emnlp emnlp2011 emnlp2011-26 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Zornitsa Kozareva ; Konstantin Voevodski ; Shanghua Teng
Abstract: Class-instance label propagation algorithms have been successfully used to fuse information from multiple sources in order to enrich a set of unlabeled instances with class labels. Yet, nobody has explored the relationships between the instances themselves to enhance an initial set of class-instance pairs. We propose two graph-theoretic methods (centrality and regularization), which start with a small set of labeled class-instance pairs and use the instance-instance network to extend the class labels to all instances in the network. We carry out a comparative study with state-of-the-art knowledge harvesting algorithm and show that our approach can learn additional class labels while maintaining high accuracy. We conduct a comparative study between class-instance and instance-instance graphs used to propagate the class labels and show that the latter one achieves higher accuracy.
Reference: text
sentIndex sentText sentNum sentScore
1 Class Label Enhancement via Related Instances Zornitsa Kozareva USC Information Sciences Institute 4676 Admiralty Way Marina del Rey, CA 90292-6695 ko z areva @ i i s . [sent-1, score-0.08]
2 edu Abstract Class-instance label propagation algorithms have been successfully used to fuse information from multiple sources in order to enrich a set of unlabeled instances with class labels. [sent-2, score-0.645]
3 Yet, nobody has explored the relationships between the instances themselves to enhance an initial set of class-instance pairs. [sent-3, score-0.365]
4 We propose two graph-theoretic methods (centrality and regularization), which start with a small set of labeled class-instance pairs and use the instance-instance network to extend the class labels to all instances in the network. [sent-4, score-0.454]
5 We carry out a comparative study with state-of-the-art knowledge harvesting algorithm and show that our approach can learn additional class labels while maintaining high accuracy. [sent-5, score-0.712]
6 We conduct a comparative study between class-instance and instance-instance graphs used to propagate the class labels and show that the latter one achieves higher accuracy. [sent-6, score-0.819]
7 1 Introduction Many natural language processing applications use and rely on semantic knowledge resources. [sent-7, score-0.094]
8 Since manually built lexical repositories such as WordNet (Fellbaum, 1998) cover a limited amount of knowledge and are tedious to maintain over time, researchers have developed algorithms for automatic knowledge extraction from structured and unstructured texts. [sent-8, score-0.423]
9 There is a substantial body of work on extracting is-a relations (Etzioni et al. [sent-9, score-0.095]
10 , 2003; Pantel and Pennacchiotti, 2006) and general facts (Lin and Pantel, 2001 ;Davidov and Rappoport, 118 Konstantin Voevodski Boston University Shang-Hua Teng University of Southern California 111Cummington St. [sent-12, score-0.048]
11 The usefulness of the generated resources has been shown to be valuable to information extraction (Riloff and Jones, 1999), question answering (Katz et al. [sent-16, score-0.09]
12 Among the most common knowledge acquisition approaches are those based on lexical patterns (Hearst, 1992; Etzioni et al. [sent-19, score-0.051]
13 , 2008) and clustering (Lin and Pantel, 2002; Davidov and Rappoport, 2008). [sent-21, score-0.046]
14 While clustering can find instances and classes that are not explicitly expressed in text, they often may not generate the granularity needed by the users. [sent-22, score-0.182]
15 In contrast, pattern-based approaches generate highly accurate lists, but they are constraint to the information matched by the pattern and often suffer from recall. [sent-23, score-0.036]
16 , 2006; Kozareva and Hovy, 2010) have shown that complete lists of semantic classes and instances are valuable for the enrichment of existing resources like WordNet and for taxonomy induction. [sent-25, score-0.315]
17 Therefore, researchers have focused on the development of methods that can automatically augment the initially extracted class-instance pairs. [sent-26, score-0.1]
