DOCSLIB.ORG

End-To-End Reinforcement Learning for Automatic Taxonomy Induction

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
End-To-End Reinforcement Learning for Automatic Taxonomy Induction End-to-End Reinforcement Learning for Automatic Taxonomy Induction Yuning Mao1, Xiang Ren2, Jiaming Shen1, Xiaotao Gu1, Jiawei Han1 1Department of Computer Science, University of Illinois Urbana-Champaign, IL, USA 2Department of Computer Science, University of Southern California, CA, USA 1fyuningm2, js2, xiaotao2, [email protected] [email protected] Abstract a set of terms into a taxonomy based on relevant resources such as text corpora. We present a novel end-to-end reinforce- ment learning approach to automatic tax- Prior studies on automatic taxonomy induc- onomy induction from a set of terms. tion (Gupta et al., 2017; Camacho-Collados, 2017) While prior methods treat the problem as a often divide the problem into two sequential sub- two-phase task (i.e., detecting hypernymy tasks: (1) hypernymy detection (i.e., extracting pairs followed by organizing these pairs term pairs of “is-a” relation); and (2) hyper- into a tree-structured hierarchy), we ar- nymy organization (i.e., organizing is-a term pairs gue that such two-phase methods may suf- into a tree-structured hierarchy). Methods devel- fer from error propagation, and cannot ef- oped for hypernymy detection either harvest new fectively optimize metrics that capture the terms (Yamada et al., 2009; Kozareva and Hovy, holistic structure of a taxonomy. In our ap- 2010) or presume a vocabulary is given and study proach, the representations of term pairs term semantics (Snow et al., 2005; Fu et al., 2014; are learned using multiple sources of in- Tuan et al., 2016; Shwartz et al., 2016). The hy- formation and used to determine which pernymy pairs extracted in the first subtask form a term to select and where to place it on the noisy hypernym graph, which is then transformed taxonomy via a policy network. All com- into a tree-structured taxonomy in the hypernymy ponents are trained in an end-to-end man- organization subtask, using different graph prun- ner with cumulative rewards, measured ing methods including maximum spanning tree by a holistic tree metric over the train- (MST) (Bansal et al., 2014; Zhang et al., 2016), ing taxonomies. Experiments on two pub- minimum-cost flow (MCF) (Gupta et al., 2017) lic datasets of different domains show that and other pruning heuristics (Kozareva and Hovy, our approach outperforms prior state-of- 2010; Velardi et al., 2013; Faralli et al., 2015; the-art taxonomy induction methods up to Panchenko et al., 2016). 19.6% on ancestor F1. 1 However, these two-phase methods encounter 1 Introduction two major limitations. First, most of them ig- nore the taxonomy structure when estimating the Many tasks in natural language understanding probability that a term pair holds the hypernymy (e.g., information extraction (Demeester et al., relation. They estimate the probability of differ- 2016), question answering (Yang et al., 2017), and ent term pairs independently and the learned term textual entailment (Sammons, 2012)) rely on lexi- pair representations are fixed during hypernymy cal resources in the form of term taxonomies (cf. organization. In consequence, there is no feed- rightmost column in Fig. 1). However, most exist- back from the second phase to the first phase and ing taxonomies, such as WordNet (Miller, 1995) possibly wrong representations cannot be rectified and Cyc (Lenat, 1995), are manually curated and based on the results of hypernymy organization, thus may have limited coverage or become un- which causes the error propagation problem. Sec- available in some domains and languages. There- ondly, some methods (Bansal et al., 2014; Zhang fore, recent efforts have been focusing on auto- et al., 2016) do explore the taxonomy space by matic taxonomy induction, which aims to organize regarding the induction of taxonomy structure as 1Code and data can be found at https://github. inferring the conditional distribution of edges. In com/morningmoni/TaxoRL other words, they use the product of edge proba- Sennenhunde Appenzeller Sennenhunde Appenzeller Sennenhunde Appenzeller shepherd_dog collie root working_dog working_dog working_dog miniature_pinscher miniature_pinscher miniature_pinscher pinscher pinscher pinscher t=0 t=5 t=6 affenpinscher t=8 affenpinscher Figure 1: An illustrative example showing the process of taxonomy induction. The input vocabulary V0 is f“working dog”, “pinscher”, “shepherd dog”, ...g, and the initial taxonomy T0 is empty. We use a virtual “root” node to represent T0 at t = 0. At time t = 5, there are 5 terms on the taxonomy T5 and 3 terms left to be attached: Vt = f“shepherd dog”, “collie”, “affenpinscher”g. Suppose the term “affenpinscher” is selected and put under “pinscher”, then the remaining vocabulary Vt+1 at next time step becomes f“shepherd dog”, “collie”g. Finally, after jV0j time steps, all the terms are attached to the fg taxonomy and VjV0j = V8 = . A full taxonomy