Research On Clustering Method Of Enterprise Microblog Fans' Interest Based On Topic Interaction Supernetwork

Microblog is a kind of broadcast social network platform that shares short real-time information through a mechanism of following.It has extremely open forms of communication and active user groups.In the microblogging platform,interpersonal relationships were build by following each other,the published information was spread by forwarding.As the representative of virtual community,microblogging is a kind of system network with a large number of nodes and complex structure.Enterprise microblog,a new commercial network communication tool along with microblog,is a platform of information dissemination and sharing based on customer relationships.Fans are the core of corporate social assets.Tens of thousands of fans are the potential customers for the enterprise,so it is increasingly important to focus on the data mining of fans.What are the characteristics of these fans? At present,there have been many studies about the method of user clustering.However,most of the literature only took the microblogging content,interest tags,personal information as clustering indicators,which were lack of objectivity.According to the existing researches and characteristics of micro-blogs,a clustering method based on fans' topic interaction super-network was proposed from the perspective of their interests,which is based on unnormalized spectral clustering algorithm.By taking the interaction between fans and topic into account,this method is more real and accurate to acquire the fans' characteristics than the usual ways which singly uses these data of tags and personal information.Surrounding to the research,this paper had done the following jobs.First,a four layer supernetwork model was build which contained users,fans,topic,keywords based on the analysis of microblog.Through the detailed analysis of topic network,interaction network,following network,the paper put forward the method for calculating the degree of interactive interest and used the vector space model to get three different fan interest feature vector.Second,introduced the step of unnormalized spectral clustering algorithm in detail.At the same time,the paper proposed the calculation method of similarity between fans' interests and construction method of diagram for similarity between fans' interests.We also choose Silhouette value index and Davies-Bouldin index to evaluate the clustering effect.Third,in order to collect enough data,the study took the method of web crawler and proposed a way of fans recognition based on the micro relationship.Through the case study,we found that the clustering method based on topic and fans' interactive behavior has a better effect on the discovery of user characteristics.Furthermore,the paper analyzed the clustering results and put forward the marketing strategies which promote the competition ability of the enterprise.