SIG-AM : 1: ( 2 I) ( 2 A) ( 2 B) ( 2 C) ( 2 A) [4] 5 ( ) ( 2 B) ( 2-➀) ( 2-➁) ( 3-➀) ( 3-➅) - 2 -

Size: px

Start display at page:

Download "SIG-AM : 1: ( 2 I) ( 2 A) ( 2 B) ( 2 C) ( 2 A) [4] 5 ( ) ( 2 B) ( 2-➀) ( 2-➁) ( 3-➀) ( 3-➅) - 2 -"

あきたけじゅふく
5 years ago
Views:

1 SIG-AM : Visualization System for Exploratory Analysis of Time-series Data: Alignment Method between Articles and Time-series Data Shun Naito 1 Ryoki Furuta 2 Mitsunori Matsushita Graduate School of Informatics, Kansai University 2 2 Suken Shuppan 3 3 Faculty of Informatics, Kansai University Abstract: The goal of our study is to support a user s analysis of time-series data in an exploratory manner. Such exploratory analysis requires repeated access to various types of information related to the user s interests such as texts and numerical data. To support such the user s analysis, we have proposed a system that visualizes temporal changes in time-series data and presents the causes of those changes with the data. In this paper, we improve the system by adding an alignment function between news articles and time-series data. By using this function, the user can find articles that relates to the time-series data easily. 1 [1] [2][3] DB DB DB WEB DB DB DB mat@res.kutc.kansai-u.ac.jp - 1 -

2 SIG-AM : 1: ( 2 I) ( 2 A) ( 2 B) ( 2 C) ( 2 A) [4] 5 ( ) ( 2 B) ( 2-➀) ( 2-➁) ( 3-➀) ( 3-➅) - 2 -

3 SIG-AM : ( 3-➁) ( 3-➃) ( 3-➂ 3-➄) 3 ( 2-➁) [3] 3 ( 3-➂) ( 3-➄) csv ( 18 ) csv / / (42)

4 SIG-AM [5] [6] Ahmad Boyd Wavelet [7][8] [9] i.e., 1 7 i.e., i.e., e.g., (e.g., (α) (β) (α) 4-A (β) 4-B (β) - 4 -

SIG-AM-13-01 4: (1) (2) d A, d B (d A < d B ) [d A, d B ] C f C d C C f [d A, d B ] d f(d) [d A, d B ] f(d A ) f(d B ) f(d) < f(d + 1)( d A d < d B ) C f C d M s (d) M e (d) 5 C d M s (d A ) M e (d B

5 SIG-AM : (1) (2) d A, d B (d A < d B ) [d A, d B ] C f C d C C f [d A, d B ] d f(d) [d A, d B ] f(d A ) f(d B ) f(d) < f(d + 1)( d A d < d B ) C f C d M s (d) M e (d) 5 C d M s (d A ) M e (d B ) C = C f C d [d A, d B ] 4.3 HP(http: // / WEB A 11 B : 1: A 1 11/1-11/ /2-11/ /4-11/ /1-11/ /2-11/ A 5 C

6 SIG-AM : Web 1 ( 15H02780) [1],,, :, (2005) [2] Naito, S., Matsushita, M.: Supporting Consecutive Data Exploration by Visualizing Spatiotemporal Trend Information, in Proceedings of the 2015 Conference on Technologies and Applications of Artificial Intelligence, pp (2015) [3], :, ARG 6 Web, No.6, pp (2015) [4] 21 3H8-3 (2007) [5] :, Vol.12, No.3, pp (2000) [6],, : :,, Vol. 48, No. 3, pp (2007) [7] Saif Ahmad, Paulo C F de Oliveira, Khurshid Ahmad: Summarization of Multimodal Information, Proc. 4th International conference on Language Resources and Evaluation, pp (2004) [8] Sarah Boyd: TREND: A System for Generating Intelligent Descriptions of Time-Series Data, Proc. IEEE International Conference on Intelligent Processing Systems (1998) [9],,, :, 22, pp (2006) - 6 -

7 SIG-AM Recommender System Using Personal Values-based User Modeling from Browsing History of Customer Reviews 1 1 Yasufumi Takama 1 Suzuto Shimizu Graduate School of System Design, Tokyo Metropolitan University Abstract: This paper proposes a method for generating a user model reflecting user s personal values from user s browsing histories of customer reviews. This paper also proposes a recommendation method using the personal values-based user model. Existing recommendation methods such as collaborative filtering and content-based filtering tend to be less accurate for new users and items due to the lack of information about them. The personal values-based recommender system is expected to realize more precise recommendations for new users. As a customer review contains reviewer s evaluation of an item and its attributes, the proposed method estimates attributes on which a target user put high priority when evaluating items from customer reviews the user refers to for his/her decision making. This paper examines the effectiveness of the proposed method with user experiments. 1 Web [11] [5] ytakama@tmu.ac.jp [2, 7] [2] [3] cold-start [2] - 7 -

