Research On Evolution Analysis Method Based On Microblog Hot Topics

With the development of Web2.0,the social media which is mainly represented by microblog,has witnessed a fast development.It has turned into an indispensable platform for the public to communicate and express feelings in their daily lives.Meanwhile,due to the broad use of social platforms,a large amount of meaningful topic data is gathered here.Facing massive amount of data,it is difficult to gather hot events generated within a specific time span.Therefore,it is of great significance to study a method which can efficiently detect hot topics.What's more,according to the need of public opinion analysis,the extraction of evaluation objects within hot topics and the analysis of emotional changes in the time series are of great value in helping the public understand the development of events in time.It is not only an important research topic in natural language processing and text mining,but also an important role in commercial application.This paper is focused on the key technologies of the following fields: hot topic detection,the extraction of evaluation objects,and the analysis of emotional changes.The main research works are as follows:1)This thesis proposes a method for detecting hot topics.To gather hot topics within a specific time span,this thesis combines the semantic dependency feature with a BTM algorithm and presents an algorithm of Topic clustering that is named as B*TM.Algorithm also takes account of forwarding times,the number of comments and the average update rate of microblog to calculate hotspot value.Finally,based on the hotspot value of topic cluster,the detection of hot topic is realized.2)This thesis presents an algorithm on extracting the evaluation object.In order to realize the work of extracting evaluation objects,this thesis adopts the conditional random fields algorithm.What's more,on the basis of the lexical feature and the syntactic feature,it proposes semantic dependency feature to complete the extraction of evaluation objects.3)In terms of extracting opinion sentences,this thesis defines specific extraction models to achieve the task of extracting opinion sentences for microblog comments.In terms of emotional change analysis,it has expanded the emotional lexicon based on the corpus and combined this data with the linguistic features of microblog texts,so as to analyze the emotional change in the time series.The study result shows that the B*TM based algorithm for detecting hot topics can detect hot events effectively.The average accuracy rate of the method for samplingevaluation objects is over 83%.The analysis of emotional changes in the time series has shown great performance.To conclude,the methods mentioned in this study are feasible.The problems that remain and the future study plans are introduced at the end of this thesis.