Efficient Event Detection in Social Media Data Streams

Given the popularity of the social networking in the recent past, Microblogging services are also developing and attracting people at a rapid pace. Microblogging facilitates users to share, comment and broadcast their daily activities and events based on the relationships they established with other users in the social networking sites. This paper focuses on event detection by analyzing the document streams of the posts extracted from microblogs, and clustering similar posts together for the purpose of enhancing the accuracy in event detection. To this end, this paper proposes a novel event detection model, named EVE (Efficient eVent dEtection). The proposed event detection model encompasses three integral components. A HITS (Hypertext Induced Topic Search) based scoring method is incorporated in the proposed model to distill high-quality posts and reduce negative impact of meaningless ordinary posts. Probabilistic Latent Semantic Analysis (PLSA) based on probabilistic topic model is used in the proposed model to find latent events in the documents stream. Finally, the EM (Expectation Maximization) algorithm is employed to train the parameters and obtain estimators for describing hot events. EVE exploits the relationships between users and their corresponding posts for the purpose of reducing the impacts of massive and noisy data. Also, this paper presents an initialization method using the authority scores of the posts to improve both the accuracy and the efficiency of the event detection process. Experimental results show that our method exhibits an improved efficiency and accuracy than the existing event detection methods, particularly PLSA.