Research On The Bursty Event Detection Method Based On Social Network

Social network refers to a series of Internet applications based on Web 2.0,which allow users to create and communicate.Events that occur in real life usually attract widespread attention in social networks.Bursty event detection method use machine learning,natural language processing to discover real-life events from massive social network data,which can timely find hot topics in society,and help government and other institutions to timely understand social public opinion and take appropriate measures.Traditional bursty event detection methods aggregate text describing the same event into clusters through online clustering method,and then identify real events by analyzing the trend and trajectory of event features in real life.However,most of the existing methods are inaccurate in extracting event features,inadequate in recognizing in time and unable to adapt to mass data processing.This thesis takes twitter data stream as the research object and bursty event detection as the research object in the following aspects:(1)Putting forward a bursty event detection method based on finite state machine.This method improves the finite state machine and realizes incremental update of new data by storing state-related information of feature words,which solves the problem that the traditional state machine model cannot efficiently process massive real-time data and can recogniz event burst features scalability.We puts forward a method of event judge,which extracte from twitter streams breaking characteristic and applied to clustering cluster emergency discriminant,will help ease the events of the early detection of problems,and use Wikipedia and the Event Registry test set was constructed incident.(2)Putting forward a bursty event detection method based on word correlation..The burst event bursty feature is constructed by considering the influence of users and the retweeted counts of tweets.The overlapping of event words is solved by using the method of multi-belonging spectral clustering in the graph partition.The number of graph partition is difficult to be determined,we compute the optimal number of graph partition is determined by using eigenvalue vector.The method identify the bursty event according to the change of word graph structure.We use real tweet data to validdate multiattribution spectral clustering method and the effect of emergency detection.