18 (Pennacchiotti and Pantel, 2009) fused information from pattern-based and distributional systems using an ensemble method and a rich set of features derived from query logs, web-crawl and Wikipedia. [sent-27, score-0.118]
19 , 2008) improved class-instance extractions exploring the relationships between the classes and the instances to propagate the initial class-labels to the remaining unlabeled instances. [sent-29, score-0.697]
20 Later on (Talukdar and Pereira, 2010) showed that class-instance extraction with label propagation can be further improved by adding semantic information Proce Ed iningbsu orfg th ,e S 2c0o1tl1an Cdo,n UfeKr,en Jcuely on 27 E–m31p,ir 2ic0a1l1 M. [sent-30, score-0.275]
21 ec th2o0d1s1 i Ans Nsoactuiartaioln La fonrg Cuaogmep Purtoatcieosnsainlg L,in pgaugies ti 1c1s8–128, in the form of instance-attribute edges derived from independently developed knowledge base. [sent-32, score-0.087]
22 , 2008) and (Talukdar and Pereira, 2010), we are interested in enriching class-instance extractions with label propagation. [sent-34, score-0.188]
23 However, unlike the previous work, we model the relationships between the instances themselves to propagate the initial set of class labels to the remaining unlabeled instances. [sent-35, score-0.806]
24 To our knowledge, this is the first work to explore the connections between instances for the task of class-label propagation. [sent-36, score-0.19]
25 Our work addresses the following question: Is it possible to effectively explore the structure of the text-mined instance-instance networks to enhance an incomplete set of class labels? [sent-37, score-0.202]
26 Our intuition is that if an instance like bear belongs to a semantic class carnivore, and the instance bear is connected to the instance fox, then it is more likely that the unlabeled instance fox is also of class carnivore. [sent-38, score-0.904]
27 To solve this problem, we propose two graph-based approaches that use the structure of the instanceinstance graph to propagate the class labels. [sent-39, score-0.651]
28 Our methods are agnostic to the sources of semantic instances and classes. [sent-40, score-0.305]
29 In this work, we carried out experiments with a state-of-the-art instance extraction system and conducted a comparative study between the original and the enhanced class-instance pairs. [sent-41, score-0.324]
30 The results show that this labeling procedure can begin to bridge the gap between the extraction power of the pattern-based approaches and the desired recall by finding class-instance pairs that are not explicitly mentioned in text. [sent-42, score-0.049]
31 The contributions of the paper are as follows: • We use only the relationships between the in- • • sWtaen cuesse t ohnelmys tehlvee rse tlaot propagate ecltwasese lnab tehles. [sent-43, score-0.475]
32 i We observe how often labels are propagated along tsheer edges wof o our s leambaenlstic a network, aatendd propose two ways to extend an initial set of class labels to all the instance nodes in the network. [sent-44, score-0.451]
33 The first approach uses a linear system to compute the network centrality relative to the initially labeled instances. [sent-45, score-0.299]
34 The second approach uses a regularization framework with respect to a random walk on the network. [sent-46, score-0.144]
35 We evaluate the proposed approaches and show tWhaet they edi tshceov preor many new cchleasss a-nindst sahnocwe pairs compared to state-of-the-art knowledge 119 harvesting algorithm, while still maintaining high accuracy. [sent-47, score-0.279]
36 • We conduct a comparative study between classiWnsetacnocned actnda oinmsptaanractei-vinessttaundcye graphs ulassesdto propagate class labels. [sent-48, score-0.748]
37 The experiments show that considering relationships between instances achieves higher accuracy. [sent-49, score-0.235]
38 Section 3 describes the Web-based knowledge harvesting algorithm used to extract the instance network and the class-instance pairs necessary for our experimental evaluation. [sent-52, score-0.37]
39 Section 4 describes the two graphtheoretic methods for class label propagation using an instance-instance network. [sent-53, score-0.337]
40 Section 5 shows a comparative study between the proposed graph algorithms and different baselines. [sent-54, score-0.256]
41 We also show a comparison between class-instance and instanceinstance graphs used in the label propagation. [sent-55, score-0.311]
