Automatic News Source Detection in Twitter Based on Text Segmentation

PACLIC 28 !195 Automatic News Source Detection in Twitter Based on Text Segmentation Takashi Inui Masaki Saito∗ Mikio Yamamoto Graduate School of Systems and Information Engineering University of Tsukuba 1-1-1 Tenoudai, Tsukuba, Ibaraki 305-8573, JAPAN {inui@,masaki@mibel.,myama@}cs.tsukuba.ac.jp Abstract In this paper, we discuss news source detec- tion (NSD), which involves finding additional information of a message generated in so- cial media to understand the original message more deeply. We propose an NSD method based on the text segmentation and two exten- sion models using web content and post times. Through the experiments using the real-world data, the proposed methods outperformed the baseline methods and exhibited an F-measure of 34.9. 1 Introduction Recently, with the advent of social-media, it has be- come easy to express opinions or comment about experiences. In particular, Twitter1 is a popular ser- vice used worldwide, and extremely large number of messages (tweets) is generated every day on it. It has been widely recognized that Twitter can poten- tially contain much useful information. Therefore, many researchers have conducted content analysis on Twitter (Java et al., 2006; Krishnamurthy et al., 2008; Pennacchiotti and Gurumurthy, 2011; Mehro- tra et al., 2013). Twitter can be regarded as a news feeder (Zhao et al., 2011). News content distributed by other me- dia are often re-distributed and diffused to more peo- ple through Twitter. For example, a user X posted a tweet as follows. ∗Currently, Fujitsu Limited. 1Twitter. https://twitter.com/ tex: Goal! Mario! http://example.football.com Many people have a chance to know the details of Mario’s fantastic goal2 through tex. Web content in- cluded in the URL http://example.football.com func- tions as an information source on tex. It can be said that tweets, such as tex, contain suitable information for news feeders. However, such cases are rare. Al- most all tweets on Twitter are unsuitable due to a va- riety of reasons, e.g. (i) X did not write the informa- tion source in her stream of tweets, (ii) a tweet mes- sage and its information source (URL) were written in separate tweets, or (iii) X included a URL that was not related to the tweet message. In these cases, tweets do not function as the news feeders and peo- ple cannot obtain any additional information from them. We discuss news source detection (NSD), which involves finding additional information of a message generated on social media to understand the original message more deeply. In Twitter, given a tweet ti, the goal with NSD is to find another tweet tj (̸= ti) that includes a reference to its information source on ti. The details of NSD are described in Section 2. We propose an NSD method based on the text seg- mentation. It is difficult to straightforwardly resolve NSD because a search space of tweet pair combina- tions is exponentially large. Therefore, we simplify the NSD problem from the viewpoint of the text seg- mentation and provide an approximate solution. We also discuss two extension models of the proposed method using web content and post times. 2Mario G¨otze …

Content truncated to 500 words