emnlp emnlp2010 emnlp2010-55 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Danish Contractor ; Govind Kothari ; Tanveer Faruquie ; L V Subramaniam ; Sumit Negi
Abstract: Recent times have seen a tremendous growth in mobile based data services that allow people to use Short Message Service (SMS) to access these data services. In a multilingual society it is essential that data services that were developed for a specific language be made accessible through other local languages also. In this paper, we present a service that allows a user to query a FrequentlyAsked-Questions (FAQ) database built in a local language (Hindi) using Noisy SMS English queries. The inherent noise in the SMS queries, along with the language mismatch makes this a challenging problem. We handle these two problems by formulating the query similarity over FAQ questions as a combinatorial search problem where the search space consists of combinations of dictionary variations of the noisy query and its top-N translations. We demonstrate the effectiveness of our approach on a real-life dataset.
Reference: text
sentIndex sentText sentNum sentScore
1 com Abstract Recent times have seen a tremendous growth in mobile based data services that allow people to use Short Message Service (SMS) to access these data services. [sent-5, score-0.206]
2 In this paper, we present a service that allows a user to query a FrequentlyAsked-Questions (FAQ) database built in a local language (Hindi) using Noisy SMS English queries. [sent-7, score-0.184]
3 We handle these two problems by formulating the query similarity over FAQ questions as a combinatorial search problem where the search space consists of combinations of dictionary variations of the noisy query and its top-N translations. [sent-9, score-0.654]
4 1 Introduction There has been a tremendous growth in the number of new mobile subscribers in the recent past. [sent-11, score-0.155]
5 4 Millions of users of instant messaging (IM) services and short message service (SMS) generate electronic content in a dialect that does not adhere to conventional grammar, punctuation and spelling standards. [sent-24, score-0.159]
6 Typical question answering systems are built for use with languages which are free from such errors. [sent-26, score-0.129]
7 Unlike other automatic question answering systems that focus on searching answers from a given text collection, Q&A; archive (Xue et al. [sent-29, score-0.146]
8 The main task is then to identify the best matching question to retrieve the relevant answer (Sneiders, 1999) (Song et al. [sent-42, score-0.15]
9 The high level of noise in SMS queries makes this a difficult problem (Kothari et al. [sent-44, score-0.152]
10 In this paper we present a FAQ-based question answering system over a SMS interface that solves this problem for two languages. [sent-48, score-0.13]
11 Multi-lingual question answering and information retrieval has been studied in the past (Sekine and Grishman, 2003)(Cimiano et al. [sent-50, score-0.14]
12 In the two language setting, it involves building a machine translation system engine and using it such that the question answering system built for a single language can be used. [sent-53, score-0.201]
13 Typical statistical machine translation systems use large parallel corpora to learn the translation probabilities (Brown et al. [sent-54, score-0.179]
14 Since the translation systems are not trained on SMS language they perform very poorly when translating noisy SMS language. [sent-57, score-0.198]
15 Parallel corpora comprising noisy sentences in one language and clean sentences in another language are not available and it would be hard to build such large parallel corpora to train a machine translation system. [sent-58, score-0.32]
16 88 Removal of noise from SMS without the use of parallel data has been studied but the methods used are highly dependent on the language model and the degree of noise present in the SMS (Contractor et al. [sent-69, score-0.183]
17 Thus, the translation ofa cleaned SMS, into a second language, will not be very accurate and it would not give good results if such a translated SMS is used to query an FAQ collection. [sent-72, score-0.25]
18 These noise-correction methods return a list of candidate terms for a given noisy token (E. [sent-74, score-0.267]
19 Also , naively replacing the noisy token in the SMS query with the top matching candidate term gives poor performance as shown by our experiments. [sent-78, score-0.436]
20 In this paper we address the challenges arising when building a cross language FAQ-based question answering system over an SMS interface. [sent-80, score-0.185]
21 Our method handles noisy representation of questions in a source language to retrieve answers across target languages. [sent-81, score-0.334]
22 To the best of our knowledge we are the first to address issues in noisy SMS based cross-language FAQ retrieval. [sent-84, score-0.126]
23 The matching is assisted by a source dictionary De consisting of clean terms in e constructed from a general English dictionary and a domain dictionary of target language Dh built from all the terms appearing gine Q lahn. [sent-95, score-0.541]
