A Relevant Content Filtering Based Framework For Data Stream Summarization (1606.07351v1)

Abstract: Social media platforms are a rich source of information these days, however, of all the available information, only a small fraction is of users' interest. To help users catch up with the latest topics of their interests from the large amount of information available in social media, we present a relevant content filtering based framework for data stream summarization. More specifically, given the topic or event of interest, this framework can dynamically discover and filter out relevant information from irrelevant information in the stream of text provided by social media platforms. It then further captures the most representative and up-to-date information to generate a sequential summary or event story line along with the evolution of the topic or event. Our framework does not depend on any labeled data, it instead uses the weak supervision provided by the user, which matches the real scenarios of users searching for information about an ongoing event. We experimented on two real events traced by a Twitter dataset from TREC 2011. The results verified the effectiveness of relevant content filtering and sequential summary generation of the proposed framework. It also shows its robustness of using the most easy-to-obtain weak supervision, i.e., trending topic or hashtag. Thus, this framework can be easily integrated into social media platforms such as Twitter to generate sequential summaries for the events of interest. We also make the manually generated gold-standard sequential summaries of the two test events publicly available for future use in the community.