emnlp emnlp2012 emnlp2012-110 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Ni Lao ; Amarnag Subramanya ; Fernando Pereira ; William W. Cohen
Abstract: We study how to extend a large knowledge base (Freebase) by reading relational information from a large Web text corpus. Previous studies on extracting relational knowledge from text show the potential of syntactic patterns for extraction, but they do not exploit background knowledge of other relations in the knowledge base. We describe a distributed, Web-scale implementation of a path-constrained random walk model that learns syntactic-semantic inference rules for binary relations from a graph representation of the parsed text and the knowledge base. Experiments show significant accuracy improvements in binary relation prediction over methods that consider only text, or only the existing knowledge base.
Reference: text
sentIndex sentText sentNum sentScore
1 Previous studies on extracting relational knowledge from text show the potential of syntactic patterns for extraction, but they do not exploit background knowledge of other relations in the knowledge base. [sent-8, score-0.355]
2 We describe a distributed, Web-scale implementation of a path-constrained random walk model that learns syntactic-semantic inference rules for binary relations from a graph representation of the parsed text and the knowledge base. [sent-9, score-0.39]
3 To fill those KB gaps, we might use general rules, ideally automatically learned, such as “if person was born in town and town is in country ∗This research was carried out during an internship at Google Research 1017 then the person is a national of the country. [sent-14, score-0.308]
4 Alternatively, we may attempt to fill KB gaps by applying relation extraction rules to free text. [sent-18, score-0.22]
5 However, the KB contains much more knowledge about other relations that could potentially be helpful in improving relation extraction accuracy and coverage, but that is not used in such purely text-based approaches. [sent-23, score-0.258]
6 In this work, we use PRA to learn weighted rules (represented as graph path patterns) that combine both semantic (KB) and syntactic information encoded respectively as edges in a graphstructured KB, and as syntactic dependency edges in dependency-parsed Web text. [sent-24, score-0.493]
7 Our approach can easily incorporate existing knowledge in extraction tasks, and its distributed implementation scales to the whole of the Freebase KB and 60 million parsed documents. [sent-25, score-0.209]
8 missing relation instances by combining the KB with parsed text. [sent-38, score-0.196]
9 AnA Ain psatathn ctyep oef i tnhe G path type ins a sequence of nodeis. [sent-46, score-0.174]
10 For instance, “the persons who were born in the same town as the query person”, and “the nationalities of persons who were born in the same town as the query person” can be reached respectively through paths matching the following types π1 π2 1. [sent-51, score-0.901]
11 with Path-Constrained Random Walks Given a query concept s ∈ C and a relation r ∈ R, PRA begins by enumerating a large set of bounded-length path types. [sent-59, score-0.512]
12 Terhaetsine path types are treated as ranking “experts,” each generating some random instance of the path type starting from s, and ranking end nodes t by their weights in the resulting distribution. [sent-60, score-0.42]
13 Finally, PRA combines the weights contributed by different “experts” by using logistic regression to predict the probability that the relation r(s, t) holds. [sent-61, score-0.179]
14 As shown in Figure 1, we extend the KB graph G with nodes and edges from text that has been syntactically analyzed with a dependency parser1 and where pronouns and other anaphoric referring expressions have been clustered with their antecedents. [sent-65, score-0.244]
15 The text nodes are word/phrase instances, and the edges are syntactic dependencies labeled by the corresponding dependency type. [sent-66, score-0.228]
16 Mentions of entities in the text are linked to KB concepts by mention edges created by an entity resolution process. [sent-67, score-0.313]
17 ), PRA produces a ranked list of answers that may have the relation Profession with the query node CharlotteBronte. [sent-69, score-0.345]
18 The features used to score answers are the random walk probabilities of reaching a certain profession node from the query node by paths with particular path types. [sent-70, score-0.948]
19 PRA can learn path types that combine background knowledge in the database with syntactic patterns in the text corpus. [sent-71, score-0.35]
20 query person as in “collaboration between McDougall and Simon 1Stanford dependencies (de Marneffe and Manning, 2008). [sent-81, score-0.225]
21 The features generated from these path types combine syntactic dependency relations (conj) and textual information relations (TW and CW) with semantic relations in the KB (Profession). [sent-97, score-0.395]
22 Experiments on three Freebase relations (profession, nationality and parents) show that exploiting existing background knowledge as path features can significantly improve the quality of extraction compared with using either Freebase or the text corpus alone. [sent-98, score-0.486]
23 Later systems for extracting arbitrary relations from text mostly use shallow surface text patterns (Etzioni et al. [sent-109, score-0.192]