24 g Fagore a token si in the SMS input, term te nin Q dictionary De and term th in dictionary Dh we dne dfiinceti a cross-lingual similarity measure α(th, te, si) that measures the extent to which term si matches th using the clean term te. [sent-96, score-1.156]
25 We consider th a cross lingual variant of si if for any te the cross language similarity measure α(th, te, si) > ? [sent-97, score-0.643]
26 We define a weight function ω(th, te, si) using the Qˆh cross lingual similarity measure and the inverse document frequency (idf) of th in the target language FAQ corpus. [sent-100, score-0.412]
27 We also define a scoring function to assign a score to each question in the corpus Qh using tshigen weight efu tnoc etaicohn. [sent-101, score-0.126]
28 For each token si, the scoring function choos∈es Qthe term from Qh having the maximum weight using possible clean representations of si; then the weight of the n chosen terms are summed up to get the score. [sent-103, score-0.354]
29 The score measures how closely the question in FAQ matches the noisy SMS string Se using the composite weights of individual tokens. [sent-104, score-0.219]
30 3 Qˆh having Noise removal from queries In order to process the noisy SMS input we first have to map noisy tokens in Se to the possible correct lex- ical representations. [sent-106, score-0.33]
31 We use a similarity measure to map the noisy tokens to their clean lexical representations. [sent-107, score-0.291]
32 This way we limit the possible variants for a particular noisy token The Longest Common Subsequence Ratio (LCSRatio) (Melamed et al. [sent-111, score-0.209]
33 Since in the SMS scenario, the dictionary term will always be longer than the SMS token, the denominator of LCSRatio is taken as the length of the dictionary term. [sent-113, score-0.294]
34 Hence the two dictionary terms “good” and “guided” have the same LCSRatio of 0. [sent-120, score-0.133]
35 Hence, for a given SMS token “byk”, the similarity measure of word “bike“ is higher than that of “break”. [sent-128, score-0.161]
36 4 Cross lingual similarity Once we have potential candidates which are the likely disambiguated representations of the noisy term, we map these candidates to appropriate terms in the target language. [sent-129, score-0.375]
37 We use a statistical dictionary to achieve this cross lingual mapping. [sent-130, score-0.339]
38 1 Statistical Dictionary In order to build a statistical dictionary we use the statistical translation model proposed in (Brown et al. [sent-132, score-0.183]
39 Yi=1jX=0 (2) Here the word translation model τ(th |te) gives the probability of translating the source t|etrm to target term and the alignment model a(j |i, m, l) gives the probability oef a translating tohed source t,emrm,l a)t g position ito a target position j. [sent-147, score-0.188]
40 Given a clean term tie in source language we get all the corresponding terms T = {t1h, . [sent-149, score-0.205]
41 We rank these terms according to the probability given by the word translation model τ(th |te) and consider only those tar- get terms that are part of domain dictionary i. [sent-159, score-0.273]
42 2 Cross lingual similarity measure For each term si in SMS input query, we find all the clean terms te in source dictionary De for which similarity measure γ(te, si) > φ. [sent-163, score-0.804]
43 Faroyr eDach of these term te, we find the cross lingual similar terms Tte using th,e w weo firdnd dtr tahnesl cartoiosns lminogdueall. [sent-164, score-0.322]
44 Wimei compute Tthe cross lingual similarity measure between these terms as α(si, te, th) = γ(te, si) . [sent-165, score-0.328]
45 τ(th, te) (3) The measure selects those terms in target language that have high probability of being translated from a noisy term through one or more valid clean terms. [sent-166, score-0.351]
46 3 Cross lingual similarity weight We combine the idf and the cross lingual similarity measure to define the cross lingual weight function ω(th, te, si) as ω(th, te, si) = α(th, te, si) . [sent-168, score-0.871]
47 For example for a given noisy token “bck” if a word translation model produces a translation output “wapas” (as in came back) or “peet” or “qamar” (as in back pain) then idf will weigh “peet” more as it is relatively more discriminative compared to “wapas” which is used frequently. [sent-171, score-0.419]
48 1 Indexing Our algorithm operates at a token level and its corresponding cross lingual variants. [sent-174, score-0.311]
49 It is therefore necessary to be able to retrieve all questions Qthh that contain a given target language term th. [sent-175, score-0.217]
50 org/java/docs/ The cross lingual similarity weight calculation requires the idf for a given term th. [sent-181, score-0.455]
51 The cross lingual similarity measure calculation requires the word translation probability for a given term te. [sent-185, score-0.45]
52 For every te in dictionary De, we store Tte in a hashmap that contains a list of terms in the target language along with their statistically determined translation probability τ(th |te) > ε, where th ∈ Dh. [sent-186, score-0.432]
