Mining Phase Evolution for Hot Topics: A Case Study from Multiple Social Media Platforms

Monitoring the evolution phases of real-time event including occurrence, development, climax, decline and ending is crucial for management department to intuitively and comprehensively understand the event and then make better decisions. However, there have been very few studies on performing phase evolution analysis of event using the number of posts at the specific time unit. The challenge of this problem is how to identify temporal pattern and mine topic of different phases automatically. In this paper, we propose a unified phase evolution mining model, it firstly identifies the temporal patterns of phases based on k-means and empirical rules, then, burst detection algorithm is adopted to discover peak interval of all phases, finally, we use a summarization technique TextRank to extract keywords from contents to summarize the topics in each phase. In addition, we perform experiments on two real-world datasets collected from different social media platform to understand the event evolution in a more comprehensive way. Experimental results show the characteristics of event evolution on different social media platforms and verify the efficacy of the proposed model.