24 The idea of using sequences of dependency edges as features for relation extraction was explored by Snow et al. [sent-111, score-0.282]
25 As in our work, they use a logistic regression model with path features. [sent-117, score-0.223]
26 Furthermore, their path patterns are used as binary-values features. [sent-119, score-0.224]
27 (2006)’s work is similar to our approach in the sense of relation extraction by discovering relational patterns. [sent-122, score-0.229]
28 However while they focus on identifying relation mentions in text (microreading),this work attempts to infer new tuples by gathering path evidence over the whole corpus (macroreading). [sent-123, score-0.396]
29 More specifically, suppose that the random walk has just reached vi by traversing edges labeled r1, . [sent-134, score-0.233]
30 Then vi+1 is drawn at random from all nodes reachable from vi by edges labeled ri+1. [sent-138, score-0.213]
31 A path type π is active for pair (s, t) if P(s → t; π) > 0. [sent-139, score-0.213]
32 , πn} be the set of all path types oBf length no greater tbhea nth e‘ tehta to occur tinh the graph together with the dummy type ⊥, which represents the bias feature. [sent-143, score-0.216]
33 θ θ Path Discovery: Given a graph and a target relation r, the total number of path types is an exponential function of the maximum path length ‘ and considering all possible paths would be computationally very expensive. [sent-150, score-0.585]
34 As a result, B is constructed using only path types that satisfy the following two constraints: 1. [sent-151, score-0.174]
35 the path type is active for more than K training query nodes, and 2. [sent-152, score-0.353]
36 the probability of reaching any correct target node t is larger than a threshold α on average for the training query nodes s. [sent-153, score-0.256]
37 In addition to making the training more efficient, these constraints are also helpful in removing low quality path types. [sent-155, score-0.174]
38 From Sr, we create the training set Dr = {(xi, yi) }, where xi = hP(si → ti; π)iπ∈B i=s t{h(ex vect)o},r of path fea=ture h Pva(lsues →for t;heπ pair (si, ti), and yi indicates whether r(si, ti) holds. [sent-157, score-0.174]
39 , for the parents relation, any pair of person concepts which are related by this relation would be valid negative examples) which not only makes training very expensive but also introduces an incorrect bias in the training set. [sent-166, score-0.464]
40 Following Lao and Cohen (2010) we use a simple biased sampling procedure to generate negative examples: first, the path types discovered in the previous (path discovery) step are used to construct an initial PRA model (all feature weights are set to 1. [sent-167, score-0.195]
41 0); then, for each query node si, this model is used to retrieve candidate answer nodes, which are then sorted in descending order by their scores; finally, nodes at the k(k + 1)/2-th positions are selected as negative samples, where k = 0, 1, 2, . [sent-168, score-0.227]
42 For each unlabeled query node s, we apply the trained PRA model to generate a list of candidate t nodes together with their scores. [sent-177, score-0.227]
43 As Freebase contains millions of concepts and edges, training on all the generated queries is computationally challenging. [sent-180, score-0.276]
44 Further, we extend the Freebase graph with parse paths of mentions of concepts in Freebase in millions of Web pages. [sent-181, score-0.335]
45 Yet another issue is that the training queries generated using Freebase are inherently biased towards the distribution of concepts in Freebase and may not reflect the distribution of mentions of these concepts in text data. [sent-182, score-0.502]
46 As one of the goals of our approach is to learn relation instances that are missing in Freebase, training on such a set biased towards the distribution of concepts in Freebase may not lead to good performance. [sent-183, score-0.297]
47 For example, for the profession relation, there are around 2 million persons in Freebase, and about 0. [sent-187, score-0.533]
48 3 million training queries (persons), each with one or more positive answers (professions), and many negative answers, which make training computationally challenging. [sent-190, score-0.178]
49 Generating all the paths for millions of queries over a graph with millions of concepts and edges further complicates the computational issues. [sent-191, score-0.501]
50 Incorporating the parse path features from the text only exacerbates the matter. [sent-192, score-0.217]
51 Finally once we have trained a PRA model for a given relation, say profession, we would like to infer the professions for all the 1. [sent-193, score-0.222]
52 7 million persons whose professions are not known to Freebase (and possibly predict changes to the profession information of the 0. [sent-194, score-0.755]
53 To circumvent this problem, we first index all parsed sentences by the concepts that they mention. [sent-203, score-0.212]
54 Therefore, to perform a random walk for a query concept s, we only load the sentences which mention s. [sent-204, score-0.37]
55 For example, for the profession relation, there are 0. [sent-207, score-0.302]
56 3 million persons for whom Freebase has profession information, and amongst these 0. [sent-208, score-0.533]
57 This may not reflect the distribution of professions of persons mentioned in Web data. [sent-210, score-0.455]
58 Using all of these as training queries will most certainly bias the trained model towards these professions as PRA is trained discriminatively. [sent-211, score-0.319]