is then constructed from scratch. bilities to represent the taxonomy quality. How- the quality of constructed taxonomies. Second, we ever, the edges are treated equally, while in reality, use the same (noisy) hypernym graph as the input they contribute to the taxonomy differently. For of all compared methods, and demonstrate that our example, a high-level edge is likely to be more RL approach does better hypernymy organization important than a bottom-out edge because it has through optimizing metrics that can capture holis- much more influence on its descendants. In ad- tic taxonomy structure. dition, these methods cannot explicitly capture the Contributions. In summary, we have made the holistic taxonomy structure by optimizing global following contributions: (1) We propose a deep metrics. reinforcement learning approach to unify hyper- To address the above issues, we propose to nymy detection and organization so as to induct jointly conduct hypernymy detection and organi- taxonomies in an end-to-end manner. (2) We de- zation by learning term pair representations and sign a policy network to incorporate semantic in- constructing the taxonomy simultaneously. Since formation of term pairs and use cumulative re- it is infeasible to estimate the quality of all pos- wards to measure the quality of constructed tax- sible taxonomies, we design an end-to-end rein- onomies holistically. (3) Experiments on two pub- forcement learning (RL) model to combine the lic datasets from different domains demonstrate two phases. Specifically, we train an RL agent that the superior performance of our approach com- employs the term pair representations using multi- pared with state-of-the-art methods. We also show ple sources of information and determines which that our method can effectively reduce error prop- term to select and where to place it on the tax- agation and capture global taxonomy structure. onomy via a policy network. The feedback from hypernymy organization is propagated back to the hypernymy detection phase, based on which the 2 Automatic Taxonomy Induction term pair representations are adjusted. All compo- nents are trained in an end-to-end manner with cu- 2.1 Problem Definition mulative rewards, measured by a holistic tree met- We define a taxonomy T = (V; R) as a tree- ric over the training taxonomies. The probability structured hierarchy with term set V (i.e., vocab- of a full taxonomy is no longer a simple aggre- ulary), and edge set R (which indicates is-a rela- gated probability of its edges. Instead, we assess tionship between terms). A term v 2 V can be ei- an edge based on how much it can contribute to ther a unigram or a multi-word phrase. The task of the whole quality of the taxonomy. end-to-end taxonomy induction takes a set of train- We perform two sets of experiments to eval- ing taxonomies and related resources (e.g., back- uate the effectiveness of our proposed approach. ground text corpora) as input, and aims to learn a First, we test the end-to-end taxonomy induction model to construct a full taxonomy T by adding performance by comparing our approach with the terms from a given vocabulary V0 onto an empty state-of-the-art two-phase methods, and show that hierarchy T0 one at a time. An illustration of the our approach outperforms them significantly on taxonomy induction process is shown in Fig. 1. 2.2 Modeling Hypernymy Relation embeddings of x and y, which capture the distri- butional semantics of two terms. Determining which term to select from V0 and where to place it on the current hierarchy requires Surface String Features. In practice, even the understanding of the semantic relationships be- embeddings of many terms are missing because tween the selected term and all the other terms. the terms in the input vocabulary may be multi- We consider multiple sources of information (i.e., word phrases, proper nouns or named entities, resources) for learning hypernymy relation rep- which are likely not covered by the external pre- resentations of term pairs, including dependency trained word embeddings. To address this issue, path-based contextual embedding and distribu- we utilize several surface features described in tional term embeddings (Shwartz et al., 2016). previous studies (Yang and Callan, 2009; Bansal Path-based Information. We extract the shortest et al., 2014; Zhang et al., 2016). Specifically, we dependency paths between each co-occurring term employ Capitalization, Ends with, Contains, Suffix pair from sentences in the given background cor- match, Longest common substring and Length dif- pora. Each path is represented as a sequence of ference. These features are effective for detecting edges that goes from term x to term y in the de- hypernyms solely based on the term pairs. pendency tree, and each edge consists of the word Frequency and Generality Features. Another lemma, the part-of-speech tag, the dependency la- feature source that we employ is the hyper- bel and the edge direction between two contiguous nym candidates from TAXI3 (Panchenko et al., words.
