Vol.56 No (Mar. 2015) Twitter 1 1,a) , Twitter SNS Twitter Twitter Visualization of the Human Relationships from Tw

Size: px

Start display at page:

Download "Vol.56 No (Mar. 2015) Twitter 1 1,a) , Twitter SNS Twitter Twitter Visualization of the Human Relationships from Tw"

もえりわにべ
5 years ago
Views:

1 Twitter 1 1,a) , Twitter SNS Twitter Twitter Visualization of the Human Relationships from Twitter Based on the Mass Audience and Consideration of Its Case Studies Akihiro Nishimura 1 Yoshinori Hijikata 1,a) Shoutarou Miwa 2 Shogo Nishida 1 Received: June 29, 2014, Accepted: September 12, 2014 Abstract: Recently, it is popular to exploit users comments or reviews in SNS for marketing research. Twitter, which is one of micro blog services, is one of the popular information sources used for the above purpose. We especially try to extract the information according to celebrities like politicians, actors, comedians, singers and so on. This will be useful for understanding the celebrities. In our study, we use ordinary people s tweet according to a celebrity and the user profile of the tweet s author. We develop a system that visualizes the relationship of celebrities from the viewpoint of ordinary people. For this purpose, we delete some noise text from the tweets, extract latent topics that present the celebrities characteristics and allocate them on two dimensional plane based on the similarities of topics. We examine the visualization results from the viewpoint of validity and discovery. Keywords: Twitter, sentiment analysis, topic model, visualization 1. Web 1 Graduate School of Engineering Science, Osaka University, Toyonaka, Osaka , Japan 2 Advanced Technology R&D Center, Mitsubishi Electric Corporation, Amagasaki, Hyogo , Japan a) hijikata@sys.es.osaka-u.ac.jp 1 Web c 2015 Information Processing Society of Japan 972

2 Wikipedia SNS 1 NAVER *1 Togetter *2 SNS 1 Favotter *3 2ch *4 SNS 1 Twitter *1 *2 Twitter *3 Twitter Favorite *4 2 Twitter 1 tweet tweet 5 [15] 8 tweet [16] tweet tweet Twitter 1 A. tweet B. tweet 2 tweet 2 C. tweet D. ID A B C Latent Dirichlet Allocation LDA [1] 2 2. Twitter tweet [10], [12] Meng Twitter [5] tweet N c 2015 Information Processing Society of Japan 973

3 Twitter SNS Park SNS Cyworld [7] Guy SNS 9 [4] Koga Twitter LDA [6] Web URL URL LDA [9] [8] 3. Twitter 3 1. tweet d i d i D B. i tweet A *5 d i d i D tweet C. tweet i tweet d i d i D tweet D. ID i ID ID d i d i D 1 B ID B A B tweet 3.1 A. tweet B. C. tweet D. ID A. tweet i tweet 3.2 tweet blog URL tweet URL tweet 2 *5 160 c 2015 Information Processing Society of Japan 974

4 tweet (1) URL tweet URL A R (2) A e 1 N N/2 N/2 N e N 4 6 (3) e A (2) B (4) e (2) (4) e N B B (5) R A (2) (6) (6) R tweet tweet tweet tweet URL URL tweet tweet M *6 A C A B C MeCab [13] MeCab Wikipedia TermExtract [14] tweet *6 tweet 3 42 bag-of-words d i d i D D ID 3.3 D d i i d TF-IDF Latent Dirichlet Allocation LDA LDA LDA Blei [1] N d w d =(w 1,w 2,...,w Nd ) D Θ d =(θ 1,θ 2,...,θ T ) Φ t =(φ 1,φ 2,...,φ W ) D N d N d N W T LDA 1. Dir(β) Φ t 2. a. Dir(α) Θ d b. i. Mult(Θ d ) wi d zi d ii. Mult(Φ zd i ) 1 D T Θ Φ α β α β Griffiths [2] D Θ Φ Blei [1] Griffiths Gibbs Sampling [2] c 2015 Information Processing Society of Japan 975

5 Fig. 1 1 LDA A graphical model representation of the LDA. Gibbs Sampling 1 LDA zi d 1 1 [3] 1 j j z j j w j Gibbs Sampling P (z j = t z \j, w) nwj \j,t + β n ( ) \j,t + Wβ n dj \j,t + α n dj \j,( ) + Tα (1) \j j ( ) n wj \j,t w j j t n ( ) \j,t t n dj \j,t d j j t n dj \j,( ) d j j ẑ ˆθ t d ˆφ t w ˆθ t d = nd t + α n d + Tα ˆφ t w = nt w + β n t + Wβ PCA MDS SOM 3 MDS (2) (3) PCA PCA PCA MDS 2 PCA MDS 3.3 CMDS MDS JS 0 Kullback-Leibler divergence KL 2 KL Jensen- Shannon divergence JS JS A B Θ A Θ B JS D JS (A, B) = 1 T θt A 2 k log θa T t k θt R + θt B k log θb t k k k=1 k=1 ( ) θt R k = 1 2 θ A tk + θt B k DJS SOM 3.3 T SOM θ R t k c 2015 Information Processing Society of Japan 976

6 1 Table 1 List of target celebrities. GACKT ZUN Wikipedia (a) (e) (a) 4.1 Twitter API 1 A D *7 Twitter A * tweet B A tweet C ,200 tweet D LDA T [11] *7 *8 Wikipedia Wikipedia 17 LDA α =50/ T β =0.1 1,000 SOM MDS A 2 2 2(d) 2(b) 2(a) (a) 2(c) B 3 BOT 3 3(a) 3(c) A A 3(b) 3(d) c 2015 Information Processing Society of Japan 977

AKB V AKB NEWS V SKE48 MLB NMB48 SAMURAI BOT TPP 2ch Fig.