59 In other words, training directly with this data would lead to a model that is more likely to predict professions that are popular in Freebase. [sent-212, score-0.222]
60 For each relation r and concept t ∈ C, we count the number roefl r edges pointing etpot t t Nr,t = |{(s, r, t) ∈ T}| Given a training according to query . [sent-214, score-0.423]
61 If we take the profession relation as an example, the above implies that for popular professions, we only sample about pNr,t out of the Nr,t possible queries that end in t, wphereas for the less popular professions we would accept all the training queries. [sent-217, score-0.751]
62 3 Text Graph Construction As we are processing Web text data (see following section for more detail), the number of mentions of a concept follows a somewhat heavy-tailed distribution: there are a small number of very popular concepts (head) and a large number of not so popular concepts (tail). [sent-219, score-0.452]
63 To prevent the text graph from being dominated by the head concepts, for each sentence that mentions concept c ∈ C, we accept site as part o thf athte m teexntti graph wnicethp probability: Pc= min? [sent-222, score-0.244]
64 3 As of this writing, it contains more than 21 million concepts and 70 million labeled edges. [sent-234, score-0.238]
65 We also collect a large Web corpus and identify 60 million pages that mention concepts relevant to this study. [sent-236, score-0.231]
66 In each of the parsed documents, we use POS tags and dependency edges to identify potential referring noun phrases (NPs). [sent-239, score-0.178]
67 For each Freebase concept c ∈ C, we compute N(c, m), tehaec hn Fumreebbera oef c otinmceesp t ch e∈ concept c pisu rteef Nerr(ce,dm by mention m by using both the alias information in Freebase and the anchors of the corresponding Wikipedia page for that concept. [sent-251, score-0.213]
68 Iafl u bisa a ctylu pst(ecr| w)it =h mention sePt M(u) in the document, and C(m) the set of concepts in KB with name or alias m, we assign u to concept c∗ = argmax p(c|m), c∈C(m),m∈M(u) provided that there exists at least one c ∈ C(m) and m ∈ M(u) stu thcher teh eaxt p(c|m) > o0n. [sent-253, score-0.291]
69 5 Results We use three relations profession, nationality and parents for our experiments. [sent-256, score-0.275]
70 However, the parents relation yields 350k triples after stratified sampling, so to reduce experimental effort we further randomly sub-sample 10% of that as input to the train-test split. [sent-260, score-0.369]
71 To encourage PRA to find paths involving the text corpus, we do not count relation M (which connects concepts to their mentions) or M−1 when calculating path lengths. [sent-262, score-0.558]
72 1 Evaluation with Existing Knowledge Previous work in relation extraction from parsed text (Mintz et al. [sent-287, score-0.279]
73 Table 2 shows a comparison of the results obtained using the PRA algorithm trained using only Freebase (KB), using only the text corpus graph (Text), trained with both Freebase and the text corpus (KB+Text) and the binarized PRA algorithm using both Freebase and the text corpus (KB+Text[b]). [sent-290, score-0.21]
74 We report Mean Reciprocal Rank (MRR) where, given a set of queries Q, MRR =|Q1|qX∈Qrank of q’s first1 correct answer Comparing the results of first three columns we see that combining Freebase and text achieves significantly better results than using either Freebase or text alone. [sent-291, score-0.183]
75 It can also be seen that the MRR for the parents relation is lower than those for other relations. [sent-294, score-0.233]
76 This is mainly because there are larger number of potential answers for each query node of Parent relation than for each query node of the other two relations all persons in Freebase versus all professions or nationalities. [sent-295, score-0.988]
77 Table 2 only reports results for the maximum path length ‘ = 4 case. [sent-298, score-0.174]
78 We found that shorter maximum path lengths give worse results: for instance, with ‘ = 3 for the profession relation, MRR drops to 0. [sent-299, score-0.476]
79 Further we find that using longer path length takes much longer time to train and test, but does not lead to significant improvements over – — the ‘ = 4 case. [sent-304, score-0.174]
80 For the profession and nationality tasks, the conjunction dependency relation (in group 1,4) plays an important role: these features first find persons mentioned in conjunction with the query Table 3: Top weighted path types involving text edges for each task grouped according to functionality. [sent-309, score-1.348]
81 M relations connect each concept in knowledge base to its mentions in the corpus. [sent-310, score-0.235]
82 CW relations connect each concept in knowledge base to the words in the text representation of this concept. [sent-312, score-0.229]
83 pFmrioesnfte fiso n idoednpseirnosf sitnmhseiwlrai cth wisaldiymrse,nilt/ahprean rea nmgt ser/saedogvratiesotrhse: starting from the concept BarackObama, words such as “Obama ”, “leader”, “president”, and “he ” are reachable through path hM, TWi Nationality Top Weighted Features Comments 4? [sent-330, score-0.273]
84 conjunction with the query person: “McDougall and Simon Phillips collaborated . [sent-336, score-0.209]
85 The parents of persons with similar names or mentioned in similar ways: starting from the concept CharlotteBronte words such as “Bronte ”, “Charlotte ”, “Patrick’’, and “she ” are reachable through path hM, TWi. [sent-361, score-0.637]