Load more

Recommended publications
  • Arxiv:1809.06223V1 [Cs.CL] 17 Sep 2018
    Unsupervised Sense-Aware Hypernymy Extraction Dmitry Ustalov†, Alexander Panchenko‡, Chris Biemann‡, and Simone Paolo Ponzetto† †University of Mannheim, Germany fdmitry,[email protected] ‡University of Hamburg, Germany fpanchenko,[email protected] Abstract from text between two ambiguous words, e.g., apple fruit. However, by definition in Cruse In this paper, we show how unsupervised (1986), hypernymy is a binary relationship between sense representations can be used to im- senses, e.g., apple2 fruit1, where apple2 is the prove hypernymy extraction. We present “food” sense of the word “apple”. In turn, the word a method for extracting disambiguated hy- “apple” can be represented by multiple lexical units, pernymy relationships that propagate hy- e.g., “apple” or “pomiculture”. This sense is dis- pernyms to sets of synonyms (synsets), tinct from the “company” sense of the word “ap- constructs embeddings for these sets, and ple”, which can be denoted as apple3. Thus, more establishes sense-aware relationships be- generally, hypernymy is a relation defined on two tween matching synsets. Evaluation on two sets of disambiguated words; this modeling princi- gold standard datasets for English and Rus- ple was also implemented in WordNet (Fellbaum, sian shows that the method successfully 1998), where hypernymy relations link not words recognizes hypernymy relationships that directly, but instead synsets. This essential prop- cannot be found with standard Hearst pat- erty of hypernymy is however not used or modeled terns and Wiktionary datasets for the re- in the majority of current hypernymy extraction spective languages. approaches. In this paper, we present an approach that addresses this shortcoming.
    [Show full text]
  • Roberto Navigli Curriculum Vitæ
    Roberto Navigli Curriculum Vitæ Roma, 18 marzo 2021 Part I { General Information Full Name Roberto Navigli Citizenship Italian E-mail [email protected] Spoken languages Italian, English, French Sito Web http://wwwusers.di.uniroma1.it/∼navigli Part II { Education Type Year Institution Notes (Degree, Experience, etc.) Ph.D. 2007 Sapienza PhD in Computer Science, advisor: prof. Paola Velardi. Thesis: "Structural Semantic Interconnections: a Knowledge-Based WSD Algorithm, its Evaluation and Applications", Winner of the 2017 AI*IA \Marco Cadoli" national prize for the best PhD thesis in AI. University 2001 Sapienza Master degree in Computer Science, 110/110 cum laude. graduation Thesis: "An automatic algorithm for domain ontology learning", supervisor: prof. Paola Velardi. Part III { Appointments III(A) { Academic Appointments Start End Institution Position 09/2017 oggi Sapienza full professor, Dipartimento di Informatica, Universit`adi Roma \La Sapienza". 12/2010 08/2017 Sapienza associate professor, Dipartimento di Informatica, Universit`adi Roma \La Sapienza". 03/2007 12/2010 Sapienza assistant professor, Dipartimento di Informatica, Sapienza. 05/2003 03/2007 Sapienza research fellow, Dipartimento di Informatica, Sapienza. III(B) { Other Appointments Start End Institution Position 05/2000 04/2003 YH Reply software engineer S.p.A. Roma 1 III(C) { Research visits and stays Start End Institution Position 01/2015 01/2015 Center for Advanced Studies Invited Visiting Fellow (CAS), LMU, Germany 2010 2012 University of Wolverhampton Visiting
    [Show full text]
  • Ontology Learning and Its Application to Automated Terminology Translation