7 2 Table 2 A tweet Major topics extracted from tweets referring to the celebrities. AKB ( ) AKB 3 B Table 3 Major topics extracted from profiles of users who have referred to the celebrities. AKB V AKB NEWS V SKE48 MLB NMB48 SAMURAI BOT TPP 2ch Fig. 3 3 B - MDS Visualization results using MDS of topics extracted from profiles of users who have referred to the celebrities. Fig. 2 2 A - MDS Visualization results using MDS of topics extracted from tweets referring to the celebrities. 3(e) V 2 B A c 2015 Information Processing Society of Japan 978

5 5 D - MDS Visualization results using MDS of topics extracted from follower of the celebrities.

8 4 C tweet Table 4 Major topics extracted from tweets the celebrities post. Blog Official QT Fig. 4 4 C - MDS Visualization results using MDS of topics extracted from tweets the celebrities post. Fig. 5 5 D - MDS Visualization results using MDS of topics extracted from follower of the celebrities. 3(a) 3(b) A 3(d) B A A C tweet 4 4 A B 4(a) 4(c) V tweet (c) 4(b) A B 4(b) A D ID ID 2 5 c 2015 Information Processing Society of Japan 979

6 6 B - PCA Visualization results using PCA of topics extracted from profiles of users who have referred to the

9 Table 5 5 Summaries of the results for each topic resource. A B C D A B 5(a) 5(d) A B C 5(c) V ID B 5(b) GACKT A B C B 5 A B B A B C A D C B Fig. 6 6 B - PCA Visualization results using PCA of topics extracted from profiles of users who have referred to the celebrities. 4.3 B B PCA 6 MDS 3 SOM 7 PCA 3 PCA Fig. 7 7 B - SOM Visualization results using SOM of topics extracted from profiles of users who have referred to the celebrities. c 2015 Information Processing Society of Japan 980

10 6 Table 6 Summaries of the results for each visualization method. PCA MDS SOM 6(a) 6(c) 6(b) V 6(d) (e) MDS PCA 4.2 B 3 PCA 6 PCA MDS PCA 4.2 B PCA SOM 7 7(a) 7(b) (c) PCA MDS MDS (c) PCA MDS V 7(c) 6 MDS PCA MDS PCA MDS SOM PCA 5. Twitter 2 tweet 2 MDS [1] Blei, D.M., Ng, A.Y. and Jordan, M.I.: Latent Dirichlet Allocation, The Journal of Machine Learning Research, Vol.3, pp (2003). [2] Griffiths, T.L. and Steyvers, M.: Finding Scientific Topics, Proc. National Academy of Sciences, Vol.101, pp (2004). [3] Hofmann, T.: Probabilistic Latent Semantic Indexing, Proc. ACM SIGIR, pp (1999). [4] Guy, I., Jacovi, M., Perer, A., Ronen, I. and Uziel, E.: Same Places, Same Things, Same People? Mining User Similarity on Social Media, Proc. ACM CSCW, pp (2010). [5] Meng, X., Wei, F., Liu, X. and Zhou, M.: Entity-centric topic-oriented opinion summarization in twitter, Proc. ACM SIGKDD, pp (2012). [6] Koga, H. and Taniguchi, T.: Developing a user recommendation engine on twitter using estimated latent topics, Proc. HCII (The 14th International Conference on Human-computer Interaction), pp (2011). [7] Park, S.J., Lim, Y.S., Sams, S., Sang, M.N. and Park, H.W.: Networked Politics on Cyworld: The Text and Sentiment of Korean Political Profiles, The Journal of Social Science Computer Review, Vol.29, No.3, pp (2011). [8] Web Vol.20, No.1, pp (2005). [9] NTT DoCoMo Vol.19, No.3, pp (2011). [10] c 2015 Information Processing Society of Japan 981

Twitter & Vol.2, No.1, pp.82 93 (2012). [11] (2011). [12] Twitter (2013).

com/svn/trunk/mecab/doc/ index.html (accessed 2013-02-18). [14] TermExtract Perl http://gensen.dl.itc.u-tokyo.ac.jp/ termextract.

[15] abc NEWS: Twitter s IPO filing shows 215 million monthly active users, available from http://abcnews.go.

[16] 5 Twitter https://www.biglobe.co.jp/pressroom/release/2014/ 06/140609-a 2014-06-09.

11 Twitter & Vol.2, No.1, pp (2012). [11] (2011). [12] Twitter (2013). [13] MeCab: Japanese morphological analyzer, available from index.html (accessed ). [14] TermExtract Perl termextract.html [15] abc NEWS: Twitter s IPO filing shows 215 million monthly active users, available from com/business/twitter-ipo-filing-reveals-500-milliontweets-day/story?id= (accessed ). [16] 5 Twitter 06/ a GroupLens Research ACM IUI Best Paper Award DEWS WebDB WebDB WebDB MIT IEEE Fellow c 2015 Information Processing Society of Japan 982

TF-IDF TDF-IDF TDF-IDF Extracting Impression of Sightseeing Spots from Blogs for Supporting Selection of Spots to Visit in Travel Sat

TF-IDF TDF-IDF TDF-IDF Extracting Impression of Sightseeing Spots from Blogs for Supporting Selection of Spots to Visit in Travel Sat 1 1 2 1. TF-IDF TDF-IDF TDF-IDF. 3 18 6 Extracting Impression of Sightseeing Spots from Blogs for Supporting Selection of Spots to Visit in Travel Satoshi Date, 1 Teruaki Kitasuka, 1 Tsuyoshi Itokawa 2