[7] [10] Web Web RDF Resource Description Framework subjectpredicate object Web Web Web Web Web 2 Web Web MUC(Message Understanding Confere

Size: px

Start display at page:

Download "[7] [10] Web Web RDF Resource Description Framework subjectpredicate object Web Web Web Web Web 2 Web 3 4 5 6 2. Web MUC(Message Understanding Confere"

ゆりかつくとの
7 years ago
Views:

1 THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS TECHNICAL REPORT OF IEICE. Web {tjstkm,jmori,ishizuka}@mi.ci.i.u-tokyo.ac.jp Web Web Web Web Web Web Web Extracting Relationships among Named Entities from the Web Tsujishita TAKUMI, Junichiro MORI, and Mitsuru ISHIZUKA Faculty of Engineering, University of Tokyo Hongo 7 3 1, Bunkyo-ku, Tokyo, Japan Graduate School of Information Science and Technology, University of Tokyo Hongo 7 3 1, Bunkyo-ku, Tokyo, Japan {tjstkm,jmori,ishizuka}@mi.ci.i.u-tokyo.ac.jp Abstract With the currently huge amount of information on the Web, Web mining methods that obtain useful information and structures from the Web have been gained interest. We propose a novel Web mining method that automatically extracts relational information among named entities from the Web. The basic idea is to cluster similar pairs of named entities based on their contextual similarity on a Web document. Relational information among named entityes is obtained from the result of clustering process. Our experiments conducting on entity pairs of politicians and places achives clustering of the entity pairs with high recall and precision, and find appropriate relational information among the entities. Key words Information Extraction, Web Mining, Search engine, Entity pair model, Clustering 1. Web Web Web Web Web Web Web Web Web Web Web Web [1] [4] Web Web Web [5]. Web [6] Web 1

1, Bunkyo-ku, Tokyo, 113 8656 Japan Graduate School of Information Science and Technology, University of Tokyo Hongo 7 3 1, Bunkyo-ku, Tokyo, 113 8656 Japan E-mail: {tjstkm,jmori,ishizuka}@mi.ci.i.u-tokyo.

2 [7] [10] Web Web RDF Resource Description Framework subjectpredicate object Web Web Web Web Web 2 Web Web MUC(Message Understanding Conference) Template Relation Task ACE(Automatic Content Extraction) meetings Relation Detection and Characterization 1 1 ACE Person, Organization, Fa- Part-Whole ACE located, near, part-whole business, family executive, staff...person GPE Web (PER- SON) (GRE) Web Web 1 AND AND AND AND 4 bag of words tfidf AND AND AND AND AND AND Web Web [11]. cility, Location, GPE, Vehicle, Weapon 2 GPE Geo political entity 2

Characterization 1 1 ACE Person, Organization, Fa- Part-Whole ACE located, near, part-whole business, family executive, staff...person GPE 2 2.

3 1 AND AND AND AND tfidf Table 1 Keyword list obtained from the search results with a search query: Junichiro Koizumi AND JAPAN, Yoshiro Mori AND JAPAN, Junichiro Mori AND Kanagawa, and Yoshiro Mori AND Ishikawa AND AND tfidf,, AND AND Web 2. 2 Web (PERSON) (ORGANIZATION) (PERSON) (GPE) Web Web 2. 3 IREX 8 [12] Web [1] [4] 2. 4 Web Raghavan entity language [13] 3

Yoshiro Mori AND Ishikawa AND AND tfidf,, AND AND Web 2.

4 Fig. 1 小ブッシュ-アメリカ <PER>-<GPE> 森泉善純朗一 - 郎日 - 日本本エンティティペア集合各ペアをWeb 検索検エンティティペアのコンテクストベクトル重索要結語果抽出のページから P1: P2: c( c(ブッシュ,アメリカ) 小泉純一郎, 日本 ) { 政大治統, 領政, 権政, 権首,イラク} 相, }. P3: c( 森善朗, 日本 ) = {ラグビー, 総理, 会長, } コンテクストベクトル間の C2 C1 類似度に基づいてクラスタリング P1 P3 C3 ラベルの抽出 C2: C1: エンティティペアのクラスタ関... 大首統相領係情報 1 Web Extraction of relational information among named entities from the Web AND Web Web Web e1 e2 t tfidf (Term Frequency-Inverse Document Frequency) tfidf(t) = tf(t) idf(t) tf(t) e1 e2 t idf(t) Web Web t e1,e2 C(e1, e2) C(e1, e2) = {t 1, t 2,..., t k,...} t k tfidf(t k ) 2. 5 C i cos(c i, C j ) = C ic j C i C j tfidf tf 3. Web (PERSON) (GPE) Web Web Web 2 n m n m n 30 m 10 3 google 4