    Natural Language Processing Ontology Learning and Its Application to Automated Terminology Translation Roberto Navigli and Paola Velardi, Università di Roma La Sapienza Aldo Gangemi, Institute of Cognitive Sciences and Technology lthough the IT community widely acknowledges the usefulness of domain ontolo- A gies, especially in relation to the Semantic Web,1,2 we must overcome several bar- The OntoLearn system riers before they become practical and useful tools. Thus far, only a few specific research for automated ontology environments have ontologies. (The “What Is an Ontology?” sidebar on page 24 provides learning extracts a definition and some background.) Many in the and is part of a more general ontology engineering computational-linguistics research community use architecture.4,5 Here, we describe the system and an relevant domain terms WordNet,3 but large-scale IT applications based on experiment in which we used a machine-learned it require heavy customization. tourism ontology to automatically translate multi- from a corpus of text, Thus, a critical issue is ontology construction— word terms from English to Italian. The method can identifying, defining, and entering concept defini- apply to other domains without manual adaptation. relates them to tions. In large, complex application domains, this task can be lengthy, costly, and controversial, because peo- OntoLearn architecture appropriate concepts in ple can have different points of view about the same Figure 1 shows the elements of the architecture. concept. Two main approaches aid large-scale ontol- Using the Ariosto language processor,6 OntoLearn a general-purpose ogy construction. The first one facilitates manual extracts terminology from a corpus of domain text, ontology engineering by providing natural language such as specialized Web sites and warehouses or doc- ontology, and detects processing tools, including editors, consistency uments exchanged among members of a virtual com- checkers, mediators to support shared decisions, and munity.
    [Show full text]
  • Incorporating Trustiness and Collective Synonym/Contrastive Evidence Into Taxonomy Construction
    Incorporating Trustiness and Collective Synonym/Contrastive Evidence into Taxonomy Construction #1 #2 3 Luu Anh Tuan , Jung-jae Kim , Ng See Kiong ∗ #School of Computer Engineering, Nanyang Technological University, Singapore [email protected], [email protected] ∗Institute for Infocomm Research, Agency for Science, Technology and Research, Singapore [email protected] Abstract Previous works on the task of taxonomy con- struction capture information about potential tax- Taxonomy plays an important role in many onomic relations between concepts, rank the can- applications by organizing domain knowl- didate relations based on the captured information, edge into a hierarchy of is-a relations be- and integrate the highly ranked relations into a tax- tween terms. Previous works on the taxo- onomic structure. They utilize such information nomic relation identification from text cor- as hypernym patterns (e.g. A is a B, A such as pora lack in two aspects: 1) They do not B) (Kozareva et al., 2008), syntactic dependency consider the trustiness of individual source (Drumond and Girardi, 2010), definition sentences texts, which is important to filter out in- (Navigli et al., 2011), co-occurrence (Zhu et al., correct relations from unreliable sources. 2013), syntactic contextual similarity (Tuan et al., 2) They also do not consider collective 2014), and sibling relations (Bansal et al., 2014). evidence from synonyms and contrastive terms, where synonyms may provide ad- They, however, lack in the three following as- ditional supports to taxonomic relations, pects: 1) Trustiness: Not all sources are trust- while contrastive terms may contradict worthy (e.g. gossip, forum posts written by them.
