Topic Detection Using a Critical Term Graph on News-Related Tweets.

Social media and online social networks are playing an increasingly important role in our lives, as they attract millions of users around the world. Twitter, one of the most popular micro-blogging services, holds a special position among them, since information shared through this service spreads faster than it would have been possible with traditional sources. There are many interesting works analyzing the information that flows through Twitter. Most of such research focuses on trending topic detection, i.e. what are the people talking about right now. We propose a new method to detect topics using a graph, where nodes correspond to terms and edges correspond to co-occurrence of the two terms in the tweet stream. Dense subgraphs, of this graph, pose special interest, as the nodes that are highly connected share a special relation. Thus, the corresponding terms potentially share a relation too. To explore this fact, we apply a community detection algorithm on the graph. The resulting communities correspond to topics related to various real world events. Experimental evaluation of the results of this technique is also provided on both synthetic and real data.