.. 大首統相領係情報 1 Web Extraction of relational information among named entities from the Web AND Web Web Web e1 e2 t tfidf (Term Frequency-Inverse Document Frequency) tfidf(t) = tf(t)

5 2 Table 2 Relational Information obtained from a cluster of entity pairs () Table 3 3 Clustering performance in relation to the number of POS in a contextual vector Precision Recall F window size between entities window size at sides of entities Precision Recall cl N correct,cl N incorrect,cl r N correct,r r N true,r Precision(P ) Recall(R) N correct,cl N correct,r P = Σ cl, R = Σ r N correctmcl + N incorrect,cl N true,r P R F 2 n m F n 30 m 10 F 3 Precision Recall 99% F 2 averaged F measure window size (the number of POS) 2 F Fig. 2 The number of POS in a contextual vector vs. F measure of Clustering results 4 Table 4 Clustering performance in relation to scoring methods of a contextual vector Precision Recall F tfidf Table 5 Clustering performance in relation to similarity measures between contextual vectors Precision Recall F cosine tfidf 5

6 tfidf [10] 4 tfidf (n m 10,5 ) tfidf 5 (n m 10,5 tfidf ) [13] Web 6. [14], [15] Hasegawa [16] Web Web Web Web Web Web [1], [3], [5], [7] 7. Web Web Web [1] P. Cimiano, G. Ladwig, and S. Staab, Gimme the context: Context-driven automatic semantic annotation with cpankow, Proc. of the 14th World Wide Web Conference, [2] P. Cimiano, S.Handschuh, and S. Staab, Towards the selfannotating web, Proc. of the 13th World Wide Web Conference, [3] O. Etzioni, M. Cafarella, D. Downey, S. Kok, A. Popescu, T. Shaked, S. Soderland, D. Weld, and A. Yates, Webscale information extraction in knowitall(preliminary results, Proc. of the 13th World Wide Web Conference, pp , [4] O. Etzioni, M. Cafarella, D. Downey, A. Popescu, T. Shaked, S. Soderland, D. Weld, and A. Yates, Methods for domain-independent information extraction from the web: An experimental comparison, Proc. of the AAAI Conference, [5] Web vol.20 no.1 pp [6] Web, SIG-KBS [7] P. Mika, Flink:semantic web technology for the extraction and analysis of social networks, Journal of Web Semantics, vol.3, no.2, [8] A. Culotta, R. Bekkerman, and A. McCallum, Extracting social networks and contact information from and the web, Proc. of CEAS, [9] Web key person, DBS-130/FI [10] Web vol.20 no.5 pp [11] G.A. Miller, and W.G. Charles, Contextual correlates of semantic similarity, Language and Cognitive Processes, vol.6, no.1, pp.1-28, [12] vol.43 no.6 pp [13] H. Raghavan, J. Allan, and A. McCallum, An exploration of entity models, collective classification and relation description, Proc. of LinkKDD, [14] E. Agichtein, and L. Gravano, Extracting relations from large plain-text collections, Proc. of the 5th ACM International Conference on Digital Libraries (ACMDL00), pp.85-94, [15] D. Zelenko, C. Aone, and A. Richardella, Kernel methods for relation extraction, Proc. of the Conference on Empirical Methods in Natural Language Processing, pp.71-78, [16] T. Hasegawa, S. Sekine, and R. Grishman, Discovering relations among named entities from large corpora, Proc. of the Annual Meeting of Association of Computational Linguistics (ACL 04),

Staab, Towards the selfannotating web, Proc. of the 13th World Wide Web Conference, 2004. [3] O. Etzioni, M. Cafarella, D. Downey, S. Kok, A. Popescu, T. Shaked, S. Soderland, D. Weld, and A.

1., 1 COOKPAD 2, Web.,,,,,,.,, [1]., 5.,, [2].,,.,.,, 5, [3].,,,.,, [4], 33,.,,.,,.. 2.,, 3.., 4., 5., ,. 1.,,., 2.,. 1,,

1., 1 COOKPAD 2, Web.,,,,,,.,, [1]., 5.,, [2].,,.,.,, 5, [3].,,,.,, [4], 33,.,,.,,.. 2.,, 3.., 4., 5., ,. 1.,,., 2.,. 1,, THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS TECHNICAL REPORT OF IEICE.,, 464 8601 470 0393 101 464 8601 E-mail: matsunagah@murase.m.is.nagoya-u.ac.jp, {ide,murase,hirayama}@is.nagoya-u.ac.jp,