    [Show full text]
  • Stefano Faralli Curriculum Vitæ
    Stefano Faralli Curriculum Vitæ PART 1 - General Information Full name Stefano Faralli orcid http://orcid.org/0000-0003-3684-8815 E-mail [email protected] Home page https://www.unitelmasapienza.it/it/contenuti/personale/stefano-faralli Spoken languages Italian (native language), English and French. PART 2 - Education Oct. 2014 PhD in Computer Science, Sapienza University of Rome. May. 2005 BS + MS in Computer Science, Sapienza University of Rome. PART 3 - Appointments Current • 01/11/2017 - Researcher at University of Rome Unitlema Sapienza, Italy Duties included research and teaching. Research fellowships and PhD • 01/06/2015 - 30/10/2017 Post-doctoral position under the supervision of Professor Simone Paolo Ponzetto University of Mannheim, Germany, Institut f¨urInformatik und Wirtschaftsinformatik Duties included research, undergraduate and graduate teaching, student mentoring. • 01/01/2014 - 31/12/2014 Research fellow, under the supervision of Professors: Roberto Navigli and Paola Velardi Sapienza University of Rome, Dip. Informatica • 01/11/2011 - 20/02/2015 PhD student under the supervision of Professors: Roberto Navigli and Paola Velardi Sapienza University of Rome, Dip. Informatica • 01/01/2011 -31/12/2013 Research fellow under the supervision of Professors: Roberto Navigli and Paola Velardi Sapienza University of Rome, Dip. Informatica • 01/04/2008 -31/03/2009 Research fellow under the supervision of Professor: Augusto Celentano Universit`a"Ca' Foscari" Venezia, Dip. Informatica Research collaboration activities and
    [Show full text]
  • Clustering Enterprise Networks by Patent Analysis
    Clustering Enterprise Networks by Patent Analysis Fulvio D´Antonio1, Simone Orsini1, Alessandro Cucchiarelli1, Paola Velardi2 1 Polytechnic University of Marche, Italy, email: cucchiarelli, dantonio, [email protected] 2 Sapienza Università di Roma, Italy, email: dantonio,[email protected] Abstract. The analysis of networks of enterprises can lead to some important insights concerning strategic aspects that can drive the decision making process of different players: business analysts, entrepreneurs, public administrators. In this paper we present the current development status of an integrated methodology to automatically extract enterprise networks from public textual data and analyzing them. We show an application to the enterprises operating in the Italian region of Marche. Keywords: Social Network Analysis, Natural Language Processing, Clustering 1 Introduction Networks of Enterprises [2] are a special kind of social networks in which the nodes represent enterprises and the links indicate some form of relationship among them. The relationships that have been traditionally represented through links are business collaborations, enterprise similarity, mutual exchange of capitals, information flows, or hierarchical relationships like the ones representing supply chains or enterprises aggregation into districts. Social Network Analysis [6] defines a number of measures and techniques that can be used for the evaluation and analysis of enterprise networks. Such measures, if examined by a business analyst, an entrepreneur or a public administrator can lead to some important insights concerning some strategic aspects of the network. We describe here few scenarios in which the analysis can be conveniently applied: Domain analysis The analyst inspects the network in order to understand which are the main productive sectors, the groups of similar enterprises, the relative strengths of such groups and their inter-relationships.
    [Show full text]
  • Word Sense Disambiguation Based on Word Similarity Calculation Using Word Vector Representation from a Knowledge-Based Graph
    Word Sense Disambiguation Based on Word Similarity Calculation Using Word Vector Representation from a Knowledge-based Graph Dongsuk O*1, Sunjae Kwon*2, Kyungsun Kim1 and Youngjoong Ko3 1R&D Center, Diquest Inc., Seoul 08390, Republic of Korea 2School of Elec. & Comp. Engineering, UNIST, Ulsan 44919, Republic of Korea 3Dept. of Computer Engineering, Dong-A University, Busan 49315, Republic of Korea [email protected], [email protected], [email protected], [email protected] Abstract Word sense disambiguation (WSD) is the task to determine the sense of an ambiguous word according to its context. Many existing WSD studies have been using an external knowledge- based unsupervised approach because it has fewer word set constraints than supervised ap- proaches requiring training data. In this paper, we propose a new WSD method to generate the context of an ambiguous word by using similarities between an ambiguous word and words in the input document. In addition, to leverage our WSD method, we further propose a new word similarity calculation method based on the semantic network structure of BabelNet. We evaluate the proposed methods on the SemEval-2013 and SemEval-2015 for English WSD dataset. Ex- perimental results demonstrate that the proposed WSD method significantly improves the base- line WSD method. Furthermore, our WSD system outperforms the state-of-the-art WSD sys- tems in the Semeval-13 dataset. Finally, it has higher performance than the state-of-the-art un- supervised knowledge-based WSD system in the average performance of both datasets. 1 Introduction In natural language text, it is very common for a word to have more than one sense.