53 S∈in Dce the query and the FAQs use terms from different languages, the computation of IDF becomes a challenge (Pirkola, 1998) (Oard et al. [sent-187, score-0.145]
54 We on the other hand rely on a statistical dictionary that has translation probabilities. [sent-190, score-0.183]
55 We therefore calculate IDFs for target language term (translation) and use it in the weight measure calculation. [sent-192, score-0.149]
56 A list Lie is created for each token si using terms in the monolingual dictionary De. [sent-202, score-0.341]
57 A term te from De is included in Lie if it satisfies the threshold condition γ(te, si) > φ (5) The threshold value φ is determined experimentally. [sent-204, score-0.238]
58 For every te ∈ Lie we retrieve Tte and then retrieve the idf scores f oLr every th ∈ Tte . [sent-205, score-0.339]
59 Using the word translation probabilities and ∈the T idf score we compute the cross lingual similarity weight to create a new list Lih. [sent-206, score-0.518]
60 A term th is included in the list only if τ(th |te) > 0. [sent-207, score-0.159]
61 If more than one term te has the same translation th, then th can occur more than once in a given list. [sent-209, score-0.313]
62 The terms th in Lih are sorted in decreasing order of their similarity weights. [sent-211, score-0.165]
63 3, for each token we create a list of possible correct dictionary words by dictionary look up. [sent-214, score-0.341]
64 Thus for token “cst” we get dictionary words lik “cost, cast, case, close”. [sent-215, score-0.218]
65 For each dictionary word we get a set of possible words in Hindi by looking at statistical translation table. [sent-216, score-0.207]
66 , Lnh containing terms from the domain dictionary and sorted according to their cross lingual weights as explained in the previous section. [sent-223, score-0.397]
67 A naive approach would be to query the index using each term appearing in all Lih to build a Collection set C of questions. [sent-224, score-0.195]
68 h iWnge compute the score of each question in C using Score(Q) and the question ewacithh highest score issi ntrge aStecdo as nHdo twheever the naive approach suffers from high runtime cost. [sent-227, score-0.186]
69 In each iteration, it picks the term that has maximum weight among all the terms appearing in the Qˆh QQˆ)h . [sent-232, score-0.127]
70 As the lists are sorted in the descending order of the weights, this amounts to picking the maximum weight term amongst the first terms of the n lists. [sent-237, score-0.163]
71 After this the chosen term th is removed from the list and the next iteration is carried out. [sent-241, score-0.159]
72 u 92 Figure 4: Search Algorithm with Pruning 6 Experiments To evaluate our system we used noisy English SMS queries to query a collection of 10, 000 Hindi FAQs. [sent-259, score-0.357]
73 We found 60 SMS queries created by the evaluators, had answers in our FAQ collection and we designated these as the in-domain queries. [sent-263, score-0.148]
74 Figure 5: Sample SMS queries 2005) and the rest from the “out-of-domain” queries created by the human evaluators. [sent-269, score-0.156]
75 Our objective was to retrieve the correct Hindi FAQ response given a noisy English SMS query. [sent-272, score-0.178]
76 A given English SMS query was matched against the list of indexed FAQs and the best matching FAQ was returned by the Pruning Algorithm described in Section 5. [sent-273, score-0.191]
77 A score of 1 was assigned if the retrieved answer was indeed the response to the posed SMS query else we assigned a score of 0. [sent-274, score-0.177]
78 In case of out of domain queries a score of 1 was assigned if the output was NULL else we assigned a score of 0. [sent-275, score-0.132]
79 Since an MT system trained solely on a collection of sentences would not be very accurate in translating questions, we trained the system on an EnglishHindi parallel question corpus. [sent-283, score-0.131]
80 As it was difficult to find a large collection of parallel text consisting of questions, we created a small collection of par- allel questions using 240 FAQs and multiplied them to create a parallel corpus of 50, 000 sentences. [sent-284, score-0.201]
81 For our experiments the lexical translation probabilities generated by Moses toolkit were used to build the word translation model. [sent-288, score-0.144]
82 The Hindi FAQ collection was indexed using Lucene and a domain dictionary Dh was created from the Hindi words in the FAQ collection. [sent-292, score-0.141]
83 2 Experiment 2 For Experiment 2 the noisy SMS query was cleaned using the following approach. [sent-304, score-0.304]
84 Given a noisy token in the SMS query it’s similarity (Equation 1) with each word in the Dictionary is calculated. [sent-305, score-0.388]
85 The noisy token is replaced with the Dictionary word with the maximum similarity score. [sent-306, score-0.265]
86 For each token in the cleaned English SMS query, we create a list of possible Hindi translations of the token using the statistical translation table. [sent-308, score-0.329]
87 3 Experiment 3 In this experiment, for each token in the noisy En- glish SMS we obtain a list of possible English variations. [sent-313, score-0.245]