42 2 Related Work In the past decade, we have reached a good understanding on the knowledge harvesting technology from structured (Suchanek et al. [sent-57, score-0.246]
43 Researchers have harvested with varying success semantic lexicons (Riloff and Shepherd, 1997) and concept lists (Katz et al. [sent-59, score-0.236]
44 Many efforts have also focused on the extraction of is-a relations (Hearst, 1992; Pas ¸ca, 2004; Etzioni et al. [sent-61, score-0.097]
45 , 2003; Pantel and Pennacchiotti, 2006) and general facts (Etzioni et al. [sent-64, score-0.048]
46 Various approaches have been proposed following the patterns of (Hearst, 1992) and clustering (Lin and Pantel, 2002; Davidov and Rappoport, 2008). [sent-66, score-0.046]
47 A substantial body of work has explored issues such as reranking the harvested knowledge using mutual information (Etzioni et al. [sent-67, score-0.245]
48 , 2009), estimating the goodness of textmining seeds (Vyas et al. [sent-69, score-0.044]
49 , 2007b) and inducing term taxonomies with WordNet (Snow et al. [sent-72, score-0.041]
50 , 2006) or starting from scratch (Kozareva and Hovy, 2010). [sent-73, score-0.039]
51 Since pattern-based approaches tend to be highprecision and low-recall in nature, recently of great interest to the research community is the development of approaches that can increment the recall of the harvested class-instance pairs. [sent-74, score-0.188]
52 (Pennacchiotti and Pantel, 2009) proposed an ensemble semantic framework that mixes distributional and patternbased systems with a large set of features from a web-crawl, query logs, and Wikipedia. [sent-75, score-0.151]
53 , 2008) combined extractions from free text and structured sources using graph-based label propagation algorithm. [sent-77, score-0.395]
54 (Talukdar and Pereira, 2010) conducted a comparative study of graph algorithms and showed that class-instance extraction can be improved using additional information that can be modeled as instance-attribute edges. [sent-78, score-0.305]
55 , 2008; Talukdar and Pereira, 2010) who model classinstance relations to propagate class-labels. [sent-80, score-0.375]
56 Although these algorithms can be applied to other relations (Alfonseca et al. [sent-81, score-0.048]
57 , 2010), to our knowledge yet nobody has modeled the connections between the instances themselves for the task of class-label propagation. [sent-82, score-0.323]
58 We propose regularization and centrality graph-theoretic methods, which exploit the instanceinstance network and a small set of class-instance pairs to propagate the class-labels to the remaining unlabeled instances. [sent-83, score-0.851]
59 While objectives similar to regularization have been used for class-label propagation, the application of node centrality for this task is also novel. [sent-84, score-0.241]
60 3 Knowledge Harvesting from the Web Our proposed class-label enhancement approaches are agnostic to the sources of semantic instances and classes. [sent-86, score-0.387]
61 Several methods have been developed to harvest instances from the Web (Pa ¸sca, 2004; Etzioni et al. [sent-87, score-0.216]
62 , 2005; Pas ¸ca, 2007), it is easy to implement and requires minimum supervision (only one seed instance and a 120 lexico-syntactic pattern). [sent-92, score-0.135]
63 For a given semantic class of interest say animals, the algorithm starts with a seed example of the class, say whales. [sent-93, score-0.265]
wordName wordTfidf (topN-words)
[('talukdar', 0.329), ('kozareva', 0.309), ('propagate', 0.278), ('pantel', 0.205), ('etzioni', 0.188), ('davidov', 0.178), ('instanceinstance', 0.171), ('comparative', 0.166), ('harvesting', 0.159), ('centrality', 0.154), ('class', 0.154), ('harvested', 0.147), ('rappoport', 0.146), ('pennacchiotti', 0.139), ('pas', 0.139), ('instances', 0.136), ('extractions', 0.116), ('propagation', 0.111), ('hearst', 0.104), ('relationships', 0.099), ('sca', 0.098), ('network', 0.093), ('fox', 0.089), ('regularization', 0.087), ('katz', 0.082), ('nobody', 0.082), ('enhancement', 0.082), ('logs', 0.077), ('jain', 0.077), ('ca', 0.074), ('pereira', 0.073), ('label', 0.072), ('labels', 0.071), ('girju', 0.069), ('ensemble', 0.069), ('maintaining', 0.069), ('hovy', 0.068), ('graphs', 0.068), ('seed', 0.068), ('unlabeled', 0.068), ('instance', 