86 1024 person, and then find their professions or nationalities. [sent-387, score-0.222]
87 Features in group 3 first find persons with similar names or mentioned in similar ways to the query person, and then aggregate the professions of their children, parents, or advisors. [sent-396, score-0.647]
88 f the mentions of the query node have poss (stands for possessive modifier, e. [sent-411, score-0.229]
89 Features in group 8 generally find persons with similar names or mentioned in similar ways to the query person. [sent-417, score-0.425]
90 The annotator has access to the Freebase and Wikipedia pages for the concepts (and is able to issue search queries about the concepts). [sent-423, score-0.243]
91 In the first evaluation, we compared the performance of two PRA models, one trained using the stratified sampled queries and another trained using a randomly sampled set of queries for the profession relation. [sent-424, score-0.634]
92 We found that the PRA model trained with stratified sampled queries has 0. [sent-426, score-0.21]
93 We also evaluated the new beliefs proposed by the models trained for all the three relations using stratified sampled queries. [sent-441, score-0.2]
94 For instance, for the profession relation, we are able to predict professions for the around 2 million persons in Freebase. [sent-447, score-0.755]
95 The top 1000 profession facts extracted by our system involve 970 distinct people, the top 10,000 facts involve 8,726 distinct people, and the top 100,000 facts involve 79,885 people. [sent-448, score-0.404]
96 6 Conclusion We have shown that path constrained random walk models can effectively infer new beliefs from a large scale parsed text corpus with background knowledge. [sent-449, score-0.462]
97 Evaluation by human annotators shows that by combining syntactic patterns in parsed text with semantic patterns in the background knowledge, our model can propose new beliefs with high accuracy. [sent-450, score-0.291]
98 First, bidirectional search from both query and target nodes can be an efficient way to discover long paths. [sent-453, score-0.218]
99 Second, relation paths that contain constant nodes (lexicalized features) and conjunction of random walk features are potentially very useful for extraction tasks. [sent-455, score-0.442]
100 Integrating probabilistic extraction models and data mining to discover relations and patterns in text. [sent-470, score-0.177]
wordName wordTfidf (topN-words)
[('pra', 0.465), ('kb', 0.372), ('freebase', 0.316), ('profession', 0.302), ('professions', 0.222), ('persons', 0.185), ('path', 0.174), ('lao', 0.153), ('concepts', 0.146), ('query', 0.14), ('relation', 0.13), ('nationality', 0.116), ('cw', 0.105), ('parents', 0.103), ('walk', 0.098), ('queries', 0.097), ('stratified', 0.088), ('person', 0.085), ('edges', 0.085), ('mrr', 0.076), ('tw', 0.069), ('concept', 0.068), ('parsed', 0.066), ('paths', 0.065), ('relational', 0.059), ('conj', 0.056), ('relations', 0.056), ('patterns', 0.05), ('mentions', 0.049), ('mcdougall', 0.049), ('mentioned', 0.048), ('triples', 0.048), ('nodes', 0.047), ('nsubj', 0.047), ('million', 0.046), ('walks', 0.044), ('town', 0.044), ('text', 0.043), ('title', 0.043), ('graph', 0.042), ('nationalities', 0.042), ('triple', 0.042), ('extraction', 0.04), ('node', 0.04), ('mention', 0.039), ('binarized', 0.039), ('active', 0.039), ('alias', 0.038), ('conjunction', 0.037), ('mintz', 0.036), ('answers', 0.035), ('facts', 0.034), ('sc', 0.033), ('millions', 0.033), ('snow', 0.033), ('knowledge', 0.032), ('barackobama', 0.032), ('charlottebronte', 0.032), ('collaborated', 0.032), ('mapreduce', 0.032), ('subpath', 0.032), ('discover', 0.031), ('reachable', 0.031), ('beliefs', 0.031), ('cohen', 0.031), ('connect', 0.03), ('president', 0.03), ('reaching', 0.029), ('suchanek', 0.029), ('wikipedia', 0.029), ('born', 0.028), ('phillips', 0.028), ('names', 0.028), ('rules', 0.028), ('np', 0.027), ('simon', 0.027), ('dependency', 0.027), ('web', 0.026), ('syntactic', 0.026), ('obama', 0.026), ('taxonomic', 0.025), ('background', 0.025), ('vi', 0.025), ('distributed', 0.025), ('logistic', 0.025), ('sampled', 0.025), ('random', 0.025), ('regression', 0.024), ('group', 0.024), ('friedman', 0.023), ('reciprocal', 0.023), ('fractional', 0.023), ('bornin', 0.023), ('si', 0.023), ('agichtein', 0.022), ('dean', 0.022), ('bollacker', 0.022), ('fill', 0.022), ('predictions', 0.022), ('biased', 0.021)]
simIndex simValue paperId paperTitle
same-paper 1 1.0000001 110 emnlp-2012-Reading The Web with Learned Syntactic-Semantic Inference Rules
Author: Ni Lao ; Amarnag Subramanya ; Fernando Pereira ; William W. Cohen
Abstract: We study how to extend a large knowledge base (Freebase) by reading relational information from a large Web text corpus. Previous studies on extracting relational knowledge from text show the potential of syntactic patterns for extraction, but they do not exploit background knowledge of other relations in the knowledge base. We describe a distributed, Web-scale implementation of a path-constrained random walk model that learns syntactic-semantic inference rules for binary relations from a graph representation of the parsed text and the knowledge base. Experiments show significant accuracy improvements in binary relation prediction over methods that consider only text, or only the existing knowledge base.
