Graph clustering-based emerging event detection from twitter data stream


Event detection from online social media is nowadays important to many fields, such as crisis notification, health epidemic identification, and trending topic extraction. To deal with the problem, in this paper we propose a new methodology to capture emerging events from Twitter data stream. We define a tweet graph representing tweet term vectors as vertices associated by their content similarities. Based on the assumption that an event denotes a set of similar tweets, we therefore employ the Markov clustering algorithm on the tweet graph to group related tweets. Then, the connected of similar events between consecutive time intervals are classified as an event trend line. Finally, the first one of those connected events will be considered as the emerging event. Performance evaluation of the proposed approach has been done on thirty days of extracted Twitter data stream. The results of detected emerging events have been studied and evaluated by fifteen volunteers with 70-80% precision.

Proceedings of the 5th International Conference on Network, Communication and Computing (ICNCC)