WII-A 2017 Web SNS Tweet 2. [ 02] [ 10, 14, 07, 12, 12]. Matsumoto et al. [Matsumoto 11] [ 15] / Support Vector Machine 2 / [ 01, 16] [ 01] [ 1

Size: px

Start display at page:

Download "WII-A 2017 Web SNS Tweet 2. [ 02] [ 10, 14, 07, 12, 12]. Matsumoto et al. [Matsumoto 11] [ 15] / Support Vector Machine 2 / [ 01, 16] [ 01] [ 1"

ひでかてらわ
5 years ago
Views:

1 1 Web 2017 Conversion of Japanese Slang into Standard Japanese Considering Sensibility Kazuyuki Matsumoto Seiji Tsuchiya Misako Imono Minoru Yoshida Kenji Kita Tokushima University Doushisha University Daido University Tokushima University keywords: Internet slang, standard word, sensibility, concept of word Summary With the recent spread of communication using social media, exchanging opinions each other on web has become more common irrespective of age and sex. On the other hand, a problem called as Internet flaming often occurs along with the increase of social network service users. One of the reasons might be that the users do not recognize meanings/intentions/emotions expressed by other users words. In this study, we focused on slangs (Internet slangs) that are often used on SNS but are not registered in dictionaries, then tried to convert them into standard words. We also intended to output more appropriate candidates by considering not only semantic similarity but also affective similarity. The proposed method conducts filtering and re-ranking over the semantically similar candidates obtained based on distributed representations to detect the inappropriate candidates as standard word by focusing on two points: (1) features of slang/standard word and (2) affective similarity between the inputted word and the candidate words. In the evaluation experiment, the proposed method obtained a higher MRR than the baseline method. 1. Web Web Web [ 13, 14]. etc.

2 WII-A 2017 Web SNS Tweet 2. [ 02] [ 10, 14, 07, 12, 12]. Matsumoto et al. [Matsumoto 11] [ 15] / Support Vector Machine 2 / [ 01, 16] [ 01] [ 16] 3. [ 98]

3 Mapping plot さささおママキマドド引きキキキ KY タタタ意意草ウウあああああリリ充キイイドラカラカ構っっっっお 1 GUI オオキ草草草草 1 3 1 [Matsumoto 12a, Ren 15] [] 671 2 1 300 2 16 50 GUI 1 1 16 (SOM: Self-Organizing Map) [Kohonen

3 3 Mapping plot さささおママキマドド引きキキキ KY タタタ意意草ウウあああああリリ充キイイドラカラカ構っっっっお 1 GUI オオキ草草草草 [Matsumoto 12a, Ren 15] [] GUI (SOM: Self-Organizing Map) [Kohonen 82] かっかさキキキキカカカカカフフフ男激おおおおおお丸 WordNet [Bond 12] [ 97] [ 04] EDR [] positive/negative/neutral [ 06] [ 05] [ 11] positive/negative/neutral 3 positive/negative/neutral

4 WII-A 2017 Number of frequency p n 0 Impression of Youth Slang positive negative neutral 66.8% positive/negative positive negative 18 negative positive Tweet SNS Twitter 1 Twitter Web [] [] [Ren 15] 1, Twitter Tweet 3,875,507 86,932,177 11,290, MeCab ver [MeCab] / / 4 2 Tweet Twitter API [ 11] [ 93] [ 04] Web [Web Corpus] Web Web N-gram 15,322 Tweet Tweet Tweet

5 5 3 Tweet 5,291, ,714,808 6,163, ID Tweet Tweet 2 Tweet [Mikolov 13, Pennington 14]. word2vec [word2vec] skip-gram Tweet T c w j SY M 5 2 No [Matsumoto 14] 4, No.1 No.8 1,567 [ 03, 08] w i Sc(w i ) (1) sim(w i,ys j ) w i ys j ys j = (ys 1,ys 2,...,ys k ) w i k stw j = (stw 1,stw 2,...,stw k ) w i k Sc(w i ) 1.0 Sc(w i ) T h w i

