Community detection based tweet summarization

Abstract

In today’s world, online social media is one of the popular platforms of information exchange. Twitter is one of the most popular social networking sites wherein people post and interact with the help of short messages known as tweets. During huge number of messages are exchanged between a large set of users, huge information are exchanged daily in microblogs during important events such as cricket match, election, budget, movie release, natural disaster and so on. So, this huge source of information open doors to the researchers to handle different challenges. Twitter employs the searching felicity based on trending/non-trending topic. So, topic-wise grouping of information is required for these tweets so that the stories go under specific groups or ‘clusters’ of relevant title or heading. Along with this, summarization is another challenging task due to information overload and irrelevant information in many of the posts. In this work, a graph-based approach has been proposed for summarizing tweets, where a tweet similarity graph is first constructed, and then community detection methodology is applied on the similarity graph to cluster similar tweets. Finally, to represent the summarization of tweets, from each cluster, important tweets are identified based on different graph centrality measures such as degree centrality, closeness centrality, betweenness centrality, eigenvector centrality and PageRank. The proposed methodology achieves better performance than Sumbasic, which is an existing summarization method.