Research On Interest Recognition And User Segmentation Of Enterprise Microblog Fans

In recent years,with the development of Web2.0 technology,social network suddenly emerge.As a new type of social networking media,microblog has accumulated a large number of users in just a few years,among which especially take Sina microblog as representative.Relying on whose huge user groups and traffic,Sina microblog has created a huge commercial value,which has developed into an important marketing tool for enterprises to expand customer relations.Through microblog,enterprises can establish fans network to achieve real-time rapid business information sharing and dissemination.Meanwhile,enterprises can accurately control of the needs of fans by analyzing their relevant microblog information,accordingly provide personalized information recommendation and marketing service.Recognizing fans interest accurately is the key of achieving personalized marketing recommendations based on microblog platform for enterprises,and the massive user information and text contents stored in microblog provide rich data sources for user interest recognition.Achieving user segmentation based on the similarity of fans interest can help companies understand the concerns and needs of fans user groups,so as to carry out personalized information recommendation and marketing services to sub-user groups pertinently to improve the efficiency and effect of microblog marketing.Based on the above background,this paper puts forward a kind of user interest recognition and user segmentation method based on topic model and machine learning.In the construction of microblog user interest model,firstly,through the analysis of information characteristics in microblog,we select the information of special users who is followed by microblog users and microblog contents published,forwarded and commented by microblog users as information sources to mine user interest.Secondly,according to the characteristics of microblog text,Biterm Topic Model suitable for short text information mining is introduced.Considering the timeliness of microblog content and the drift of user interest,we propose time-sensitive T-BTM microblog user interest recognition model to fully mine user long-term interest,short-term interest and overall interest.Collecting real data set of enterprise microblog fans through the Sina microblog API and web crawler,the experimental results prove that the T-BTM model can effectively recognize user interest and its changes,which is superior to the BTM and LDA model in the interest extraction performance.Based on recognizing user interest,the MUSC(Microblog User Spectral Clustering)algorithm based on interest similarity is proposed to divide the fans groups.The user interest topic probability distribution vector is used as the characteristic term,constructing a linear kernel function based on Jensen-Shannon divergence as similar function to build user interest similarity matrix.Solving normalized Laplacian matrix corresponding to similar matrix,and performing spectral decomposition of Laplacian matrix.Aiming at the problem that the number of clusters can't be determined in the traditional spectral clustering algorithm,this paper determines the number of clusters adaptively by analyzing the eigengap between eigenvalues of Laplacian matrix,and the eigenvector space is constructed accordingly.On this basis,carry out K-means clustering on the eigenvector space to complete the clustering of enterprise microblog fans,which are divided into several user clusters with similar interests,and the group interests of user clusters are described by the set of topic feature terms.The results of experiments on real data set show that the MUSC algorithm can effectively divide microblog user groups and discover interests of user groups.Based on user segmentation results,the enterprise microblog can generate a personalized recommendation list by calculating similarity between the candidate document and the interest feature vectors of user clusters.