Evolution Analysis Of Hot Topics On Weibo Based On Topic Model

In the era of mobile Internet,the way of news communication emerges in an endless stream,and micro-blog is one of the most important media.On the micro-blog platform,news events are endless,the news media have been tracking reports,release real-time news;people,each one airs his own views on the micro-blog platform,and it is this behavior will make the platform generate a lot of arguments and discussions related to the event,event related views and discussions,the views reflect the main ideas,thoughts and opinions of the users who involved in the discussion.In order to better analyze the aspirations of the people,to understand public opinion and resolve disputes,it is so urgent to strengthen the analysis and evolution of the micro-blog hot topic.With regard to micro-blog hot topics,existing research is focused on three aspects:simple keywords frequency statistics,Web text clustering and topic model method.However,although these methods applied in traditional text better,but for micro-blog platform which are emergeing network words at any time,often tend to ignore the importance of network words,so,this paper proposes a calculation method of semantic similarity for micro-blog,using Baidu encyclopedia as the semantic vocabulary expansion base for network words of micro-blog,combined with the "HowNet" method to compute semantic similarity of micro-blog,and extends to the micro-blog hot topic discovery,can increase semantics of micro-blog topics for the text clustering,and improve the accuracy rate on the level of semantics.On the basis of finding hot topics,this paper puts forward the topic evolution analysis process,first of all,through the text clustering find hot topic of the micro-blog,and then,the topic evolution model is constructed for the micro-blog platform,analysis topic evolution of hot topics from the level of the documents.The method of topic evolution analysis takes time information as discrete variable,and the topic evolution analysis method is based on the discrete analysis method of time information,using the time information classified document collections,followed to sub topics found each time slice of the document,and put forward a time range of subject content correlation analysis method,analysis the evolution of topics for a certain time range.In addition to the correlation analysis of the content of the sub topic,the topic intensity is measured in the sub topic of the mainstream opinion in the hot topic.In the course of experiment analysis,in order to avoid the meaningless topic,this paper puts forward the strategy of meaningless topic filtering to improve the accuracy of topic evolution analysis.Finally,Through the collection of the 2016 full year micro-blog platform social news category micro-blog Bowen data,carried on the hot topic discovery experiment and hot topic evolution analysis,analysis of experimental results,summarized the evolvement social hot topics of social news based on information life cycle theory,the laws can be divided into two categories basically:conventional social hot topic evolution regular pattern and unconventional social hot topic evolution regular pattern,based on the experimental results,from the beginning of the outbreak of the incident,the outbreak of the event and the demise of the event,proposes advance prevention,control and post event analysis of public opinion control methods for the adoption of relevant agencies and departments.