2 0.14779817 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers
Author: Jayant Krishnamurthy ; Tom Mitchell
Abstract: We present a method for training a semantic parser using only a knowledge base and an unlabeled text corpus, without any individually annotated sentences. Our key observation is that multiple forms ofweak supervision can be combined to train an accurate semantic parser: semantic supervision from a knowledge base, and syntactic supervision from dependencyparsed sentences. We apply our approach to train a semantic parser that uses 77 relations from Freebase in its knowledge representation. This semantic parser extracts instances of binary relations with state-of-theart accuracy, while simultaneously recovering much richer semantic structures, such as conjunctions of multiple relations with partially shared arguments. We demonstrate recovery of this richer structure by extracting logical forms from natural language queries against Freebase. On this task, the trained semantic parser achieves 80% precision and 56% recall, despite never having seen an annotated logical form.
3 0.13472712 97 emnlp-2012-Natural Language Questions for the Web of Data
Author: Mohamed Yahya ; Klaus Berberich ; Shady Elbassuoni ; Maya Ramanath ; Volker Tresp ; Gerhard Weikum
Abstract: The Linked Data initiative comprises structured databases in the Semantic-Web data model RDF. Exploring this heterogeneous data by structured query languages is tedious and error-prone even for skilled users. To ease the task, this paper presents a methodology for translating natural language questions into structured SPARQL queries over linked-data sources. Our method is based on an integer linear program to solve several disambiguation tasks jointly: the segmentation of questions into phrases; the mapping of phrases to semantic entities, classes, and relations; and the construction of SPARQL triple patterns. Our solution harnesses the rich type system provided by knowledge bases in the web of linked data, to constrain our semantic-coherence objective function. We present experiments on both the . in question translation and the resulting query answering.
4 0.12199519 47 emnlp-2012-Explore Person Specific Evidence in Web Person Name Disambiguation
Author: Liwei Chen ; Yansong Feng ; Lei Zou ; Dongyan Zhao
Abstract: In this paper, we investigate different usages of feature representations in the web person name disambiguation task which has been suffering from the mismatch of vocabulary and lack of clues in web environments. In literature, the latter receives less attention and remains more challenging. We explore the feature space in this task and argue that collecting person specific evidences from a corpus level can provide a more reasonable and robust estimation for evaluating a feature’s importance in a given web page. This can alleviate the lack of clues where discriminative features can be reasonably weighted by taking their corpus level importance into account, not just relying on the current local context. We therefore propose a topic-based model to exploit the person specific global importance and embed it into the person name similarity. The experimental results show that the corpus level topic in- formation provides more stable evidences for discriminative features and our method outperforms the state-of-the-art systems on three WePS datasets.
