Multi-tweet Summarization of Real-Time Events

Abstract

Popular real-time public events often cause upsurge of traffic in Twitter while the event is taking place. These posts range from real-time update of the event's occurrences highlights of important moments thus far, personal comments and so on. A large user group has evolved who seeks these live updates to get a brief summary of the important moments of the event so far. However, major social search engines including Twitter still present the tweets satisfying the Boolean query in reverse chronological order, resulting in thousands of low quality matches agglomerated in a prosaic manner. To get an overview of the happenings of the event, a user is forced to read scores of uninformative tweets causing frustration. In this paper, we propose a method for multi-tweet summarization of an event. It allows the search users to quickly get an overview about the important moments of the event. We have proposed a graph-based retrieval algorithm that identifies tweets with popular discussion points among the set of tweets returned by Twitter search engine in response to a query comprising the event related keywords. To ensure maximum coverage of topical diversity, we perform topical clustering of the tweets before applying the retrieval algorithm. Evaluation performed by summarizing the important moments of a real-world event revealed that the proposed method could summarize the proceeding of different segments of the event with up to 81.6% precision and up to 80% recall.