Automatic Generation of Personalized Annotation Tags for Twitter Users

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 689–692, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Automatic Generation of Personalized Annotation Tags for Twitter Users Wei Wu, Bin Zhang, Mari Ostendorf Electrical Engineering University of Washington, Seattle, WA {weiwu, binz, ostendor}@uw.edu Abstract This paper introduces a system designed for automatically generating personalized annota- tion tags to label Twitter user’s interests and concerns. We applied TFIDF ranking and TextRank to extract keywords from Twitter messages to tag the user. The user tagging pre- cision we obtained is comparable to the preci- sion of keyword extraction from web pages for content-targeted advertising. 1 Introduction Twitter is a communication platform which com- bines SMS, instant messages and social networks. It enables users to share information with their friends or the public by updating their Twitter messages. A large majority of the Twitter users are individ- ual subscribers, who use Twitter to share informa- tion on “what am I doing” or “what’s happening right now”. Most of them update their Twitter mes- sages very frequently, in which case the Twitter mes- sages compose a detailed log of the user’s everyday life. These Twitter messages contain rich informa- tion about an individual user, including what s/he is interested in and concerned about. Identifying an individual user’s interests and concerns can help po- tential commercial applications. For instance, this information can be employed to produce “follow- ing” suggestions, either a person who shares simi- lar interests (for expanding their social network) or a company providing products or services the user is interested in (for personalized advertisement). In this work, we focus on automatically gener- ating personalized annotation tags to label Twitter user’s interests and concerns. We formulate this problem as a keyword extraction task, by selecting words from each individual user’s Twitter messages as his/her tags. Due to the lack of human generated annotations, we employ an unsupervised strategy. Specifically, we apply TFIDF ranking and TextRank (Mihalcea and Tarau, 2004) keyword extraction on Twitter messages after a series of text preprocess- ing steps. Experiments on randomly selected users showed good results with TextRank, but high vari- ability among users. 2 Related Work Research work related to Twitter message analysis includes a user sentiment study (Jansen et al., 2009) and information retrieval indexing. To our knowl- edge, no previously published research has yet ad- dressed problems on tagging user’s personal inter- ests from Twitter messages via keyword extraction, though several studies have looked at keyword ex- traction using other genres. For supervised keyword extraction, (Turney, 2000; Turney, 2003; Hulth, 2003; Yih et al., 2006; Liu et al., 2008) employed TFIDF or its variants with Part-of-Speech (POS), capitalization, phrase and sentence length, etc., as features to train key- word extraction models, and discriminative training is usually adopted. Yih et al. (2006) use logis- tic regression to extract keywords from web pages for content-targeted advertising, which has the most similar application to our work. However, due to the …

Content truncated to 500 words