A graph based clustering technique for tweet summarization

Twitter is a very popular online social networking site, where hundreds of millions of tweets are posted every day by millions of users. Twitter is now considered as one of the fastest and most popular communication mediums, and is frequently used to keep track of recent events or news-stories. Whereas tweets related to a particular event / news-story can easily be found using keyword matching, many of the tweets are likely to contain semantically identical information. If a user wants to keep track of an event / news-story, it is difficult for him to have to read all the tweets containing identical or redundant information. Hence, it is desirable to have good techniques to summarize large number of tweets. In this work, we propose a graph-based approach for summarizing tweets, where a graph is first constructed considering the similarity among tweets, and community detection techniques are then used on the graph to cluster similar tweets. Finally, a representative tweet is chosen from each cluster to be included into the summary. The similarity among tweets is measured using various features including features based on WordNet synsets which help to capture the semantic similarity among tweets. The proposed approach achieves better performance than Sumbasic, an existing summarization technique.