    [Show full text]
  • Learning Domain Ontologies from Document Warehouses and Dedicated Web Sites
    Learning Domain Ontologies from Document Warehouses and Dedicated Web Sites Roberto Navigli∗ Paola Velardi Universit`a di Roma “La Sapienza” Universit`a di Roma “La Sapienza” We present a method and a tool, OntoLearn, aimed at the extraction of domain ontologies from Web sites, and more generally from documents shared among the members of virtual organiza- tions. OntoLearn first extracts a domain terminology from available documents. Then, complex domain terms are semantically interpreted and arranged in a hierarchical fashion. Finally, a general-purpose ontology, WordNet, is trimmed and enriched with the detected domain concepts. The major novel aspect of this approach is semantic interpretation, that is, the association of a complex concept with a complex term. This involves finding the appropriate WordNet concept for each word of a terminological string and the appropriate conceptual relations that hold among the concept components. Semantic interpretation is based on a new word sense disambiguation algorithm, called structural semantic interconnections. 1. Introduction The importance of domain ontologies is widely recognized, particularly in relation to the expected advent of the Semantic Web (Berners-Lee 1999). The goal of a domain on- tology is to reduce (or eliminate) the conceptual and terminological confusion among the members of a virtual community of users (for example, tourist operators, commer- cial enterprises, medical practitioners) who need to share electronic documents and information of various kinds. This is achieved by identifying and properly defining a set of relevant concepts that characterize a given application domain. An ontology is therefore a shared understanding of some domain of interest (Uschold and Gruninger 1996).
    [Show full text]
  • What Women Like: a Gendered Analysis of Twitter Users' Interests
    Wikipedia, a Social Pedia: Research Challenges and Opportunities: Papers from the 2015 ICWSM Workshop What Women Like: A Gendered Analysis of Twitter Users’ Interests based on a Twixonomy Stefano Faralli, Giovanni Stilo and Paola Velardi Department of Computer Science Sapienza University of Rome, Italy [email protected] Abstract al., 2010). The majority of these methods infer interests from lexical information in tweets (bigrams, named entities or la- In this paper we analyze the distribution of interests in a large tent topic models), a technique that may fall short in terms population of Twitter users (the full set of 40 million users in of computational complexity when applied to large Twitter 2009 and a sample of about 100 thousand New York users in 2014), as a function of gender. To model interests, we asso- populations, as shown in Stilo and Velardi (2014). ciate ”topical” friends in users’ friendship lists (friends repre- Only few studies investigated the characteristics of Twit- senting an interest rather than a social relation between peers) ter users regardless of specific applications. In Kim et al. with Wikipedia categories. A word-sense disambiguation al- (2010) it is shown that words extracted from Twitter lists gorithm is used for selecting the appropriate wikipage for could represent latent characteristics of the users in the re- each topical friend. Starting from the set of wikipages repre- spective lists. In Kapanipathi et al. (2014) named entities are senting the population’s interests, we extract the sub-graph of extracted from tweets, then, Wikipedia categories, named Wikipedia categories connected to these pages, and we then primitive interests, are associated to each named entity.
    [Show full text]
  • Word Sense Disambiguation: a Survey
    Word Sense Disambiguation: A Survey ROBERTO NAVIGLI Universit`a di Roma La Sapienza Word sense disambiguation (WSD) is the ability to identify the meaning of words in context in a compu- tational manner. WSD is considered an AI-complete problem, that is, a task whose solution is at least as hard as the most difficult problems in artificial intelligence. We introduce the reader to the motivations for solving the ambiguity of words and provide a description of the task. We overview supervised, unsupervised, 10 and knowledge-based approaches. The assessment of WSD systems is discussed in the context of the Sen- seval/Semeval campaigns, aiming at the objective evaluation of systems participating in several different disambiguation tasks. Finally, applications, open problems, and future directions are discussed. Categories and Subject Descriptors: I.2.7 [Artificial Intelligence]: Natural Language Processing—Text analysis; I.2.4 [Artificial Intelligence]: Knowledge Representation Formalisms and Methods General Terms: Algorithms, Experimentation, Measurement, Performance Additional Key Words and Phrases: Word sense disambiguation, word sense discrimination, WSD, lexical semantics, lexical ambiguity, sense annotation, semantic annotation ACM Reference Format: Navigli, R. 2009. Word sense disambiguation: A survey. ACM Comput. Surv. 41, 2, Article 10 (February 2009), 69 pages DOI = 10.1145/1459352.1459355 http://doi.acm.org/10.1145/1459352.1459355 1. INTRODUCTION 1.1. Motivation Human language is ambiguous, so that many words can be interpreted in multiple ways depending on the context in which they occur. For instance, consider the following sentences: (a) I can hear bass sounds. (b) They like grilled bass. The occurrences of the word bass in the two sentences clearly denote different mean- ings: low-frequency tones and a type of fish, respectively.
