Causal Knowledge Flow Model And User Behavior Analysis Method For Public Opinion Web Event

With the repaid development of the Internet,a variety of social media are constantly emerging,such as Weibo and Wechat,which make it possible to find public opinion data(e.g.,news and comments)of events on Web.Moreover,public opinion data presents the characteristics of volume,diversity,value sparsity and fragmentation.Therefore,it is difficult to distinguish which event is urgent or hot event,mine causality knowledge of Web events,and find influential users of Web event.As a result,Users cannot quickly understand the Web events they are interested in and government cannot make comprehensive and rapid public opinion analysis on Web events.Aiming at the above problems,a causal knowledge flow model and user behavior analysis method for Web event is propose to solve the problems of volume,diversity,value sparsity and fragmentation of public opinion data of Web event.1)A Bayesian based method for type discrimination of Web event is proposed.The definition of web event is given,and Web events are divided into three types(i.e.,emergency events,hot events,general events).And then,several important characteristic parameters(i.e.,degree of outbreak,distribution skewness coefficient,volatility,distribution kurtosis coefficient,outliers).are proposed.Finally,a Bayesian based probability algorithm combining characteristic parameters is introduced to discriminate the type of Web event.Solving the problem of volume of public opinion data.2)The extraction of causality and the establishment of knowledge flow of causal events are introduced.The proposed algorithm consists of the following three main steps: First,building cue phrase set and proposing Chinese causal pattern template to extract causal sentences from news corpus.Second,extracting casual event from causal sentences base on grammatical dependencies and part-of-speech tagging.Finally,causal knowledge flows are constructed by the method of spreading activation.Solving the problem of diversity and fragmentation of public opinion data.3)The algorithm of user social influence analysis is proposed.User behavior network and user content network are constructed,respectively,and two networks are associated as a two-layer network.Based on the two-layer network,an iteration algorithm is proposed to find influential users.Solving the problem of value sparsity of public opinion data.4)User behavior network based user role mining method is proposed.This paper defines three user roles(information makers,information promoters,and information bridges),and uses four centralities(degree centrality,closeness centrality,betwenness centrality,and eigenvector centrality)as basic characteristics for user role mining.In additional,different statistical characteristics are introduced for different user role mining.Solving the problem of value sparsity of public opinion data.This paper proposes a causal knowledge flow model and user behavior analysis method for Web event,which not only helps users quickly understand the Web events they are interested in,but also helps the government to timely and accurately control the development of Web events.