-Boundary Models for Grammar Induction
19 0.29485568 92 emnlp-2012-Multi-Domain Learning: When Do Domains Matter?
20 0.29477033 89 emnlp-2012-Mixed Membership Markov Models for Unsupervised Conversation Modeling