8 SIG-AM [12] Web Amazon TOP [9] [8] cold-start[11] sparsity 2.2 Rokeach Value Survey[10] Big Five[1] [4] Wu [13] rate matching rate, RMRate [2] u i x j p ij a k p k ij u i a k RMR ik 1 I i u i δ(x, y) x y 1 0 RMR ik = x j I i δ(p ij, p k ij ) I i (1). [2] - 8 -

SIG-AM-13-02 CF [6] CF CF [6] 3 3.1 [2] 1:. 1 1 1 1.

9 SIG-AM CF [6] CF CF [6] [2] 1: (1) (2) (3) 5 [12] : - 9 -

10 SIG-AM 致率更新終了条件? 終了 2: [2] (1) 2 r Score(r, u i ) = k { ek r e r } RMR 2 ik N r log n r (2) u i r a k e k r r e r n r N r [12] u i x j 3 x j c a k e k c x j a k e k j u i Score(x j, u i, c) = A i = {e k j e k c } RMRik 2 (3) a k A i { a a k RMR ik l A RMR } il (4) A travel

11 SIG-AM travel 2: A B A B 1: A B A B A 5 3, 4 ID B

12 SIG-AM : A ID (10) (7) (12) (15) (10) (11) (8) (6) (15) (17) (8) (10) (8) (8) (12) (17) (7) (10) (10) (10) (12) (17) (7) (7) (15) (4) (14) (15) (11) (4) 4: A ID (6) (2) (10) (11) (11) (15) (7) (6) (11) (14) (12) (13) (2) (8) (9) (14) (12) (15) (10) (8) (7) (11) (14) (18) (6) (7) (11) (14) (13) (16) : / / A 5/4/11 2/6/14 B 4/2/10 7/4/7 A 1/6/12 4/3/9 B 5/4/5 8/5/2 15/16/38 21/18/32 6: A B A B

13 SIG-AM : : JSPS JP16K12535 [1] P. T. Costa, R. R. McCrae: Revised NEO Personality Inventory (NEO-PI-R) and NEO Five- Factor Inventory (NEO-FFI), Psychological Assessment Resources, [2] S. Hattori, Y. Takama: Recommender system employing personal-value-based user model, Journal of Advanced Computational Intelligence and Intelligent Informatics, Vol. 18, No. 2, pp , [7] M. A. S. N. Nunes, R. Hu: Personality-based Recommender Systems: an Overview, Rec- Sys 12, pp. 5 6, [8] F. Pachet, P. Roy, D. Cazaly: A combinatorial approach to content-based music selection, IEEE International Conference on Multimedia Computing and Systems, pp , [9] P. Resnick, N. Iacovou, M. Suchak, P. Bergstrom, J. Riedl: GroupLens: an open architecture for collaborative filtering of netnews, 1994 Conference on Computer Supported Cooperative Work, pp , [10] M. Rokeach: The Nature of Human Values, New York: The Free Press, [11] A. I. Schein, A. Popescul, L. H. Ungar, D. M. Pennock: Methods and metrics for cold-start recommendations, Special Interest Group on Information Retrieval, pp , [12],, :, 28, 3B4-OS-10b-3, [13] W. Wu, L. Chen, L. He: Using Personality to Adjust Diversity in Recommender Systems, 24th ACM Conf. on Hypertext and Social Media, pp , [3], :, 97, No. 14, [4],, :,, vol. WI , pp , [5] : (1),, Vol. 22, No. 6, pp , [6],, :, 28 (JSAI2014), 1H4-NFC-01a-5,

14 SIG-AM Recent Research Trends and Proposal of Distributional Word Representations for Hypernymy Detection 1 Koki Washio Graduate School of Arts and Sciences, The University of Tokyo Abstract: Distributional representation for words is an important model of word senses which reflects several types of semantic relations, not only similarity relations. In this paper, we provide an overview of unsupervised/supervised approaches with word distributional representations to detect hypernym-hyponym relations and a recently reported problem associated with these approaches. Moreover, we propose a future research direction to solve this problem and prove the validity of this direction with small experiments. 1 [22][7] kkwashio3333@gmail.com [9] w V W c V C w w c c w c (w, c) D f