0.067), ('agnostic', 0.066), ('bear', 0.064), ('cafarella', 0.064), ('unstructured', 0.064), ('sources', 0.06), ('riloff', 0.059), ('snow', 0.057), ('walk', 0.057), ('connections', 0.054), ('boston', 0.054), ('wordnet', 0.053), ('initially', 0.052), ('knowledge', 0.051), ('extraction', 0.049), ('marina', 0.049), ('repositories', 0.049), ('bloom', 0.049), ('admiralty', 0.049), ('animals', 0.049), ('classinstance', 0.049), ('enrichment', 0.049), ('fused', 0.049), ('lnab', 0.049), ('rse', 0.049), ('twhaet', 0.049), ('wof', 0.049), ('researchers', 0.048), ('enhance', 0.048), ('facts', 0.048), ('relations', 0.048), ('graph', 0.048), ('body', 0.047), ('clustering', 0.046), ('lists', 0.046), ('rey', 0.044), ('southern', 0.044), ('enrich', 0.044), ('goodness', 0.044), ('harvest', 0.044), ('lations', 0.044), ('semantic', 0.043), ('study', 0.042), ('valuable', 0.041), ('del', 0.041), ('alfonseca', 0.041), ('increment', 0.041), ('suchanek', 0.041), ('taxonomies', 0.041), ('zanzotto', 0.041), ('zornitsa', 0.041), ('conduct', 0.04), ('scratch', 0.039), ('ko', 0.039), ('tedious', 0.039), ('mixes', 0.039), ('propagated', 0.039), ('developed', 0.036), ('structured', 0.036), ('pattern', 0.036)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999976 26 emnlp-2011-Class Label Enhancement via Related Instances
Author: Zornitsa Kozareva ; Konstantin Voevodski ; Shanghua Teng
Abstract: Class-instance label propagation algorithms have been successfully used to fuse information from multiple sources in order to enrich a set of unlabeled instances with class labels. Yet, nobody has explored the relationships between the instances themselves to enhance an initial set of class-instance pairs. We propose two graph-theoretic methods (centrality and regularization), which start with a small set of labeled class-instance pairs and use the instance-instance network to extend the class labels to all instances in the network. We carry out a comparative study with state-of-the-art knowledge harvesting algorithm and show that our approach can learn additional class labels while maintaining high accuracy. We conduct a comparative study between class-instance and instance-instance graphs used to propagate the class labels and show that the latter one achieves higher accuracy.
2 0.11051948 113 emnlp-2011-Relation Acquisition using Word Classes and Partial Patterns
Author: Stijn De Saeger ; Kentaro Torisawa ; Masaaki Tsuchida ; Jun'ichi Kazama ; Chikara Hashimoto ; Ichiro Yamada ; Jong Hoon Oh ; Istvan Varga ; Yulan Yan
Abstract: This paper proposes a semi-supervised relation acquisition method that does not rely on extraction patterns (e.g. “X causes Y” for causal relations) but instead learns a combination of indirect evidence for the target relation semantic word classes and partial patterns. This method can extract long tail instances of semantic relations like causality from rare and complex expressions in a large Japanese Web corpus in extreme cases, patterns that occur only once in the entire corpus. Such patterns are beyond the reach ofcurrent pattern based methods. We show that our method performs on par with state-of-the-art pattern based methods, and maintains a reasonable level of accuracy even for instances — — acquired from infrequent patterns. This ability to acquire long tail instances is crucial for risk management and innovation, where an exhaustive database of high-level semantic relations like causation is of vital importance.
3 0.08992736 145 emnlp-2011-Unsupervised Semantic Role Induction with Graph Partitioning
Author: Joel Lang ; Mirella Lapata
Abstract: In this paper we present a method for unsupervised semantic role induction which we formalize as a graph partitioning problem. Argument instances of a verb are represented as vertices in a graph whose edge weights quantify their role-semantic similarity. Graph partitioning is realized with an algorithm that iteratively assigns vertices to clusters based on the cluster assignments of neighboring vertices. Our method is algorithmically and conceptually simple, especially with respect to how problem-specific knowledge is incorporated into the model. Experimental results on the CoNLL 2008 benchmark dataset demonstrate that our model is competitive with other unsupervised approaches in terms of F1 whilst attaining significantly higher cluster purity.
