Research On Topic Evolution In Social Networks

Online social network is a virtual social network formed by mapping the interactions between individuals and individuals,individuals and groups,groups and groups of the real world into the Internet.With the rapid development of information technology and the Internet,online social networks have become an important way for people to obtain information,express opinions and feelings.The information about the event which happens in the real world will post in online social network,which forms a topic being discussed widely.As time goes on and the number of relationship among users increases,the public opinion emerged by evolution of the topic will react on the real world and influence event development.What's more,some topics eventually evolve to complaints of social injustice and suspicions of the credibility and execution of the government,which endangered security of state and society.Therefore,the research of topic evolution on social network plays an important role in public security and stability.How to represent the topic,analyze the causes of topic evolution,measure the intensity of topic evolution and track the content of topic evolution are the core contents of our researches.Topic evolution is our key point and is studied from four aspects: evolutiony representation — topic representation model,evolution incentives — user's interest,evolution intensity— topic popularity,evolution content —topics concern.The main contributions are as follows:(1)In the research of topic representation model,this paper proposes a topic representation model based on information entropy and LDA model.The traditional vector space model is based on semantic similarity calculation,which cannot handle some semantic problems well such as polysemy and synonym.And probability model does not consider the word distribution uniformity,which is not suitable for application scenarios such as word collection of topic.What's more,the topic representation model based on behaviors is proposed,combining the analysis of behaviors in discussion of the internal weight.Experiments based on Sina Weibo show that the topic representation models proposed can choose words with more discriminating and representative,which is in line with practical application scenarios of the topic representation model.(2)In the research of user topic interest mining,this paper proposes a method of user interest mining via tags and bidirectional interactions.The traditional analysis of user topic interest is based on text mining technology,that is,user topic interests are acquired through mining technology from the perspective of analyzing semantics.But the results of these methods usually have poor interpretability and can't be shown to users directly.What' more,interaction among users is on account of information,so it is useful to analyze of user interest.The proposed method takes full advantage of tags that are marked or created by users themselves,so results of the method have a good interpretability.Based on Sina Weibo,the tags are discussed under different numbers of interactions respectively.The results show that forward spread has a greater impact on tag spread and the method has a better performance than traditional mining methods.(3)In the research of topic popularity evolution,this paper proposes a prediction method of topic popularity based on similarity relation and co-occurrence relation.Most of the traditional methods analyze the correlation of information popularity between early stage and late stage or predict exact value.But whether the topic will become popular is cared more in actual scene,rather than its exact popular value.The definition of popularity in Weibo is given and popularity is divided into four levels.After that final popularity of the topic is predicted by combining similarity relations and co-occurrence relations of topics.The experiment is designed on Sina Weibo and the result proves the effectiveness of prediction method.(4)In the research of topic focus evolution,this paper proposes an analysis method of topic focus evolution based on density estimation.Most of the existing methods of topic focus evolution are based on the coarse granularity as time slice and documents posted on time slice are analyzed,that is,there is few research on finer granularity as point-in-time.What's more,only documents posted at time point be used is not realistic,because there may be a few documents even no document posed at time point.In our method,the idea that documents posted before time point also have influence to time point is absorbed.On this basis,an analysis framework of topic focus evolution is proposed.Experiment results show that the analysis framework can greatly improve the efficiency of the focus analysis.In summary,this paper aims at topic representation model,user topic interest mining,topic popularity evolution and topic focus evolution in online social network.We carry out experiments on real world dataset,and the experiment results show the feasibility of our methods.The research for topic detection and evolution in online social network is significant to theoretical research and practical applications...