<> <name> </name> <body> <></> <> <title> </title> <item> </item> <item> 11 </item> </>... </body> </> 1 XML Web XML HTML 1 name item 2 item item HTML

Size: px

Start display at page:

Download "<> <name> </name> <body> <></> <> <title> </title> <item> </item> <item> 11 </item> </>... </body> </> 1 XML Web XML HTML 1 name item 2 item item HTML"

さわとどろき
9 years ago
Views:

1 DEWS2008 C6-4 XML XML XML XML, Abstract Person Retrieval on XML Documents by Coreference that Uses Structural Features Yumi YONEI, Mizuho IWAIHARA, and Masatoshi YOSHIKAWA Department of Social Informatics, Graduate School of Informatics, Kyoto University Yoshidahonmachi, Sakyo-ku, Kyoto, Japan Present retrieval by keywords is based on the occurrence frequency and the occurrence position of the keywords. As for retrieval by two or more keywords, semantic relation between keywords is important. For retrieving information about a person, it is common to search by pair of keywords consisting of the person s name and his/her attribute. However, if semantic relation between keywords is not considered, the documents that describe different person s attribute may be retrieved. By using dependency analysis and coreference analysis, it is possible to retrieve the contents in which query keywords have semantic dependencies and improve search precision. However, it is costly. On the other hand, as for structural documents such as the XML, correspondence is often influenced by the document structure. In this paper, we confirm it by the coreference that uses structural features of XML documents, and we describe our person retrieval that uses the structual coreference. Key words XML, coreference, person retrieval, structural features, 1. XXX XXX

jp XML XML XML, Abstract Person Retrieval on XML Documents by Coreference that Uses Structural Features Yumi YONEI, Mizuho IWAIHARA, and Masatoshi YOSHIKAWA Department of Social Informatics, Graduate

2 <> <name> </name> <body> <></> <> <title> </title> <item> </item> <item> 11 </item> </>... </body> </> 1 XML Web XML HTML 1 name item 2 item item HTML [1] [17] [3] [18] Wikipedia Web HTML XML Web Web XML Wikipedia XML 5 2. Web [14] [8] [14] [8]. Web caption [1] [17] [1] - [17] [3] [18] [3] 3 [18] Web 3 Web Web Web [16] [16] Web Web HTML XML

Wikipedia Web HTML XML Web Web XML 2 3 4 Wikipedia XML 5 2. Web [14] [8] [14] [8].

3 2 (a) name body 3. XML 3. 1 XML (linguistic features) [4] 4 ( ) ( ) ( ) ( ) PER- SON, LOCATION Cabocha [7] ( ) 2 (structural features) XML HTML XML 照応詞 title p title item title title item (b) 照応詞 title p 2 (a) (b) name body name body 照応詞照応詞 item item (c) name body 照応詞 item item (d) 照応詞 name body // 3 k- (k=2) XML 2 2(a) XML title XML 2(b) XML

1 1 (linguistic features) [4] 4 ( ) ( ) ( ) ( ) PER- SON, LOCATION Cabocha [7] ( ) 2

4 k- k 3 k=2 k- 3(a),(b),(c) name body 3(d) 2 k- (k=2) k- k- ( (name )(body )) 3(d) k Support Vector Machine 1 x, y C(x, y) p(x, y) [6] [9] (x, y) 1,0 x ( (name )(body )) k- y { 1 ifx y f i (x, y) = 0 otherwise 1 (1) p(x, y) = 1 z(x) e λ if i (x,y) i z(x) = e λ if i (x,y) i y λ z(x). P (f i) = p(x, y)f i(x, y) (3) x,y (1) (2) P (f i ) = p(x)p(x, y)f i (x, y) (4) x,y P (f i) = P (f i) (5) P (f i ) P (f i ) (6) H(P ) = p(x, y)logp(x, y) (6) x,y λ 2 SVM SVM [2] 2 SVM [5] T 1, T 2 V 1, V 2 E 1, E 2 T 1 = (V 1, E 1), T 2 = (V 2, E 2) K(T 1, T 2 )= K S (s 1, s 2 ) (7) v 1 V 1 v 2 V 2 s 1 S v1 (T 1 ) s 2 S v2 (T 2 ) K S (s 1, s 2 ) = I(s 1 = s 2 ) (8) S v(t ) v V K S (2) I() 1 0 s 1 = s 2 2 [12] RNA HTML XML Web SVM

5 SVM 1 0 k (2) SVM 3. 2 ( (name )(body )) body name body XML XML Wikipedia 1 XML 4 1: 2: 3: (item ) 2 (p ) 3 (item (normalist )) 4 ( (name )(body )) 5 ( (title )( )) 6 (body (p )( )) 7 (body ( )( )) 8 (normalist (item )(item )) 9 ( (normalist ) (normalist )) 10 ((p ) (p )) 11 ( ( )( )) 4 k- (k=2) 4: [4] [11] Cabocha [7] EDR [13] EDR

org/ 1 1 2 3 4 333 521 6 72 75 96 4 4 12597 22758 18 33 240 489 7 14 12260 22269 11 19 1 (item ) 2 (p ) 3 (item (normalist

