Research On Bursty Events Detection In MicroBlog

Microblog as a new social network media, with the advantages of fast transmission, strong timeliness and comprehensive content, has become an important channel for the rapid gathering and dissemination of bursty event information. However, the exponential growth of the microblog data makes it difficult for users to understand the details of event timely. Beside, due to the high degree of expression freedom in microblog, the bursty event is easy to spread maliciously. The rumor may bring great hidden danger to our national security and social stability. Therefore, it is of great significance to detect the unexpected bursty events accurately and efficiently from massive microblog data. It can not only help users to obtain critical emergency information in real-time and eliminate the panic psychology, but also assist in the emergency management agencies to grasp the development trend of event, control and guide the direction of public opinion reasonably, finally, provide decision-making information to support emergency management of public opinion.Because microblog contains numerous noises, and it is short text with non-standard writing, the bursty events detection faces many challenges. This paper analyzed bursty features of event and considered the characteristics of microblog data. The bursty-feature-pivot techniques of events detection and the hot degree analysis of public opinion were conducted in the study. In bursty events detection, firstly, the study considered the topic expression ability and bursty performance of words. A reference time window mechanism was introduced. According to word frequency, document frequency, topic label Hashtag, word frequency growth rate feature selection and calculation methods, the algorithm based on dynamic threshold was proposed to extract bursty words. The experimental results show that the algorithm can extract bursty words accurately and effectively. Secondly, an event detection algorithm based on bursty words and agglomerative hierarchical clustering was proposed. The algorithm used the bursty words to represent microblog text as a feature vector. And filtering rules of event three elements were introduced to retain high-quality feature vectors. The Jaccard coefficient was chosen to measure the text similarity. Then, with the microblog text similarity matrix constructed, the agglomerative hierarchical clustering algorithm was used to achieve the detection of bursty events in microblog. The experimental results show that the bursty detection method achieved 80% accuracy rate, which verified the feasibility and effectiveness of the proposed method. According to bursty events detected, the study analyzed user network characteristics and communication modes of microblog. From two perspectives of the user influence and the micro dissemination influence, a calculation model of public opinion hot degree was proposed. Furthermore, public opinion hot degree was analyzed by dividing unit time piece. Finally, through case analysis, found that the model can divide the life cycle stage of public opinion accurately, and reveal the bursty event development trend and change law.