88 The reciprocal rank of a query response is the multiplicative inverse of the rank of the first correct answer. [sent-326, score-0.21]
89 The mean reciprocal rank is the average of the reciprocal ranks of results for a sample of queries Q. [sent-327, score-0.186]
90 As the SMS based FAQ retrieval system will be used via mobile phones where screen size is a major constraint it is crucial to have the correct result on the top. [sent-329, score-0.133]
91 From Table 3 we can see the difference in perplexity for noisy and clean SMS data for the English FAQ data-set. [sent-348, score-0.259]
92 “hw 2 prvnt typhd” we manually created a clean SMS query “how to prevent typhoid”. [sent-352, score-0.21]
93 A character level language model using the questions in the clean English FAQ dataset was created to quantify the level of noise in our SMS dataset. [sent-353, score-0.232]
94 We computed the perplexity of the language model on clean and noisy SMS queries. [sent-354, score-0.259]
95 Automatic cross-lingual QA over SMS is challenging because of inherent noise in the query and the lack of cross language resources for noisy processing. [sent-358, score-0.402]
96 In this paper we present a cross-language FAQ retrieval system that handles the inherent noise in source language to retrieve FAQs in a target language. [sent-359, score-0.205]
97 Our system does not require an end-to-end machine translation system and can be implemented using a simple dictionary which can be static or constructed statistically using a moderate sized parallel corpus. [sent-360, score-0.218]
98 We present an efficient algorithm to search and match the best question in the large FAQ corpus of target language for a noisy input question. [sent-362, score-0.236]
99 Optimizing predictive text entry for short message service on mobile phones. [sent-413, score-0.181]
100 In Proceedings of the 4th international conference on mobile technology, applications, and systems and the 1st international symposium on Computer human interaction in mobile technology, pp. [sent-435, score-0.198]
wordName wordTfidf (topN-words)
[('sms', 0.695), ('faq', 0.385), ('hindi', 0.187), ('lingual', 0.149), ('noisy', 0.126), ('query', 0.123), ('te', 0.118), ('dictionary', 0.111), ('faqs', 0.103), ('qh', 0.103), ('mobile', 0.099), ('gud', 0.09), ('si', 0.089), ('clean', 0.087), ('token', 0.083), ('cross', 0.079), ('queries', 0.078), ('services', 0.077), ('noise', 0.074), ('translation', 0.072), ('term', 0.072), ('questions', 0.071), ('question', 0.066), ('idf', 0.066), ('editdistancesms', 0.064), ('lcsratio', 0.064), ('mrr', 0.064), ('similarity', 0.056), ('cleaned', 0.055), ('retrieve', 0.052), ('th', 0.051), ('perplexity', 0.046), ('dh', 0.044), ('tte', 0.044), ('message', 0.044), ('reciprocal', 0.043), ('answering', 0.04), ('answers', 0.04), ('oard', 0.04), ('contractor', 0.038), ('faruquie', 0.038), ('kothari', 0.038), ('lcs', 0.038), ('lih', 0.038), ('negi', 0.038), ('qth', 0.038), ('subramaniam', 0.038), ('tanveer', 0.038), ('thk', 0.038), ('venkata', 0.038), ('service', 0.038), ('list', 0.036), ('sorted', 0.036), ('parallel', 0.035), ('retrieval', 0.034), ('weight', 0.033), ('ub', 0.033), ('sumit', 0.033), ('india', 0.033), ('matching', 0.032), ('guided', 0.032), ('collection', 0.03), ('thresholding', 0.03), ('tremendous', 0.03), ('pruning', 0.028), ('se', 0.028), ('score', 0.027), ('de', 0.027), ('acharyya', 0.026), ('byun', 0.026), ('choudhury', 0.026), ('cimiano', 0.026), ('fagin', 0.026), ('govind', 0.026), ('kobus', 0.026), ('kopparapu', 0.026), ('lnh', 0.026), ('neef', 0.026), ('peet', 0.026), ('pirkola', 0.026), ('prochasson', 0.026), ('roc', 0.026), ('schusteritsch', 0.026), ('smses', 0.026), ('subscribers', 0.026), ('tej', 0.026), ('wapas', 0.026), ('song', 0.026), ('get', 0.024), ('subsequence', 0.024), ('interface', 0.024), ('lie', 0.024), ('threshold', 0.024), ('built', 0.023), ('handles', 0.023), ('target', 0.022), ('measure', 0.022), ('rank', 0.022), ('search', 0.022), ('terms', 0.022)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999952 55 emnlp-2010-Handling Noisy Queries in Cross Language FAQ Retrieval
Author: Danish Contractor ; Govind Kothari ; Tanveer Faruquie ; L V Subramaniam ; Sumit Negi
Abstract: Recent times have seen a tremendous growth in mobile based data services that allow people to use Short Message Service (SMS) to access these data services. In a multilingual society it is essential that data services that were developed for a specific language be made accessible through other local languages also. In this paper, we present a service that allows a user to query a FrequentlyAsked-Questions (FAQ) database built in a local language (Hindi) using Noisy SMS English queries. The inherent noise in the SMS queries, along with the language mismatch makes this a challenging problem. We handle these two problems by formulating the query similarity over FAQ questions as a combinatorial search problem where the search space consists of combinations of dictionary variations of the noisy query and its top-N translations. We demonstrate the effectiveness of our approach on a real-life dataset.
