Research On Detection And Credibility Analysis Of Burst Events In Social Media

Recently, social media is developing rapidly, providing people a much morebroader platform than ever to express their opinions. Social media, such as Weibo, isbecoming the starting media of more and more burst events, which makes thedetection of burst events in social media very important to public opinion analysis.Meanwhile, there are many rumors speading in social media. These rumors couldcause great damages if they are widely spead. The earlier the rumor is detected, theless damage it could cause.By now, reasearch on detection of burst events mainly mine events through hotwords, got problems of wrongly recognization of cyclical burst events and burstadvertisements. On credibility analysis of burst events in social media, currentmethods mainly adopt ranking or classifer. But almost all the reasearch haven’ttaken user opinions into consideration when judging credibility of an event.Although message content is analyzed, the analysis for the sentiment information isinadequate. Besides, the features used for user evaluating are not well designed.To solve these problems, the paper studied the approach for detection andcredibility analysis of burst events. The paper designed and implemented a approachfor burst events detection, which based on burst hot words discovery and topicoriginality filtering. First discover burst hot words using characteristics frommessage content and propagation, then cluster these hot words. Each cluster respondto one event descripted by responding word set at last. And a originality threshold isused to filter advertisements out of the clusters during the procedure. On the basis ofthe detection part, the paper studied credibility analysis of burst events basing oncharacteristics mining. The approach mines characteristics of message content, users,topics and propagation structure, constructs a classifer to judge the credibility of aburst event.The main contribution of this study as follows: Firstly, the paper designed amethod which improved the problem of wrongly recognization of cyclical burstevents, as the sake of using a look-back window and considering both wordfrequency and grow speed when discover burst hot words; Secondly, propose theconcept of topic originality to filter advertisement events, which increased theprecision of the detection. At last, the paper proposed a approach for credibilityanalysis based on characteristics mining, which could effciently discover rumors insocial media, and the features proposed when judging rumors are good promotionsfor relevant reasearch.