4 0.087058522 57 emnlp-2011-Extreme Extraction - Machine Reading in a Week
Author: Marjorie Freedman ; Lance Ramshaw ; Elizabeth Boschee ; Ryan Gabbard ; Gary Kratkiewicz ; Nicolas Ward ; Ralph Weischedel
Abstract: We report on empirical results in extreme extraction. It is extreme in that (1) from receipt of the ontology specifying the target concepts and relations, development is limited to one week and that (2) relatively little training data is assumed. We are able to surpass human recall and achieve an F1 of 0.5 1 on a question-answering task with less than 50 hours of effort using a hybrid approach that mixes active learning, bootstrapping, and limited (5 hours) manual rule writing. We compare the performance of three systems: extraction with handwritten rules, bootstrapped extraction, and a combination. We show that while the recall of the handwritten rules surpasses that of the learned system, the learned system is able to improve the overall recall and F1.
5 0.074993163 128 emnlp-2011-Structured Relation Discovery using Generative Models
Author: Limin Yao ; Aria Haghighi ; Sebastian Riedel ; Andrew McCallum
Abstract: We explore unsupervised approaches to relation extraction between two named entities; for instance, the semantic bornIn relation between a person and location entity. Concretely, we propose a series of generative probabilistic models, broadly similar to topic models, each which generates a corpus of observed triples of entity mention pairs and the surface syntactic dependency path between them. The output of each model is a clustering of observed relation tuples and their associated textual expressions to underlying semantic relation types. Our proposed models exploit entity type constraints within a relation as well as features on the dependency path between entity mentions. We examine effectiveness of our approach via multiple evaluations and demonstrate 12% error reduction in precision over a state-of-the-art weakly supervised baseline.
6 0.072438642 28 emnlp-2011-Closing the Loop: Fast, Interactive Semi-Supervised Annotation With Queries on Features and Instances
7 0.067255482 70 emnlp-2011-Identifying Relations for Open Information Extraction
8 0.064098001 78 emnlp-2011-Large-Scale Noun Compound Interpretation Using Bootstrapping and the Web as a Corpus
9 0.058761157 64 emnlp-2011-Harnessing different knowledge sources to measure semantic relatedness under a uniform model
10 0.056943957 14 emnlp-2011-A generative model for unsupervised discovery of relations and argument classes from clinical texts
11 0.052659094 109 emnlp-2011-Random Walk Inference and Learning in A Large Scale Knowledge Base
12 0.04876684 9 emnlp-2011-A Non-negative Matrix Factorization Based Approach for Active Dual Supervision from Document and Word Labels
13 0.04712452 88 emnlp-2011-Linear Text Segmentation Using Affinity Propagation
14 0.044089217 98 emnlp-2011-Named Entity Recognition in Tweets: An Experimental Study
15 0.042496238 1 emnlp-2011-A Bayesian Mixture Model for PoS Induction Using Multiple Features
16 0.041353349 96 emnlp-2011-Multilayer Sequence Labeling
17 0.039016042 144 emnlp-2011-Unsupervised Learning of Selectional Restrictions and Detection of Argument Coercions
18 0.038664233 40 emnlp-2011-Discovering Relations between Noun Categories
19 0.037654079 4 emnlp-2011-A Fast, Accurate, Non-Projective, Semantically-Enriched Parser
20 0.036753953 129 emnlp-2011-Structured Sparsity in Structured Prediction
topicId topicWeight
[(0, 0.135), (1, -0.104), (2, -0.095), (3, 0.052), (4, -0.022), (5, -0.12), (6, 0.039), (7, -0.02), (8, -0.053), (9, 0.106), (10, 0.045), (11, -0.009), (12, -0.044), (13, -0.02), (14, -0.043), (15, 0.061), (16, -0.075), (17, -0.076), (18, -0.044), (19, 0.066), (20, 0.111), (21, -0.008), (22, -0.003), (23, 0.077), (24, 0.089), (25, 0.049), (26, -0.016), (27, 0.071), (28, -0.06), (29, 0.057), (30, -0.055), (31, 0.017), (32, -0.121), (33, 0.141), (34, 0.012), (35, -0.136), (36, -0.131), (37, 0.049), (38, -0.112), (39, -0.118), (40, 0.066), (41, -0.003), (42, -0.015), (43, 0.076), (44, 0.259), (45, -0.03), (46, 0.094), (47, 0.245), (48, -0.175), (49, -0.118)]