2 0.08610142 74 emnlp-2010-Learning the Relative Usefulness of Questions in Community QA
Author: Razvan Bunescu ; Yunfeng Huang
Abstract: We present a machine learning approach for the task of ranking previously answered questions in a question repository with respect to their relevance to a new, unanswered reference question. The ranking model is trained on a collection of question groups manually annotated with a partial order relation reflecting the relative utility of questions inside each group. Based on a set of meaning and structure aware features, the new ranking model is able to substantially outperform more straightforward, unsupervised similarity measures.
3 0.085520163 73 emnlp-2010-Learning Recurrent Event Queries for Web Search
Author: Ruiqiang Zhang ; Yuki Konda ; Anlei Dong ; Pranam Kolari ; Yi Chang ; Zhaohui Zheng
Abstract: Recurrent event queries (REQ) constitute a special class of search queries occurring at regular, predictable time intervals. The freshness of documents ranked for such queries is generally of critical importance. REQ forms a significant volume, as much as 6% of query traffic received by search engines. In this work, we develop an improved REQ classifier that could provide significant improvements in addressing this problem. We analyze REQ queries, and develop novel features from multiple sources, and evaluate them using machine learning techniques. From historical query logs, we develop features utilizing query frequency, click information, and user intent dynamics within a search session. We also develop temporal features by time series analysis from query frequency. Other generated features include word matching with recurrent event seed words and time sensitivity of search result set. We use Naive Bayes, SVM and decision tree based logistic regres- sion model to train REQ classifier. The results on test data show that our models outperformed baseline approach significantly. Experiments on a commercial Web search engine also show significant gains in overall relevance, and thus overall user experience.
4 0.078442805 51 emnlp-2010-Function-Based Question Classification for General QA
Author: Fan Bu ; Xingwei Zhu ; Yu Hao ; Xiaoyan Zhu
Abstract: In contrast with the booming increase of internet data, state-of-art QA (question answering) systems, otherwise, concerned data from specific domains or resources such as search engine snippets, online forums and Wikipedia in a somewhat isolated way. Users may welcome a more general QA system for its capability to answer questions of various sources, integrated from existed specialized sub-QA engines. In this framework, question classification is the primary task. However, the current paradigms of question classification were focused on some specified type of questions, i.e. factoid questions, which are inappropriate for the general QA. In this paper, we propose a new question classification paradigm, which includes a question taxonomy suitable to the general QA and a question classifier based on MLN (Markov logic network), where rule-based methods and statistical methods are unified into a single framework in a fuzzy discriminative learning approach. Experiments show that our method outperforms traditional question classification approaches.
5 0.07820297 32 emnlp-2010-Context Comparison of Bursty Events in Web Search and Online Media
Author: Yunliang Jiang ; Cindy Xide Lin ; Qiaozhu Mei
Abstract: In this paper, we conducted a systematic comparative analysis of language in different contexts of bursty topics, including web search, news media, blogging, and social bookmarking. We analyze (1) the content similarity and predictability between contexts, (2) the coverage of search content by each context, and (3) the intrinsic coherence of information in each context. Our experiments show that social bookmarking is a better predictor to the bursty search queries, but news media and social blogging media have a much more compelling coverage. This comparison provides insights on how the search behaviors and social information sharing behaviors of users are correlated to the professional news media in the context of bursty events.