5 0.11112683 93 emnlp-2012-Multi-instance Multi-label Learning for Relation Extraction
Author: Mihai Surdeanu ; Julie Tibshirani ; Ramesh Nallapati ; Christopher D. Manning
Abstract: Distant supervision for relation extraction (RE) gathering training data by aligning a database of facts with text – is an efficient approach to scale RE to thousands of different relations. However, this introduces a challenging learning scenario where the relation expressed by a pair of entities found in a sentence is unknown. For example, a sentence containing Balzac and France may express BornIn or Died, an unknown relation, or no relation at all. Because of this, traditional supervised learning, which assumes that each example is explicitly mapped to a label, is not appropriate. We propose a novel approach to multi-instance multi-label learning for RE, which jointly models all the instances of a pair of entities in text and all their labels using a graphical model with latent variables. Our model performs competitively on two difficult domains. –
6 0.10514478 98 emnlp-2012-No Noun Phrase Left Behind: Detecting and Typing Unlinkable Entities
7 0.10172338 103 emnlp-2012-PATTY: A Taxonomy of Relational Patterns with Semantic Types
8 0.090247221 30 emnlp-2012-Constructing Task-Specific Taxonomies for Document Collection Browsing
9 0.089810401 40 emnlp-2012-Ensemble Semantics for Large-scale Unsupervised Relation Extraction
10 0.084922239 5 emnlp-2012-A Discriminative Model for Query Spelling Correction with Latent Structural SVM
11 0.073121883 78 emnlp-2012-Learning Lexicon Models from Search Logs for Query Expansion
12 0.072274745 76 emnlp-2012-Learning-based Multi-Sieve Co-reference Resolution with Knowledge
13 0.066479094 62 emnlp-2012-Identifying Constant and Unique Relations by using Time-Series Text
14 0.059517041 96 emnlp-2012-Name Phylogeny: A Generative Model of String Variation
15 0.057316601 100 emnlp-2012-Open Language Learning for Information Extraction
16 0.056787822 84 emnlp-2012-Linking Named Entities to Any Database
17 0.053877827 41 emnlp-2012-Entity based QA Retrieval
18 0.053667698 85 emnlp-2012-Local and Global Context for Supervised and Unsupervised Metonymy Resolution
19 0.053039305 112 emnlp-2012-Resolving Complex Cases of Definite Pronouns: The Winograd Schema Challenge
20 0.050235748 10 emnlp-2012-A Statistical Relational Learning Approach to Identifying Evidence Based Medicine Categories
topicId topicWeight
[(0, 0.205), (1, 0.154), (2, 0.026), (3, -0.048), (4, -0.007), (5, -0.1), (6, 0.153), (7, 0.212), (8, -0.122), (9, 0.052), (10, 0.017), (11, -0.026), (12, 0.007), (13, 0.011), (14, -0.046), (15, -0.013), (16, -0.053), (17, 0.064), (18, 0.021), (19, 0.102), (20, -0.016), (21, 0.044), (22, -0.004), (23, -0.132), (24, 0.092), (25, 0.027), (26, -0.078), (27, 0.062), (28, -0.028), (29, 0.064), (30, -0.081), (31, -0.074), (32, -0.065), (33, -0.183), (34, -0.078), (35, -0.132), (36, 0.005), (37, -0.115), (38, 0.098), (39, 0.058), (40, 0.031), (41, 0.052), (42, 0.089), (43, 0.122), (44, 0.029), (45, -0.12), (46, 0.064), (47, 0.087), (48, -0.038), (49, -0.036)]
simIndex simValue paperId paperTitle
same-paper 1 0.9471271 110 emnlp-2012-Reading The Web with Learned Syntactic-Semantic Inference Rules
Author: Ni Lao ; Amarnag Subramanya ; Fernando Pereira ; William W. Cohen
Abstract: We study how to extend a large knowledge base (Freebase) by reading relational information from a large Web text corpus. Previous studies on extracting relational knowledge from text show the potential of syntactic patterns for extraction, but they do not exploit background knowledge of other relations in the knowledge base. We describe a distributed, Web-scale implementation of a path-constrained random walk model that learns syntactic-semantic inference rules for binary relations from a graph representation of the parsed text and the knowledge base. Experiments show significant accuracy improvements in binary relation prediction over methods that consider only text, or only the existing knowledge base.
2 0.74015284 30 emnlp-2012-Constructing Task-Specific Taxonomies for Document Collection Browsing
Author: Hui Yang
Abstract: Taxonomies can serve as browsing tools for document collections. However, given an arbitrary collection, pre-constructed taxonomies could not easily adapt to the specific topic/task present in the collection. This paper explores techniques to quickly derive task-specific taxonomies supporting browsing in arbitrary document collections. The supervised approach directly learns semantic distances from users to propose meaningful task-specific taxonomies. The approach aims to produce globally optimized taxonomy structures by incorporating path consistency control and usergenerated task specification into the general learning framework. A comparison to stateof-the-art systems and a user study jointly demonstrate that our techniques are highly effective. .
3 0.57945091 97 emnlp-2012-Natural Language Questions for the Web of Data
Author: Mohamed Yahya ; Klaus Berberich ; Shady Elbassuoni ; Maya Ramanath ; Volker Tresp ; Gerhard Weikum
Abstract: The Linked Data initiative comprises structured databases in the Semantic-Web data model RDF. Exploring this heterogeneous data by structured query languages is tedious and error-prone even for skilled users. To ease the task, this paper presents a methodology for translating natural language questions into structured SPARQL queries over linked-data sources. Our method is based on an integer linear program to solve several disambiguation tasks jointly: the segmentation of questions into phrases; the mapping of phrases to semantic entities, classes, and relations; and the construction of SPARQL triple patterns. Our solution harnesses the rich type system provided by knowledge bases in the web of linked data, to constrain our semantic-coherence objective function. We present experiments on both the . in question translation and the resulting query answering.