simIndex simValue paperId paperTitle
same-paper 1 0.97109705 26 emnlp-2011-Class Label Enhancement via Related Instances
Author: Zornitsa Kozareva ; Konstantin Voevodski ; Shanghua Teng
Abstract: Class-instance label propagation algorithms have been successfully used to fuse information from multiple sources in order to enrich a set of unlabeled instances with class labels. Yet, nobody has explored the relationships between the instances themselves to enhance an initial set of class-instance pairs. We propose two graph-theoretic methods (centrality and regularization), which start with a small set of labeled class-instance pairs and use the instance-instance network to extend the class labels to all instances in the network. We carry out a comparative study with state-of-the-art knowledge harvesting algorithm and show that our approach can learn additional class labels while maintaining high accuracy. We conduct a comparative study between class-instance and instance-instance graphs used to propagate the class labels and show that the latter one achieves higher accuracy.
2 0.52723289 113 emnlp-2011-Relation Acquisition using Word Classes and Partial Patterns
Author: Stijn De Saeger ; Kentaro Torisawa ; Masaaki Tsuchida ; Jun'ichi Kazama ; Chikara Hashimoto ; Ichiro Yamada ; Jong Hoon Oh ; Istvan Varga ; Yulan Yan
Abstract: This paper proposes a semi-supervised relation acquisition method that does not rely on extraction patterns (e.g. “X causes Y” for causal relations) but instead learns a combination of indirect evidence for the target relation semantic word classes and partial patterns. This method can extract long tail instances of semantic relations like causality from rare and complex expressions in a large Japanese Web corpus in extreme cases, patterns that occur only once in the entire corpus. Such patterns are beyond the reach ofcurrent pattern based methods. We show that our method performs on par with state-of-the-art pattern based methods, and maintains a reasonable level of accuracy even for instances — — acquired from infrequent patterns. This ability to acquire long tail instances is crucial for risk management and innovation, where an exhaustive database of high-level semantic relations like causation is of vital importance.
3 0.46025211 57 emnlp-2011-Extreme Extraction - Machine Reading in a Week
Author: Marjorie Freedman ; Lance Ramshaw ; Elizabeth Boschee ; Ryan Gabbard ; Gary Kratkiewicz ; Nicolas Ward ; Ralph Weischedel
Abstract: We report on empirical results in extreme extraction. It is extreme in that (1) from receipt of the ontology specifying the target concepts and relations, development is limited to one week and that (2) relatively little training data is assumed. We are able to surpass human recall and achieve an F1 of 0.5 1 on a question-answering task with less than 50 hours of effort using a hybrid approach that mixes active learning, bootstrapping, and limited (5 hours) manual rule writing. We compare the performance of three systems: extraction with handwritten rules, bootstrapped extraction, and a combination. We show that while the recall of the handwritten rules surpasses that of the learned system, the learned system is able to improve the overall recall and F1.
4 0.43838882 64 emnlp-2011-Harnessing different knowledge sources to measure semantic relatedness under a uniform model
Author: Ziqi Zhang ; Anna Lisa Gentile ; Fabio Ciravegna
Abstract: Measuring semantic relatedness between words or concepts is a crucial process to many Natural Language Processing tasks. Exiting methods exploit semantic evidence from a single knowledge source, and are predominantly evaluated only in the general domain. This paper introduces a method of harnessing different knowledge sources under a uniform model for measuring semantic relatedness between words or concepts. Using Wikipedia and WordNet as examples, and evaluated in both the general and biomedical domains, it successfully combines strengths from both knowledge sources and outperforms stateof-the-art on many datasets.
