Hot Events Detection And Sentiment Analysis For Chinese Microblog Based On TH-LDA Model

Information which relates to life and interpersonal communication on microblog is growing at an unprecedented rate and spreading in geometric growth.Many bursty topics are firstly appearing in microblog,rapidly spreading,quickly becoming hot spots,causing widespread social resonance,then spreading to traditional media,thus producing huge social influence.It is positive and significant that hot event detection technologies are used to discover the latest social hot spots,becoming timely aware of network public opinion,detect public sentiment,handle paroxysmal things on microblog platform.Massive emotional text information is formed by which users express their views and emotions on hot events and discuss with each other on microblog platform.The valuable information is mined through the analysis of emotional text information.In this paper,the main encountered problems associated with microblog research are pointed out by analyzing the existing research on microblog hotspot detection and sentiment analysis,one of which are the processing of microblog that doesn't include topic tag when microblog topic tag is modeled to detect hot events and sub-events detection,the second are the acquisition of network buzzword and the determination of emotion polarity in sentiment analysis.On this basis,TH-LDA model is proposed to solve the hot event detection,and sentiment analysis method based on dictionary and characteristics of network catchword emotional polarity is put forward.The research work includes the following:(1)TH-LDA model is presented to detect hotspot events,which combines the topic tag(Hashtag),time factor(Time)and topic model(LDA).The microblog texts without topic tag are retrieved to get a more comprehensive set of texts for the same hot event by this model.At the same time,the sub-events of hot event are detected,and thus the evolution of event is tracked.(2)The network catchwords filter(NCF)rule is put forward to construct network buzzwords dictionary and network buzzwords emotional dictionary.We can get the network buzzwords in the microblog texts according to network buzzwords dictionary,determine the emotional polarity of network buzzwords by network buzzwords emotional dictionary.(3)The sentiment analysis of microblog is achieved.We determine the microblog polarity by combining the basic emotional lexicon,emoticons dictionary and network buzzwords emotional dictionary,fusing emotional polarity of network buzzwords.The experimental results show that TH-LDA model can more accurately obtain the microblog texts without topic tag in the same event,achieve the detection of sub-events,and better track the evolution of hot event.The NCF rule can be used to build a more perfect network buzzwords dictionary.The emotional polarity of network buzzwords has a good effect on the rectification of the sentiment polarity of micrblog,and the effectiveness of the method is verified by experiments.