    [Show full text]
  • Syntactically Aware Neural Architectures for Definition Extraction
    Syntactically Aware Neural Architectures for Definition Extraction Luis Espinosa-Anke and Steven Schockaert School of Computer Science and Informatics Cardiff University fespinosa-ankel,[email protected] Abstract lexicographic information such as definitions con- stitutes the cornerstone of important language re- Automatically identifying definitional knowl- sources for NLP, such as WordNet (Miller et al., edge in text corpora (Definition Extraction or 1990), BabelNet (Navigli and Ponzetto, 2012), DE) is an important task with direct appli- Wikidata (Vrandeciˇ c´ and Krotzsch¨ , 2014) and ba- cations in, among others, Automatic Glos- sary Generation, Taxonomy Learning, Ques- sically any Wikipedia-derived resource. tion Answering and Semantic Search. It is In this context, systems able to address the prob- generally cast as a binary classification prob- lem of Definition Extraction (DE), i.e., identifying lem between definitional and non-definitional definitional information spanning in free text, are sentences. In this paper we present a set of of great value both for computational lexicogra- neural architectures combining Convolutional phy and for NLP. In the early days of DE, rule- and Recurrent Neural Networks, which are based approaches leveraged linguistic cues ob- further enriched by incorporating linguistic in- formation via syntactic dependencies. Our ex- served in definitional data (Rebeyrolle and Tan- perimental results in the task of sentence clas- guy, 2000; Klavans and Muresan, 2001; Malaise´ sification, on two benchmarking DE datasets et al., 2004; Saggion and Gaizauskas, 2004; Stor- (one generic, one domain-specific), show that rer and Wellinghoff, 2006). However, in order these models obtain consistent state of the to deal with problems like language dependence art results.
    [Show full text]
  • What to Write? a Topic Recommender for Journalists
    What to Write? A topic recommender for journalists Giovanni Stilo and Paola Velardi Sapienza University of Rome, Italy stilo, velardi @di.uniroma1.it { } Alessandro Cucchiarelli and Giacomo Marangoni and Christian Morbidoni Universita` Politecnica delle Marche, Italy a.cucchiarelli, g.marangoni, c.morbidoni @univpm.it { } Abstract media events and performances, or for disasters (Lehmann et al., 2012). Assessing the focus, dura- In this paper we present a recommender tion and outcomes of news stories on public atten- 3 system, What To Write and Why (W ), tion is paramount for both public bodies and me- capable of suggesting to a journalist, for dia in order to determine the issues around which a given event, the aspects still uncov- the public opinion forms, and in framing the issues ered in news articles on which the read- (i.e., how they are being considered) (Brooker and ers focus their interest. The basic idea is Schaefer, 2005). Futhermore, real-time analysis to characterize an event according to the of public reaction to news items may provide use- echo it receives in online news sources ful feedback to journalists, such as highlighting as- and associate it with the corresponding pects of a story that needs to be further addressed, readers’ communicative and informative issues that appear to be of interest for the public patterns, detected through the analysis of but have been ignored, or even to help local news- Twitter and Wikipedia, respectively. Our papers echo international press releases. methodology temporally aligns the results The aim of this paper is to present a news me- of this analysis and recommends the con- dia recommender, What to Write and Why (W 3), cepts that emerge as topics of interest from for analyzing the impact of news stories on the Twitter and Wikipedia, either not covered readers, and finding aspects – still uncovered in or poorly covered in the published news news articles – on which the public has focused articles.
    [Show full text]