6 WII-A 2017 若者言葉 ys i 概念ベクトル DB 概念ベクトル wv i の取得単語 w j (j=1,2,...,n) と, 若者言葉 ys i との概念ベクトル間の i 概念類似度 wsim を計算し j, 概念類似語の集合 i {(w 1, wsim ), (w i 1 2, wsim ), (w i 2 n, wsim n )} を取得概念類似語の取得 N wsim j i T c 俗語 / 標準語らしさ推定モデル N Y 俗語らしさのスコア Sc(w j ) を計算 -1.0 Sc(w j ) T h 俗語らしさ ( 標準語らしさ ) に基づくフィルタリング Y 感性評価ベクトルの推定 ev j ev i 単語感性評価ベクトル推定モデル ( 感性評価重み行列 M A ) 感性類似度 ( esim i,j ) の計算感性類似度に基づくフィルタリング N esim i,j T e 出力候補から除外 Y λ i,j を計算し, 降順ソート変換候補集合 W 標準語らしさ, 感性類似度, 概念類似度に基づく並べ替え 4 S(w i ) = 1 k H(w i ) = 1 k k sim(w i,ys j ) j=1 k sim(w i,stw j ) j=1 Sc(w i ) = S(w i ) H(w i ) (1) 5 3 [Matsumoto 12b] /, / 2 [ 11] M A M A (2) ev i = (e i 1,e i 2,...,e i 16) i wv i = (v i 1,v i 2,...,v i d ) ys i d

7 7 0.9 M A = = m i=1 ev T i m i=1 e i 1 e i 2. wv i ( v1 i v2 i... vd i ) (2) 率解正標準語俗語 e i d M A (ev j ) (3) k /k ev T j = M A wvj T ( ) = e j 1 ej 2... ej 16 ys i ev i (3) w j ev j esim i,j = sim(ev i,ev j ) T e w j wsim i,j = sim(wv i,wv j ) cosine esim i,j Sc(w j ) 1.0 λ i,j (4) { } esimi,j + wsim i,j λ i,j = 1 Sc(w j ) (4) 1 2, cosine cosine cosine k / k k 20 k cosine cosine window (window:dimension) window ,071

8 WII-A 2017 cosine similarity (3:10) (3:50) (3:100) (3:200) (5:10) (5:50) (5:100) (5:200) (10:10) (10:50) (10:100) (10:200) cosine 5 accuracy (3:10) (3:50) (3:100) (3:200) (5:10) (5:50) (5:100) (5:200) (10:10) (10:50) (10:100) (10:200) 7 7 window 3, window window 10 ( ) Tweet ,033

9 , (5) MRR(Mean Reciprocal Rank) N i R i R i MRR 0 MRR MRR MRR(1) MRR MRR(2) { 1 N MRR = N i=1 1 R i (N 0) 0 (otherwise) (5) Tweet window 10 (10, 50, 100, 200) 4 k 100 T h 0 1 T e 0.5 T c T c T h T e MRR MRR(1), MRR(2) MRR(2) MRR(1) R M proposed(win:10) baseline(win:10) MRR(1) MRR(2) MRR window=10, =100 baseline

10 WII-A R M R M proposed(win:10) baseline(win:10) MRR(1) MRR(2) proposed(win:10) baseline(win:10) MRR(1) MRR(2) MRR 10, 50 8 MRR(1) MRR 10 MRR(2) MRR 7. MRR /

1 6 (2002) [ 10] : (TOD), Vol.3, No. 3, pp. 68 77 (2010) [ 97] : CD-ROM (1999) [] EDR [] http://zokugo-dict.com/. [ 13] :. SLP,, 2013-SLP-96, No. 12, pp. 1 8 (2013) [ 11] :. D,, Vol. 94, No. 3, pp. 540 548 (2011) [ 05] :, Vol.