5 0.41104507 96 emnlp-2011-Multilayer Sequence Labeling
Author: Ai Azuma ; Yuji Matsumoto
Abstract: In this paper, we describe a novel approach to cascaded learning and inference on sequences. We propose a weakly joint learning model on cascaded inference on sequences, called multilayer sequence labeling. In this model, inference on sequences is modeled as cascaded decision. However, the decision on a sequence labeling sequel to other decisions utilizes the features on the preceding results as marginalized by the probabilistic models on them. It is not novel itself, but our idea central to this paper is that the probabilistic models on succeeding labeling are viewed as indirectly depending on the probabilistic models on preceding analyses. We also propose two types of efficient dynamic programming which are required in the gradient-based optimization of an objective function. One of the dynamic programming algorithms resembles back propagation algorithm for mul- tilayer feed-forward neural networks. The other is a generalized version of the forwardbackward algorithm. We also report experiments of cascaded part-of-speech tagging and chunking of English sentences and show effectiveness of the proposed method.
6 0.39513147 88 emnlp-2011-Linear Text Segmentation Using Affinity Propagation
7 0.37909526 78 emnlp-2011-Large-Scale Noun Compound Interpretation Using Bootstrapping and the Web as a Corpus
8 0.36722147 70 emnlp-2011-Identifying Relations for Open Information Extraction
9 0.36263365 144 emnlp-2011-Unsupervised Learning of Selectional Restrictions and Detection of Argument Coercions
10 0.34257564 28 emnlp-2011-Closing the Loop: Fast, Interactive Semi-Supervised Annotation With Queries on Features and Instances
11 0.33445999 145 emnlp-2011-Unsupervised Semantic Role Induction with Graph Partitioning
12 0.31187186 67 emnlp-2011-Hierarchical Verb Clustering Using Graph Factorization
13 0.29306105 81 emnlp-2011-Learning General Connotation of Words using Graph-based Algorithms
14 0.27065659 129 emnlp-2011-Structured Sparsity in Structured Prediction
15 0.23780304 109 emnlp-2011-Random Walk Inference and Learning in A Large Scale Knowledge Base
16 0.22397792 9 emnlp-2011-A Non-negative Matrix Factorization Based Approach for Active Dual Supervision from Document and Word Labels
17 0.21725771 68 emnlp-2011-Hypotheses Selection Criteria in a Reranking Framework for Spoken Language Understanding
18 0.20358054 46 emnlp-2011-Efficient Subsampling for Training Complex Language Models
19 0.18936048 12 emnlp-2011-A Weakly-supervised Approach to Argumentative Zoning of Scientific Documents
20 0.18660522 87 emnlp-2011-Lexical Generalization in CCG Grammar Induction for Semantic Parsing
topicId topicWeight
[(8, 0.45), (23, 0.089), (33, 0.01), (36, 0.037), (37, 0.015), (45, 0.055), (54, 0.012), (57, 0.038), (62, 0.026), (66, 0.037), (69, 0.013), (79, 0.033), (82, 0.02), (96, 0.064)]
simIndex simValue paperId paperTitle
same-paper 1 0.75547814 26 emnlp-2011-Class Label Enhancement via Related Instances
Author: Zornitsa Kozareva ; Konstantin Voevodski ; Shanghua Teng
Abstract: Class-instance label propagation algorithms have been successfully used to fuse information from multiple sources in order to enrich a set of unlabeled instances with class labels. Yet, nobody has explored the relationships between the instances themselves to enhance an initial set of class-instance pairs. We propose two graph-theoretic methods (centrality and regularization), which start with a small set of labeled class-instance pairs and use the instance-instance network to extend the class labels to all instances in the network. We carry out a comparative study with state-of-the-art knowledge harvesting algorithm and show that our approach can learn additional class labels while maintaining high accuracy. We conduct a comparative study between class-instance and instance-instance graphs used to propagate the class labels and show that the latter one achieves higher accuracy.