6 0.077708215 62 emnlp-2010-Improving Mention Detection Robustness to Noisy Input
7 0.07552807 33 emnlp-2010-Cross Language Text Classification by Model Translation and Semi-Supervised Learning
8 0.068026766 56 emnlp-2010-Hashing-Based Approaches to Spelling Correction of Personal Names
9 0.063418292 79 emnlp-2010-Mining Name Translations from Entity Graph Mapping
11 0.05907663 66 emnlp-2010-Inducing Word Senses to Improve Web Search Result Clustering
12 0.052642655 47 emnlp-2010-Example-Based Paraphrasing for Improved Phrase-Based Statistical Machine Translation
13 0.052528847 63 emnlp-2010-Improving Translation via Targeted Paraphrasing
14 0.047001004 78 emnlp-2010-Minimum Error Rate Training by Sampling the Translation Lattice
15 0.046566691 57 emnlp-2010-Hierarchical Phrase-Based Translation Grammars Extracted from Alignment Posterior Probabilities
16 0.046426456 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
17 0.043641184 90 emnlp-2010-Positional Language Models for Clinical Information Retrieval
18 0.041684344 50 emnlp-2010-Facilitating Translation Using Source Language Paraphrase Lattices
19 0.039340284 58 emnlp-2010-Holistic Sentiment Analysis Across Languages: Multilingual Supervised Latent Dirichlet Allocation
20 0.038646083 5 emnlp-2010-A Hybrid Morpheme-Word Representation for Machine Translation of Morphologically Rich Languages
topicId topicWeight
[(0, 0.155), (1, -0.002), (2, -0.09), (3, 0.093), (4, 0.023), (5, 0.031), (6, -0.13), (7, 0.034), (8, -0.16), (9, 0.037), (10, -0.038), (11, 0.113), (12, 0.155), (13, 0.066), (14, -0.021), (15, 0.12), (16, -0.101), (17, -0.022), (18, 0.111), (19, -0.023), (20, -0.092), (21, 0.049), (22, -0.039), (23, 0.048), (24, 0.033), (25, -0.022), (26, 0.001), (27, -0.115), (28, -0.051), (29, -0.004), (30, 0.112), (31, 0.028), (32, -0.078), (33, 0.026), (34, -0.057), (35, -0.004), (36, -0.108), (37, -0.045), (38, 0.058), (39, 0.082), (40, -0.065), (41, -0.009), (42, -0.088), (43, 0.069), (44, -0.023), (45, -0.129), (46, -0.097), (47, -0.083), (48, -0.05), (49, 0.062)]
simIndex simValue paperId paperTitle
same-paper 1 0.94451261 55 emnlp-2010-Handling Noisy Queries in Cross Language FAQ Retrieval
Author: Danish Contractor ; Govind Kothari ; Tanveer Faruquie ; L V Subramaniam ; Sumit Negi
Abstract: Recent times have seen a tremendous growth in mobile based data services that allow people to use Short Message Service (SMS) to access these data services. In a multilingual society it is essential that data services that were developed for a specific language be made accessible through other local languages also. In this paper, we present a service that allows a user to query a FrequentlyAsked-Questions (FAQ) database built in a local language (Hindi) using Noisy SMS English queries. The inherent noise in the SMS queries, along with the language mismatch makes this a challenging problem. We handle these two problems by formulating the query similarity over FAQ questions as a combinatorial search problem where the search space consists of combinations of dictionary variations of the noisy query and its top-N translations. We demonstrate the effectiveness of our approach on a real-life dataset.
2 0.51524442 74 emnlp-2010-Learning the Relative Usefulness of Questions in Community QA
Author: Razvan Bunescu ; Yunfeng Huang
Abstract: We present a machine learning approach for the task of ranking previously answered questions in a question repository with respect to their relevance to a new, unanswered reference question. The ranking model is trained on a collection of question groups manually annotated with a partial order relation reflecting the relative utility of questions inside each group. Based on a set of meaning and structure aware features, the new ranking model is able to substantially outperform more straightforward, unsupervised similarity measures.
3 0.47684959 51 emnlp-2010-Function-Based Question Classification for General QA
Author: Fan Bu ; Xingwei Zhu ; Yu Hao ; Xiaoyan Zhu
Abstract: In contrast with the booming increase of internet data, state-of-art QA (question answering) systems, otherwise, concerned data from specific domains or resources such as search engine snippets, online forums and Wikipedia in a somewhat isolated way. Users may welcome a more general QA system for its capability to answer questions of various sources, integrated from existed specialized sub-QA engines. In this framework, question classification is the primary task. However, the current paradigms of question classification were focused on some specified type of questions, i.e. factoid questions, which are inappropriate for the general QA. In this paper, we propose a new question classification paradigm, which includes a question taxonomy suitable to the general QA and a question classifier based on MLN (Markov logic network), where rule-based methods and statistical methods are unified into a single framework in a fuzzy discriminative learning approach. Experiments show that our method outperforms traditional question classification approaches.
4 0.45013961 73 emnlp-2010-Learning Recurrent Event Queries for Web Search
Author: Ruiqiang Zhang ; Yuki Konda ; Anlei Dong ; Pranam Kolari ; Yi Chang ; Zhaohui Zheng
Abstract: Recurrent event queries (REQ) constitute a special class of search queries occurring at regular, predictable time intervals. The freshness of documents ranked for such queries is generally of critical importance. REQ forms a significant volume, as much as 6% of query traffic received by search engines. In this work, we develop an improved REQ classifier that could provide significant improvements in addressing this problem. We analyze REQ queries, and develop novel features from multiple sources, and evaluate them using machine learning techniques. From historical query logs, we develop features utilizing query frequency, click information, and user intent dynamics within a search session. We also develop temporal features by time series analysis from query frequency. Other generated features include word matching with recurrent event seed words and time sensitivity of search result set. We use Naive Bayes, SVM and decision tree based logistic regres- sion model to train REQ classifier. The results on test data show that our models outperformed baseline approach significantly. Experiments on a commercial Web search engine also show significant gains in overall relevance, and thus overall user experience.