11 11 15K16077, 15K00425, 15K00309 [ 03] : NTT 1 CD-ROM (2003) [ 08] : NTT 4 CD-ROM (2008) [Bond 12] Bond, F., Baldwin, T., Fothergill, R. and Uchimoto, K.: Japanese SemCor: A Sense-tagged Corpus of Japanese, In Proceedings of the 6th International Conference of the Global WordNet (2012) [ 01] : 7 pp (2001) [ 02] :. TL,, Vol. 102, No. 491, pp. 1 6 (2002) [ 10] : (TOD), Vol.3, No. 3, pp (2010) [ 97] : CD-ROM (1999) [] EDR [] [ 13] :. SLP,, 2013-SLP-96, No. 12, pp. 1 8 (2013) [ 11] :. D,, Vol. 94, No. 3, pp (2011) [ 05] :, Vol. 12, No. 3, pp (2005) [Kohonen 82] Kohonen, T.: Self-organized Formation of Topologically Correct Feature Maps, Biological Cybernetics, Vol. 1, pp (1982) [ 04] : (2004) [Matsumoto 11] Matsumoto, K. and Ren, F.: Construction of Wakamono Kotoba Emotion Dictionary and Its Application, In Proceedings of the 12th International Conference, CICLing2011, Part I, pp (2011) [Matsumoto 12a] Matsumoto, K., Kita, K. and Ren, F.: Emotion Estimation from Sentence Using Relation between Japanese Slangs and Emotion Expressions, In Proceedings of the 26th Pacific Asia Conference on Language, Information and Computation, pp (2012) [Matsumoto 12b] Matsumoto, K., Kita, K. and Ren, F.: Emotional Vector Distance Based Sentiment Analysis of Wakamono Kotoba, China Communications, Vol. 9, No. 3, pp (2012) [Matsumoto 14] Matsumoto, K., Akita, K., Keranmu, X., Yoshida, Y. and Kita, K.: Extraction Japanese Slang from Weblog Data based on Script Type and Stroke Count, Procedia Computer Science, Vol. 35, pp (2014) [ 14] : 76 pp (2014) [MeCab] [Mikolov 13] Mikolov, T., Sutskever, I., Chen, K., Corrado, G. and Dean, J.: Distributed Representations of Words and Phrases and their Compositionality, CoRR, abs/ (2013) [ 14] : 2014-NL-216, No. 12, pp.1 3 (2014) [ 93] : (1993) [] [ 16] : 22 pp (2016) [Pennington 14] Pennington, J., Socher, R. and Manning, C. D.: GloVe: Global Veectors for Word Representation, In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP2014), pp (2014) [Ren 15] Ren, F. and Matsumoto, K.: Semi-automatic Creation of Youth Slang Corpus and Its Application to Affective Computing, IEEE Transactions on Affective Computing, Vol. 7, No. 2, pp (2015) [ 07] : Web DBS, 2007-DBS-141(6), pp (2007) [ 11] : :,. NLC,, Vol. 110, No. 400, pp (2011) [ 06] :, Vol. 47, No. 2, pp (2006) [ 12] : Vol. 19, No. 5, pp (2012) [ 12] : Vol. 24, No. 3, pp (2012) [Web Corpus] Web, corpus/nwc2010/. [word2vec] word2vec, archive/p/word2vec/. [ 15] : Vol. 15, No. 1, pp (2015) [ 98] : (1998) [] : blog133.fc2.com/

12 WII-A

自然言語処理24_705

自然言語処理24_705 nwjc2vec: word2vec nwjc2vec nwjc2vec nwjc2vec 2 nwjc2vec 7 nwjc2vec word2vec nwjc2vec: Word Embedding Data Constructed from NINJAL Web Japanese Corpus Hiroyuki Shinnou, Masayuki Asahara, Kanako Komiya