4 0.56102633 103 emnlp-2012-PATTY: A Taxonomy of Relational Patterns with Semantic Types
Author: Ndapandula Nakashole ; Gerhard Weikum ; Fabian Suchanek
Abstract: This paper presents PATTY: a large resource for textual patterns that denote binary relations between entities. The patterns are semantically typed and organized into a subsumption taxonomy. The PATTY system is based on efficient algorithms for frequent itemset mining and can process Web-scale corpora. It harnesses the rich type system and entity population of large knowledge bases. The PATTY taxonomy comprises 350,569 pattern synsets. Random-sampling-based evaluation shows a pattern accuracy of 84.7%. PATTY has 8,162 subsumptions, with a random-sampling-based precision of 75%. The PATTY resource is freely available for interactive access and download.
5 0.52830493 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers
Author: Jayant Krishnamurthy ; Tom Mitchell
Abstract: We present a method for training a semantic parser using only a knowledge base and an unlabeled text corpus, without any individually annotated sentences. Our key observation is that multiple forms ofweak supervision can be combined to train an accurate semantic parser: semantic supervision from a knowledge base, and syntactic supervision from dependencyparsed sentences. We apply our approach to train a semantic parser that uses 77 relations from Freebase in its knowledge representation. This semantic parser extracts instances of binary relations with state-of-theart accuracy, while simultaneously recovering much richer semantic structures, such as conjunctions of multiple relations with partially shared arguments. We demonstrate recovery of this richer structure by extracting logical forms from natural language queries against Freebase. On this task, the trained semantic parser achieves 80% precision and 56% recall, despite never having seen an annotated logical form.
6 0.48158938 47 emnlp-2012-Explore Person Specific Evidence in Web Person Name Disambiguation
7 0.43998498 40 emnlp-2012-Ensemble Semantics for Large-scale Unsupervised Relation Extraction
8 0.43535361 85 emnlp-2012-Local and Global Context for Supervised and Unsupervised Metonymy Resolution
9 0.40185156 93 emnlp-2012-Multi-instance Multi-label Learning for Relation Extraction
10 0.37697592 78 emnlp-2012-Learning Lexicon Models from Search Logs for Query Expansion
11 0.3753652 100 emnlp-2012-Open Language Learning for Information Extraction
12 0.35557622 62 emnlp-2012-Identifying Constant and Unique Relations by using Time-Series Text
13 0.34777573 98 emnlp-2012-No Noun Phrase Left Behind: Detecting and Typing Unlinkable Entities
14 0.33756745 10 emnlp-2012-A Statistical Relational Learning Approach to Identifying Evidence Based Medicine Categories
15 0.30127442 5 emnlp-2012-A Discriminative Model for Query Spelling Correction with Latent Structural SVM
16 0.28919154 125 emnlp-2012-Towards Efficient Named-Entity Rule Induction for Customizability
17 0.26701489 96 emnlp-2012-Name Phylogeny: A Generative Model of String Variation
18 0.26392025 32 emnlp-2012-Detecting Subgroups in Online Discussions by Modeling Positive and Negative Relations among Participants
19 0.26302263 17 emnlp-2012-An "AI readability" Formula for French as a Foreign Language
20 0.26147228 26 emnlp-2012-Building a Lightweight Semantic Model for Unsupervised Information Extraction on Short Listings
topicId topicWeight
[(2, 0.027), (12, 0.012), (14, 0.011), (16, 0.028), (25, 0.023), (34, 0.072), (60, 0.095), (63, 0.072), (64, 0.017), (65, 0.053), (70, 0.014), (73, 0.021), (74, 0.039), (76, 0.061), (80, 0.012), (86, 0.025), (94, 0.016), (95, 0.044), (98, 0.269)]
simIndex simValue paperId paperTitle
same-paper 1 0.75940561 110 emnlp-2012-Reading The Web with Learned Syntactic-Semantic Inference Rules
Author: Ni Lao ; Amarnag Subramanya ; Fernando Pereira ; William W. Cohen
Abstract: We study how to extend a large knowledge base (Freebase) by reading relational information from a large Web text corpus. Previous studies on extracting relational knowledge from text show the potential of syntactic patterns for extraction, but they do not exploit background knowledge of other relations in the knowledge base. We describe a distributed, Web-scale implementation of a path-constrained random walk model that learns syntactic-semantic inference rules for binary relations from a graph representation of the parsed text and the knowledge base. Experiments show significant accuracy improvements in binary relation prediction over methods that consider only text, or only the existing knowledge base.