Author: Aida Khemakhem ; Bilel Gargouri ; Abdelmajid Ben Hamadou
Abstract: Electronic dictionaries covering all natural language levels are very relevant for the human use as well as for the automatic processing use, namely those constructed with respect to international standards. Such dictionaries are characterized by a complex structure and an important access time when using a querying system. However, the need of a user is generally limited to a part of such a dictionary according to his domain and expertise level which corresponds to a specialized dictionary. Given the importance of managing a unified dictionary and considering the personalized needs of users, we propose an approach for generating personalized views starting from a normalized dictionary with respect to Lexical Markup Framework LMF-ISO 24613 norm. This approach provides the re-use of already defined views for a community of users by managing their profiles information and promoting the materialization of the generated views. It is composed of four main steps: (i) the projection of data categories controlled by a set of constraints (related to the user‟s profiles), (ii) the selection of values with consistency checking, (iii) the automatic generation of the query‟s model and finally, (iv) the refinement of the view. The proposed approach was con- solidated by carrying out an experiment on an LMF normalized Arabic dictionary. 1
6 0.39933312 56 emnlp-2010-Hashing-Based Approaches to Spelling Correction of Personal Names
7 0.39891911 90 emnlp-2010-Positional Language Models for Clinical Information Retrieval
8 0.38909149 79 emnlp-2010-Mining Name Translations from Entity Graph Mapping
9 0.37705317 32 emnlp-2010-Context Comparison of Bursty Events in Web Search and Online Media
10 0.3413398 109 emnlp-2010-Translingual Document Representations from Discriminative Projections
11 0.33585384 66 emnlp-2010-Inducing Word Senses to Improve Web Search Result Clustering
12 0.31015453 62 emnlp-2010-Improving Mention Detection Robustness to Noisy Input
13 0.27463797 108 emnlp-2010-Training Continuous Space Language Models: Some Practical Issues
14 0.27108291 29 emnlp-2010-Combining Unsupervised and Supervised Alignments for MT: An Empirical Study
15 0.26570955 33 emnlp-2010-Cross Language Text Classification by Model Translation and Semi-Supervised Learning
16 0.24720903 44 emnlp-2010-Enhancing Mention Detection Using Projection via Aligned Corpora
17 0.23967296 84 emnlp-2010-NLP on Spoken Documents Without ASR
18 0.23635475 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
19 0.22912346 115 emnlp-2010-Uptraining for Accurate Deterministic Question Parsing
20 0.22458588 45 emnlp-2010-Evaluating Models of Latent Document Semantics in the Presence of OCR Errors
topicId topicWeight
[(3, 0.032), (10, 0.036), (12, 0.045), (29, 0.065), (30, 0.023), (32, 0.019), (52, 0.016), (56, 0.043), (62, 0.031), (66, 0.097), (72, 0.062), (76, 0.047), (79, 0.014), (82, 0.364), (89, 0.02), (92, 0.012)]
simIndex simValue paperId paperTitle
1 0.80139589 21 emnlp-2010-Automatic Discovery of Manner Relations and its Applications
Author: Eduardo Blanco ; Dan Moldovan
Abstract: This paper presents a method for the automatic discovery of MANNER relations from text. An extended definition of MANNER is proposed, including restrictions on the sorts of concepts that can be part of its domain and range. The connections with other relations and the lexico-syntactic patterns that encode MANNER are analyzed. A new feature set specialized on MANNER detection is depicted and justified. Experimental results show improvement over previous attempts to extract MANNER. Combinations of MANNER with other semantic relations are also discussed.
same-paper 2 0.73128289 55 emnlp-2010-Handling Noisy Queries in Cross Language FAQ Retrieval
Author: Danish Contractor ; Govind Kothari ; Tanveer Faruquie ; L V Subramaniam ; Sumit Negi
Abstract: Recent times have seen a tremendous growth in mobile based data services that allow people to use Short Message Service (SMS) to access these data services. In a multilingual society it is essential that data services that were developed for a specific language be made accessible through other local languages also. In this paper, we present a service that allows a user to query a FrequentlyAsked-Questions (FAQ) database built in a local language (Hindi) using Noisy SMS English queries. The inherent noise in the SMS queries, along with the language mismatch makes this a challenging problem. We handle these two problems by formulating the query similarity over FAQ questions as a combinatorial search problem where the search space consists of combinations of dictionary variations of the noisy query and its top-N translations. We demonstrate the effectiveness of our approach on a real-life dataset.
