An Empirical Assessment of Contemporary Online Media in Ad-Hoc Corpus Creation for Social Events

International Joint Conference on Natural Language Processing, pages 1375–1382, Nagoya, Japan, 14-18 October 2013. An Empirical Assessment of Contemporary Online Media in Ad-Hoc Corpus Creation for Social Events Kanika Narang, Seema Nagar, Sameep Mehta, L V Subramaniam, Kuntal Dey IBM Research Labs, India {kaninara,senagar3,sameepmehta,lvsubram,kuntadey}@in.ibm.com Abstract Social networking sites such as Facebook and Twitter have become favorite portals for users to discuss and express opinions. Research shows that topical discussions around events tend to evolve socially on microblogs. However, sources like Twitter have no explicit discussion thread which will link semantically similar posts. More- over, the discussion may be evolving in multiple different threads (like Facebook). Researchers have proposed the use of on- line contemporary documents to act as ex- ternal corpus to connect pairs of contex- tually related semantic topics. This moti- vates the question: given a significant so- cial event, what is a good choice of exter- nal corpus to identify evolution of discus- sion topics around the event’s context? In this work, we compare the effectiveness of contemporary blog posts, online news me- dia and forum discussions in creating ad- hoc external corpus. Using social propen- sity of evolution of topical discussions on Twitter to assess the goodness of the cre- ation, we find online news media as most effective. We evaluate on three large real- life Twitter datasets to affirm our findings. 1 Introduction Social media has become a hotbed of user gen- erated content. Multiple online platforms have emerged for users to participate, interact and dis- cuss. Social contact and activity networks like Facebook, video sharing networks like Youtube, photo/image sharing platforms like Pinterest, so- cial bookmarking platforms like Digg, and mi- croblogging platforms like Twitter have become prominent online social media platforms. Research suggests that social microblogging platforms, like Twitter, diffuse information in a manner similar to news media (Kwak and Lee, 2010). In a world of millions of people (Twitter subscribers) with inherent entropy, in absence of explicit discussion threads (unlike online forums, for example), conversations around any event are expected to move towards different directions over time. Contradictory to this apparent expectation, research suggests that these discussions tend to temporally grow and evolve along social relation- ships of people engaging in these discussions, much more strongly, compared to random evolu- tion (Narang and Nagar, 2013). Interestingly, trending events on unstructured microblogs often get built around non-traditional, temporary and contemporary factors, entities and relationships. Because of the sheer number of di- verse contemporary events, event types and asso- ciated documents, it is impossible to prepare well- defined, validated and clean corpus for each and every event. For instance, political turmoils have existed for ages; however, one may not expect a dedicated corpus to preexist for the Libya 2011 turmoil associating its places and locations, con- temporary leaders, and all the other global politi- cal factors. Hence, there is a strong need of using contemporary online media for ad-hoc corpus cre- ation in such a setting. The use of external corpus has been shown to …

Content truncated to 500 words