An Optimization Approach for Sub-event Detection and Summarization in Twitter.

In this paper, we present a system that generates real-time summaries of events using only posts collected from Twitter. The system both identifies important moments within the event and generates a corresponding textual description. First, the set of tweets posted in a short time interval is represented as a weighted graph-of-words. To identify important moments within an event, the system detects rapid changes in the graphs' edge weights using a convex optimization formulation. The system then extracts a few tweets that best describe the chain of interesting occurrences in the event using a greedy algorithm that maximizes a nondecreasing submodular function. Through extensive experiments on real-world sporting events, we show that the proposed system can effectively capture the sub-events, and that it clearly outperforms the dominant sub-event detection method.