An Efficient Event Detection Model For Online Social Networks

With the development of computer technology and the advent of Internet Web2.0, online social networks(eg. Twitter, Facebook and microblogging) gradually become a worldwide platform for social media. Microblogging services are attracting more and more people to share, comment and broadcast their daily activities and events via the online social networks they established with other users. With simple and efficient information generating and propagation mechanism, social network can produce a large number of data information which reflects the current social events.Larger user and mass real-time data makes research on event detection to be a hot topic in recent years, and makes it to be possible for social opinion analysis, rumor detection, and news recommendation.Some specific characteristics of social networks data, such as short text and informal language, bring new challenges to event detection. First, social network data streams contain lots of noisy messages which is meaningless and valueless for event detection. Second, mass data requires more efficient and accurate to detect events. There are lots of other relevant information in social networks except text message, and these relevant information has not yet been fully utilized in existing event detection methods. For noisy message, the existing event detection studies have not proposed some effective methods. In addition, the finally results of event detection methods are based on topics, and the topics are not all real-life event and need to be judged. Nevertheless, the existing event detection methods need to judge the results of event detection manually, and it is lack of efficiency and intelligence.To solve above problems, this paper proposes a novel event detection model,named EVE(Efficient e Vent d Etection) by analyzing the data streams of the posts extracted from microblogs and clustering similar posts together for the purpose of enhancing the accuracy in event detection. The main works and creations are as follows:1.Through the study of relevant information in social networks, in the EVE model, a new method(named relationship assessment method) is presented to infer the importance of posts and users in order to choose high quality posts and high influence users. The relationship assessment method exploits the relationships between users and their corresponding posts to reduce the impacts of massive and noisy data for the purpose of event detection effectively and efficiently.2.In order to further improve the efficiency of event detection, a novel initialization method using the authority scores of the posts is also proposed toinitialize target parameter of EM algorithm, and further improve both the accuracy and the efficiency of the event detection process.3.In order to detect event intelligently, the automatic identification method based on cosine similarity is developed to judge whether the topic can form an important event and to identify those key posts in each event intelligently instead of manually.Experimental results showed that our EVE model exhibits an improved efficiency and accuracy than Baseline Approaches and the automatic identification method judge whether the topic can form an important event and identify those key posts effectively and efficiently.