15 SIG-AM V W V C M w M w M 0 w w c M w c M PMI(pointwise mutual information, ) P MI(w, c) = log 2 P (w, c) P (w)p (c) = log 2 f(w, c) D f(w)f(c) PMI w c w c P MI(w, c) = PPMI(positive pointwise mutual information, ) P P MI(w, c) = { 0 (P MI(w, c) 0) P MI(w, c) (P MI(w, c) > 0) PPMI w c [12] M V W V C M V W d d c 2.2 d [14] [15] SkipGram SGNS SGNS SGNS (w, c) d w 1 0) V W one-hot c 1 V C one-hot V W d w w d V C c c 1 SGNS (w, c) D P (D = 1 w, c) = exp( w c) arg max w, c (w,c) D = arg max w, c P (D = 1 w, c) (w,c) D log exp( w c) (w, c) P (D = 1 w, c) = 1 (w, c) D k (w, c 1 ),..., (w, c k )c j P (c) 3/4 D arg max w, c (w,c) D = arg max w, c P (D = 1 w, c) P (D = 0 w, c) (w,c) D 1 log 1 + exp( w c) (w,c) D k (w,c) P (c) 3 4 D log exp( w c) d 1 w one-hot d w

16 SIG-AM vs. [2] Levy [11][10][13] [11] SGNS word2vec PMI SGNS w c w c = ( ) f(w, c) D log log k f(w) f(c) = P MI(w, c) log k SGNS W C W i C j = P MI(w i, c j ) log k M PMI SGNS PMI log k SGNS 2 [14][15][16] [10] 2 word2vec CBoW GloVe [20] [13] SGNS Levy Levy SGNS 3 2 [12]

17 SIG-AM (Distributional generality)[22] (Distributional inclusion hypotheses)[5] w 1 w 2 w i = (w i1,..., w in ) F 0 W eeds [22] i F (w W eedsp (w 1, w 2 ) = 1) F (w w 2) 1i W eedsr(w 1, w 2 ) = i F (w 1) w 1i i F (w 1) F (w 2) w 2i i F (w 2) w 2i precision recall w 1 w 2 W eedsp 1 W eedsr 0 1 W eedsp W eedsp W eedsr Weeds Clarke[3] invcl[8] i F (w ClarkeP (w 1, w 2 ) = 1) F (w min(w 2) 1i, w 2i ) ClarkeR(w 1, w 2 ) = i F (w 1) w 1i i F (w 1) F (w 2) min(w 1i, w 2i ) i F (w 2) w 2i invcl(w 1, w 2 ) = ClarkeP (w 1, w 2 )(1 ClarkeR(w 1, w 2 )) Clarke W eeds ClarkeP ClarkeP W eedsr invcl W eeds Clarke invcl invcl [21] i F (w simdiff(w 1, w 2 ) = 1) F (w min(w 2) 1i, w 2i ) i F (w 1) F (w max(w 2) 1i, w 2i ) i F (w 2) F (w w 1) 2i i F (w 1) F (w w 2) 1i i F (w 1) F (w max(w 2) 1i, w 2i ) Jaccard 1 2 w 1 w 2 AP(Average Precision) w 1 w 2 balap inc [7] Erk [4] BNC(British National Corpus) 5000 PPMI WordNet3.0 W eeds Clarke precision recall 1: W eeds Clarke invcl simdif f

18 SIG-AM [19] SLQS SLQS N c = (c 1,..., c n ) H(c) = n p(c i c) log 2 (p(c i c)) i=1 p(c i c) c MinMax 0 1 H n (c) w i E wi = Me N j=1(h n (c j )) Me SLQS SLQS(w 1, w 2 ) = 1 E w 1 E w2 SLQS(w 1, w 2 ) > 0 w 1 w SVM [18][23] [18] BNC 1 SGNS 3 BLESS[1] 200 BLESS SGNS F AP(Average Precision) [12] match error recall matcherror = recall 3 Omer Levy hyperwords

19 SIG-AM [12] 3.2 2: F AP [12] BNC 1 PPMI 3.2 BLESS cos W eedsp/r, ClarkeP/R, invcl 50 : SLQS cos W eedsp/r ClarkeP/R invcl 50 SLQS