3 0.70759451 27 emnlp-2010-Clustering-Based Stratified Seed Sampling for Semi-Supervised Relation Classification
Author: Longhua Qian ; Guodong Zhou
Abstract: Seed sampling is critical in semi-supervised learning. This paper proposes a clusteringbased stratified seed sampling approach to semi-supervised learning. First, various clustering algorithms are explored to partition the unlabeled instances into different strata with each stratum represented by a center. Then, diversity-motivated intra-stratum sampling is adopted to choose the center and additional instances from each stratum to form the unlabeled seed set for an oracle to annotate. Finally, the labeled seed set is fed into a bootstrapping procedure as the initial labeled data. We systematically evaluate our stratified bootstrapping approach in the semantic relation classification subtask of the ACE RDC (Relation Detection and Classification) task. In particular, we compare various clustering algorithms on the stratified bootstrapping performance. Experimental results on the ACE RDC 2004 corpus show that our clusteringbased stratified bootstrapping approach achieves the best F1-score of 75.9 on the subtask of semantic relation classification, approaching the one with golden clustering.
Author: Hugo Hernault ; Danushka Bollegala ; Mitsuru Ishizuka
Abstract: Several recent discourse parsers have employed fully-supervised machine learning approaches. These methods require human annotators to beforehand create an extensive training corpus, which is a time-consuming and costly process. On the other hand, unlabeled data is abundant and cheap to collect. In this paper, we propose a novel semi-supervised method for discourse relation classification based on the analysis of cooccurring features in unlabeled data, which is then taken into account for extending the feature vectors given to a classifier. Our experimental results on the RST Discourse Treebank corpus and Penn Discourse Treebank indicate that the proposed method brings a significant improvement in classification accuracy and macro-average F-score when small training datasets are used. For instance, with training sets of c.a. 1000 labeled instances, the proposed method brings improvements in accuracy and macro-average F-score up to 50% compared to a baseline classifier. We believe that the proposed method is a first step towards detecting low-occurrence relations, which is useful for domains with a lack of annotated data.
5 0.42261732 31 emnlp-2010-Constraints Based Taxonomic Relation Classification
Author: Quang Do ; Dan Roth
Abstract: Determining whether two terms in text have an ancestor relation (e.g. Toyota and car) or a sibling relation (e.g. Toyota and Honda) is an essential component of textual inference in NLP applications such as Question Answering, Summarization, and Recognizing Textual Entailment. Significant work has been done on developing stationary knowledge sources that could potentially support these tasks, but these resources often suffer from low coverage, noise, and are inflexible when needed to support terms that are not identical to those placed in them, making their use as general purpose background knowledge resources difficult. In this paper, rather than building a stationary hierarchical structure of terms and relations, we describe a system that, given two terms, determines the taxonomic relation between them using a machine learning-based approach that makes use of existing resources. Moreover, we develop a global constraint opti- mization inference process and use it to leverage an existing knowledge base also to enforce relational constraints among terms and thus improve the classifier predictions. Our experimental evaluation shows that our approach significantly outperforms other systems built upon existing well-known knowledge sources.
6 0.40223745 20 emnlp-2010-Automatic Detection and Classification of Social Events
7 0.39430016 72 emnlp-2010-Learning First-Order Horn Clauses from Web Text
8 0.39298776 35 emnlp-2010-Discriminative Sample Selection for Statistical Machine Translation
9 0.38634166 45 emnlp-2010-Evaluating Models of Latent Document Semantics in the Presence of OCR Errors
10 0.37219727 46 emnlp-2010-Evaluating the Impact of Alternative Dependency Graph Encodings on Solving Event Extraction Tasks
11 0.37181213 32 emnlp-2010-Context Comparison of Bursty Events in Web Search and Online Media
12 0.3658525 26 emnlp-2010-Classifying Dialogue Acts in One-on-One Live Chats
13 0.36545616 68 emnlp-2010-Joint Inference for Bilingual Semantic Role Labeling
14 0.36433512 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
15 0.36362299 51 emnlp-2010-Function-Based Question Classification for General QA
16 0.36320409 67 emnlp-2010-It Depends on the Translation: Unsupervised Dependency Parsing via Word Alignment
17 0.36313024 69 emnlp-2010-Joint Training and Decoding Using Virtual Nodes for Cascaded Segmentation and Tagging Tasks
18 0.36110353 92 emnlp-2010-Predicting the Semantic Compositionality of Prefix Verbs
19 0.36103669 78 emnlp-2010-Minimum Error Rate Training by Sampling the Translation Lattice
20 0.36033022 84 emnlp-2010-NLP on Spoken Documents Without ASR