Towards Effective Event Detection, Tracking and Summarization on Microblog Data

摘要

Microblogging has become one of the most popular social Web applications in recent years. Posting short messages (i.e., a maximum of 140 characters) to the Web at any time and at any place lowers the usage barrier, accelerates the information diffusion process, and makes it possible for instant publication. Among those daily user-published posts, many are related to recent or real-time events occurring in our daily life. While microblog sites usually display a list of words representing the trend topics during a time period (e.g., 24 hours, a week or even longer) on their homepages, the topical words do not make any sense to let the users have a comprehensive view of the topic, especially for those without any background knowledge. Additionally, users can only open each post in the relevant list to learn the topic details. In this paper, we propose a unified workflow of event detection, tracking and summarization on microblog data. Particularly, we introduce novel features considering the characteristics of microblog data for topical words selection, and thus for event detection. In the tracking phase, a bipartite graph is constructed to capture the relationship between two events occurring at adjacent time. The matched event pair is grouped into an event chain. Furthermore, inspired by diversity theory in Web search, we are the first to summarize event chains by considering the content coverage and evolution over time. The experimental results show the effectiveness of our approach on microblog data.