The Design And Implementation Of Topic Evolution Tracking For Micro-Blog

With the rapid development of network information technology,emerging media such as forums and microblog have become the main channels for people to obtain information.Especially in recent years,being equipped with unique and powerful dissemination mechanism,microblog has been widely involved within users.At the same time,microblog also has rapidly developed into a platform for information sharing,dissemination,and acquisition based on user relationships.Among them,the microblog topic is an important way for users to participate.It is based on microblog hotspots,personal interests,netizens discussions,and other channels,and is the topic page related to a topic word after the topic host has added and modified and edited it.Microblog users can enter this page to post a microblog discussion,and the topic page will also automatically include the relevant microblog containing the topic word.The microblog topic is an important way of reflecting user's personal preferences and discovering the user's behavior habits.Therefore,the user publishes information on the microblog,and the information forms a topic through forwarding and commenting.The tracking of microblog topic evolution has important research value in many fields such as user interest discovery,rumors detection and public opinion tracking.Therefore,how to get topics from microblog and track the evolution of the topics becomes an urgent problem to be solved.However,the tracking of microblog topics has inherent difficulties.Among them,it has a short text and a low word frequency brings great difficulty to the topic detection.Moreover,there are many problems such as topic alignment,topic similarity measure,and topic intensity measure in the tracking of topic evolution in time series,It also brings many challenges for the tracking of topic evolution in microblog topics in time series.Therefore,in order to solve the above problems,we intend to adopt the following method:First of all,this article introduces the Biterm Topic Model(abbreviated as BTM)to process microblog data.BTM is a cluster model of topic analysis,which has advantages for the topic classification of short texts.Secondly,in order to track the evolution of the topic,this thesis introduces the concept of evolutionary matrix in the Online Latent Dirichlet Allocation(OLDA)model,and obtains an online BTM(abbr.OBTM)by extending the BTM.Then,using the OBTM models the text on the time slice to get the topic.Finally,the evolution matrix is used to analyze the topic evolution,and the similarity and intensity of the topic are measured according to the two indexes of Jensen-Shannon divergence and discussion degree.Experiments have proved that the OBTM proposed in this thesis has high efficiency and accuracy in tracking the evolution of microblog topics.