Highly Applicable Linear Event Detection Algorithm on Social Media with Graph Stream

Lihua Liu, Mao Wang, Haiwei Jiang, Ziming Li, Peifan Shi, Youhuan Li

Abstract: In this paper, we model social media with graph stream and propose an efficient event detection algorithm that costs only linear time and space. Different from existing work, we propose an LIS (longest increasing subsequence)-based edge weight to evaluate the importance of keywords co-occurrence with regard to an event. LIS-based measure conquers the limitations of existing metrics, such as ignoring the inherent correlations and sensitive to noises. More importantly, we propose a linear time and space stream algorithm to detect event subgraphs from social media. The elegant theoretical results indicate the high scalability of our graph stream solution in web-scale social media data. Extensive experiments over Tweets stream from Twitter confirms the efficiency and effectiveness of our solution.