2 0.48696706 46 emnlp-2011-Efficient Subsampling for Training Complex Language Models
Author: Puyang Xu ; Asela Gunawardana ; Sanjeev Khudanpur
Abstract: We propose an efficient way to train maximum entropy language models (MELM) and neural network language models (NNLM). The advantage of the proposed method comes from a more robust and efficient subsampling technique. The original multi-class language modeling problem is transformed into a set of binary problems where each binary classifier predicts whether or not a particular word will occur. We show that the binarized model is as powerful as the standard model and allows us to aggressively subsample negative training examples without sacrificing predictive performance. Empirical results show that we can train MELM and NNLM at 1% ∼ 5% of the strtaaninda MrdE complexity LwMith a no %los ∼s 5in% performance.
3 0.32173836 78 emnlp-2011-Large-Scale Noun Compound Interpretation Using Bootstrapping and the Web as a Corpus
Author: Su Nam Kim ; Preslav Nakov
Abstract: Responding to the need for semantic lexical resources in natural language processing applications, we examine methods to acquire noun compounds (NCs), e.g., orange juice, together with suitable fine-grained semantic interpretations, e.g., squeezed from, which are directly usable as paraphrases. We employ bootstrapping and web statistics, and utilize the relationship between NCs and paraphrasing patterns to jointly extract NCs and such patterns in multiple alternating iterations. In evaluation, we found that having one compound noun fixed yields both a higher number of semantically interpreted NCs and improved accuracy due to stronger semantic restrictions.
4 0.28050804 128 emnlp-2011-Structured Relation Discovery using Generative Models
Author: Limin Yao ; Aria Haghighi ; Sebastian Riedel ; Andrew McCallum
Abstract: We explore unsupervised approaches to relation extraction between two named entities; for instance, the semantic bornIn relation between a person and location entity. Concretely, we propose a series of generative probabilistic models, broadly similar to topic models, each which generates a corpus of observed triples of entity mention pairs and the surface syntactic dependency path between them. The output of each model is a clustering of observed relation tuples and their associated textual expressions to underlying semantic relation types. Our proposed models exploit entity type constraints within a relation as well as features on the dependency path between entity mentions. We examine effectiveness of our approach via multiple evaluations and demonstrate 12% error reduction in precision over a state-of-the-art weakly supervised baseline.
5 0.27813435 28 emnlp-2011-Closing the Loop: Fast, Interactive Semi-Supervised Annotation With Queries on Features and Instances
Author: Burr Settles
Abstract: This paper describes DUALIST, an active learning annotation paradigm which solicits and learns from labels on both features (e.g., words) and instances (e.g., documents). We present a novel semi-supervised training algorithm developed for this setting, which is (1) fast enough to support real-time interactive speeds, and (2) at least as accurate as preexisting methods for learning with mixed feature and instance labels. Human annotators in user studies were able to produce near-stateof-the-art classifiers—on several corpora in a variety of application domains—with only a few minutes of effort.
6 0.27548999 116 emnlp-2011-Robust Disambiguation of Named Entities in Text
7 0.27367502 132 emnlp-2011-Syntax-Based Grammaticality Improvement using CCG and Guided Search
8 0.27265942 145 emnlp-2011-Unsupervised Semantic Role Induction with Graph Partitioning
9 0.27198988 70 emnlp-2011-Identifying Relations for Open Information Extraction
10 0.27093562 108 emnlp-2011-Quasi-Synchronous Phrase Dependency Grammars for Machine Translation
11 0.27061298 1 emnlp-2011-A Bayesian Mixture Model for PoS Induction Using Multiple Features
12 0.27031824 81 emnlp-2011-Learning General Connotation of Words using Graph-based Algorithms
13 0.27024996 136 emnlp-2011-Training a Parser for Machine Translation Reordering
14 0.27022246 8 emnlp-2011-A Model of Discourse Predictions in Human Sentence Processing
15 0.2698468 137 emnlp-2011-Training dependency parsers by jointly optimizing multiple objectives
16 0.26981279 37 emnlp-2011-Cross-Cutting Models of Lexical Semantics
17 0.26951596 57 emnlp-2011-Extreme Extraction - Machine Reading in a Week
18 0.26929981 126 emnlp-2011-Structural Opinion Mining for Graph-based Sentiment Representation
19 0.26878542 98 emnlp-2011-Named Entity Recognition in Tweets: An Experimental Study
20 0.26852468 107 emnlp-2011-Probabilistic models of similarity in syntactic context