Handling topic drift for topic tracking in microblogs

Microblogs such as Twitter have become an increasingly popular source of real-time information, where users may demand tracking the development of the topics they are interested in. We approach the problem by adapting an effective classifier based on Binomial Logistic Regression, which has shown to be state-of-art in traditional news filtering. In our adaptation, we utilize the link information to enrich tweets?? content and the social symbols to help estimate tweets?? quality. Moreover, we find that topics are very likely to drift in microblogs as a result of the information redundancy and topic divergence of tweets. To handle the topic drift over time, we adopt a cluster-based subtopic detection algorithm to help identify whether drift occurs and the detected subtopic is regarded as the current focus of the general topic to adjust topic drift. Experimental results on the corpus of TREC2012 Microblog Track show that our approach achieves remarkable performance in both T11SU and F-0.5 metrics. ? Springer International Publishing Switzerland 2015.EI