2 0.53175837 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers
Author: Jayant Krishnamurthy ; Tom Mitchell
Abstract: We present a method for training a semantic parser using only a knowledge base and an unlabeled text corpus, without any individually annotated sentences. Our key observation is that multiple forms ofweak supervision can be combined to train an accurate semantic parser: semantic supervision from a knowledge base, and syntactic supervision from dependencyparsed sentences. We apply our approach to train a semantic parser that uses 77 relations from Freebase in its knowledge representation. This semantic parser extracts instances of binary relations with state-of-theart accuracy, while simultaneously recovering much richer semantic structures, such as conjunctions of multiple relations with partially shared arguments. We demonstrate recovery of this richer structure by extracting logical forms from natural language queries against Freebase. On this task, the trained semantic parser achieves 80% precision and 56% recall, despite never having seen an annotated logical form.
Author: Jianxing Yu ; Zheng-Jun Zha ; Tat-Seng Chua
Abstract: This paper proposes to generate appropriate answers for opinion questions about products by exploiting the hierarchical organization of consumer reviews. The hierarchy organizes product aspects as nodes following their parent-child relations. For each aspect, the reviews and corresponding opinions on this aspect are stored. We develop a new framework for opinion Questions Answering, which enables accurate question analysis and effective answer generation by making use the hierarchy. In particular, we first identify the (explicit/implicit) product aspects asked in the questions and their sub-aspects by referring to the hierarchy. We then retrieve the corresponding review fragments relevant to the aspects from the hierarchy. In order to gener- ate appropriate answers from the review fragments, we develop a multi-criteria optimization approach for answer generation by simultaneously taking into account review salience, coherence, diversity, and parent-child relations among the aspects. We conduct evaluations on 11 popular products in four domains. The evaluated corpus contains 70,359 consumer reviews and 220 questions on these products. Experimental results demonstrate the effectiveness of our approach.
4 0.51904219 89 emnlp-2012-Mixed Membership Markov Models for Unsupervised Conversation Modeling
Author: Michael J. Paul
Abstract: Recent work has explored the use of hidden Markov models for unsupervised discourse and conversation modeling, where each segment or block of text such as a message in a conversation is associated with a hidden state in a sequence. We extend this approach to allow each block of text to be a mixture of multiple classes. Under our model, the probability of a class in a text block is a log-linear function of the classes in the previous block. We show that this model performs well at predictive tasks on two conversation data sets, improving thread reconstruction accuracy by up to 15 percentage points over a standard HMM. Additionally, we show quantitatively that the induced word clusters correspond to speech acts more closely than baseline models.
5 0.51614845 47 emnlp-2012-Explore Person Specific Evidence in Web Person Name Disambiguation
Author: Liwei Chen ; Yansong Feng ; Lei Zou ; Dongyan Zhao
Abstract: In this paper, we investigate different usages of feature representations in the web person name disambiguation task which has been suffering from the mismatch of vocabulary and lack of clues in web environments. In literature, the latter receives less attention and remains more challenging. We explore the feature space in this task and argue that collecting person specific evidences from a corpus level can provide a more reasonable and robust estimation for evaluating a feature’s importance in a given web page. This can alleviate the lack of clues where discriminative features can be reasonably weighted by taking their corpus level importance into account, not just relying on the current local context. We therefore propose a topic-based model to exploit the person specific global importance and embed it into the person name similarity. The experimental results show that the corpus level topic in- formation provides more stable evidences for discriminative features and our method outperforms the state-of-the-art systems on three WePS datasets.
6 0.51517379 14 emnlp-2012-A Weakly Supervised Model for Sentence-Level Semantic Orientation Analysis with Multiple Experts
7 0.51289308 71 emnlp-2012-Joint Entity and Event Coreference Resolution across Documents
8 0.50979275 103 emnlp-2012-PATTY: A Taxonomy of Relational Patterns with Semantic Types
9 0.50652242 124 emnlp-2012-Three Dependency-and-Boundary Models for Grammar Induction
10 0.50570881 97 emnlp-2012-Natural Language Questions for the Web of Data
11 0.50503445 23 emnlp-2012-Besting the Quiz Master: Crowdsourcing Incremental Classification Games
12 0.50084645 3 emnlp-2012-A Coherence Model Based on Syntactic Patterns
13 0.49850446 30 emnlp-2012-Constructing Task-Specific Taxonomies for Document Collection Browsing
14 0.49808234 18 emnlp-2012-An Empirical Investigation of Statistical Significance in NLP
15 0.49787194 93 emnlp-2012-Multi-instance Multi-label Learning for Relation Extraction
16 0.49766991 114 emnlp-2012-Revisiting the Predictability of Language: Response Completion in Social Media
17 0.49707842 107 emnlp-2012-Polarity Inducing Latent Semantic Analysis
18 0.49664116 42 emnlp-2012-Entropy-based Pruning for Phrase-based Machine Translation
19 0.49609959 51 emnlp-2012-Extracting Opinion Expressions with semi-Markov Conditional Random Fields
20 0.49605468 77 emnlp-2012-Learning Constraints for Consistent Timeline Extraction