6 2 k- (k=2) (1) (item ) 71 (2) (p ) 40 (3) (item (normalist )) 4 (4) ( (name )(body )) 93 (5) ( (title )( )) 3 (6) (body (p )( )) 29 (7) (body ( )( )) 0 (8) (normalist (item )(item )) 0 (9) ( (normalist )(normalist )) 0 (10) ((p ) (p )) 0 (11) ( ( )( )) k- (k=2) (2) (p ) 7 (10) ((p ) (p )) 0 7 Wikipedia 2 3 k=2 k (1) (2) item (3) item (4) name (5) title (6) (6) (10) 2 3 XML (3) (4) (5) name title 2 (4) name item (body (p )( ) SVM [10] SVM SV M light [15] SVM (precision) (recall) precision = recall = F (F-measure) F F -measure = precision recall precision + recall k- k- SVM XML (I) (II) (III) 4, 3.1.1(1) k=2 k- 5 5 (I) (II)

(2) item (3) item (4) name (5) title (6) (6) (10) 2 3 XML (3) (4) (5) name title 2 (4) name item (body (p )( ) 2 3 4. 1. 3 SVM [10] SVM SV M light [15] SVM 0 1 4.

7 4 (I) (II) (III) % 76.0% 51.3 % 31.2% (I) 40.8 % 66.8% 57.5 % 48.7% F 52.7 % 71.1% 54.2 % 38.0% (II) 77.0% 78.9% 69.3% 75.0% 48.1% 69.2% 91.7% 54.8% F 59.2% 73.7% 74.9% 63.3% (III) 90.6% 92.0% 82.0% 86.0% 6 k % 78.9% 69.3% 75.0% k=2 48.1% 69.2% 91.7% 54.8% F 59.2% 73.7% 74.9% 63.3% 75.4% 77.6% 69.3% 75.0% k=3 46.2% 68.6% 91.7% 54.8% F 57.3% 72.8% 74.9% 63.3% 72.1% 78.0% 69.3% 75.0% k= 49.3% 72.0% 91.7% 54.8% F 58.6% 74.9% 74.9% 63.3% % 78.0% 69.3% 75.0% 49.3% 72.0% 91.7% 54.8% F 58.6% 74.9% 74.9% 63.3% 97.4% 85.3% 63.3% 88.0% SVM 30.6% 61.2% 83.3% 30.4% F 46.6% 71.3% 719% 45.2% 38.9% 33.5% 62.0% 54.8% F 54.4% 49.1% 70.6% 66.7% (k= ) F XML (III) (II) (III) k=2 k=3 k- k=2 ( = ) 3 F 1 k=2 2 k= k=2 k [4] SVM k (k=2 - ) 3 (k=3 - ) SVM F 3 SVM

0% 69.3% 75.0% k= 49.3% 72.0% 91.7% 54.8% F 58.6% 74.9% 74.9% 63.3% 7 1 2 3 4 72.1% 78.0% 69.3% 75.0% 49.3% 72.0% 91.7% 54.8% F 58.6% 74.9% 74.9% 63.3% 97.4% 85.3% 63.3% 88.0% SVM 30.6% 61.2% 83.

8 5. XML k- XML Wikipedia XML 4 F k- k- =2 k- SVM F Wikipedia HTML (B)( ), [4],,,,, Vol 46, No [5] Vol.21, No.1,a, [6] Andrew Kehler, Probabilistic Coreference in Information Extraction,CoRR, cmp-lg/ ,1997. [7], Support Vector Machine Chunk,, Vol. 9, No. 5, pp [8] 11, 2005 [9] Adam L. Berger, Stephen A. Della Pietra, Vincent J. Della Pietra A Maximum Entropy Approach to Natural Language Processing, Computational Linguistics, [10] Zhang Le Maximum Entropy Modeling Toolkid for Python and C++, toolkit. html. [11],,,,, version 2.3.3, [12] Alessandro Moschitti, Making Tree Kernels proctical for Natural Language Learning,EACL, [13] EDR. Technical Report TR 045, [14] 2002 pp [15] SV M light moschitt/tree-kernel.htm. [16] Lan Yi,Bing Liu,Xiaoli Li, Eliminating noisy information in web pages for data mining, Conference on Knowledge Discovery in Data Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining, pp , [17] Minoru Yoshida, Kentaro Torisawa, Junichi Tsujii, Extracting ontologies from World Wide Web via HTML tables, Pacific Association for Computational Linguistics, pp ,2001. [18] Web DEWS 6-p [1] Hsin-Hsi Chen, Shih-Chung Tsai, Jin-He Tsai Mining Tables from Large Scale HTML Texts, 18th International Conference Computational Linguistics, pp [2] Nello Cristianini, John Shawe-Taylor,, [3] WWW HTML, DE

Berger, Stephen A. Della Pietra, Vincent J. Della Pietra A Maximum Entropy Approach to Natural Language Processing, Computational Linguistics, 22 1996.

2 : Open Clip Art Library [4] 2 3 4 5 6 2. 2 2. 1 Microsoft Office PowerPoint Web PowerPoint 2 Yahoo! Web [5] SlideShare 2. 1. 1 Yahoo! Web Yahoo! Web

2 : Open Clip Art Library [4] 2 3 4 5 6 2. 2 2. 1 Microsoft Office PowerPoint Web PowerPoint 2 Yahoo! Web [5] SlideShare 2. 1. 1 Yahoo! Web Yahoo! Web DEWS2008 E4-4 606-8501 E-mail: {hsato,oyama,tanaka}@dl.kuis.kyoto-u.ac.jp.. Supporting the Selection of Images Based on Referential Semantics from Surrounding Information of the Image in Presentation Files