20 SIG-AM : F AP : cos (baseline) F AP , 4 baseline cos SGNS SGNS SGNS+ SGNS 5: SGNS SGNS F AP F AP 5 [1] Baroni, M., Lenci, A.: How we BLESSed distributional semantic evaluation., In Proc. of the ofthe GEMS 2011Workshop on GEometrical Models of Natural Language Semantics, pp (2011) [2] Baroni, M., Dinu, G., Kruszewski, G.: Dont count, predict! a systematic comparison of context-counting vs. context-predicting semantic vectors., In Proc. of the Annual Meeting of the Association for Computational Linguistics(ACL), Vol. 2, long paper, pp (2014) [3] Clarke, D.: Context-theoretic semantics for natural language: an overview., In Proc. of the EACL 2009Workshop on GEMS: GEometrical Models of Natural Language Semantics, pp (2009) [4] Supporting inferences in semantic space: representing words as regions., In Proc. of the International Conference on Computational Semantics(ICCS), pp (2009) [5] Geffet, M., Dagan, I.: The distributional inclusion hypotheses and lexical entailment., In Proc. of the Annual Meeting of the Association for Computational Linguistics(ACL), pp (2005)

21 SIG-AM [6] Goldberg, Y., Levy, O.: word2vec explained: deriving mikolov et al. s negativesampling., arxiv preprint, arxiv: (2014) [7] Kotlerman, L., Dagan, I., Szpektor, I., Geffet, M., Directional Distributional Similarity for Lexical Inference., Natural Language Engineering, Vol. 16(4), pp (2010) [8] Lenci, A., Benotto, G.: Identifying hypernyms in distributional semantic space., In SEM 2012 The First Joint Conference on Lexical and Computational Semantics, Vol. 2, pp (2012) [9] Levy, O., Goldberg, Y.: Dependencybased word embeddings., In Proc. of the Annual Meeting of the Association for Computational Linguistics(ACL), Vol. 2, Short Paper (2014) [10] Levy, O., Goldberg, Y.: Linguistic regularities in sparse and explicit word representations., In Proc. of the Conference on Computational Natural Language Learning., pp (2014) [11] Levy, O., Goldberg, Y.: Neural word embeddings as implicit matrix factorization., In Advances in Neural Information Processing Systems 27: Annual Conference on Neural Information Processing Systems 2014, pp (2014) [12] Levy, O., Remus, S., Biemann, C., Dagan, I.: Do Supervised Distributional Methods Really Learn Lexical Inference Relations?, In Proc. of the 2015 North American Chapter of the Association for Computational Linguistics(NAACL): Human Language Technologies, pp (2015) [13] Levy, O., Goldberg, Y., Dagan, I. Ramat-Gan, I.: Improving distributional similarity with lessons learned from word embeddings., Transactions of the Association for Computational Linguistics, 3 (2015) [14] Mikolov, T., Chen, K., Corrado, G. S., Dean, J.: Efficient estimation of word representations in vector space., CoRR, abs/ (2013) [15] Mikolov, T., Sutskever, I., Chen, K. Corrado, G. S., Dean, J.: Distributed representations of words and phrases and their compositionality., In Advances in neural Information Processing Systems, pp (2013) [16] Mikolov, T., Yih, W., Zweig, G.: Linguistic regularities in continuous space word representations., In Proc. the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp (2013) [17] Pennington, J., Socher, R., Manning, C.: Glove: Global vectors for word representation., In Proc. of the Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, Vol. 2, Short Paper (2015) [18] Roller, S., Erk, K., Boleda, G.: Inclusive yet selective: Supervised distriubtional hypernymy detection., In Proc. of the International Conference on Computational Linguistics(COLING), pp (2014) [19] Santus, E., Lenci, A., Lu, Q., Walde, S.: Chasing hypernyms in vector spaces with entropy., In Proc. ofn the Conference of the European Chapter of the Association for Computational Linguistics, pp (2014) [20] Suzuki, J., Nagata, M.: A Unified Learning Framework of Skip-Grams and Global Vectors., In Proc. of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (ACL/IJCNLP), Vol. 2, Short Paper (2015) [21],, (2015) [22] Weeds, J., Weir, D., McCarthy, D.: Characterising measures of lexical distributional similarity., In Proc. of the International Conference on Computational Linguistics(COLING), pp (2004) [23] Weeds, J., Clarke, D., Reffin, J., Weir, D., Keller, Bill.: Learning to distinguish hypernyms and cohyponyms., In Proc. of the International Conference on Computational Linguistics(COLING): Technical Papers, pp (2014)

Computational Semantics 1 category specificity Warrington (1975); Warrington & Shallice (1979, 1984) 2 basic level superiority 3 super-ordinate catego

Computational Semantics 1 category specificity Warrington (1975); Warrington & Shallice (1979, 1984) 2 basic level superiority 3 super-ordinate category preservation 1 / 13 analogy by vector space Figure