The User Interest Model Research Based On The Context Information

With the rapid growth of information on the internet,the need for personalized services is growing faster.The large amount of data has brought great convenience to users on the internet,however,a lot of useless information drowns user's interest and increases the difficulty for people to obtain useful information.In the information service system,the introduction of user interest model can provide users with personalized service better,alleviate the problem of information overload effectively,finally improve the user experience.Traditional user interest model obtains content characteristics and interest preference through the analysis of user's browsing content and behavior,but rarely takes into account the influence of user context to user interest.This paper is mainly from the user's point of view,and gives a deep research on user's historical browsing behavior logs combined with user context.Through the acquisition and classification of user's context information,finally two kinds of user interest models based on context information are established.The first model is the hierarchical vector space model based on context(HVSM),which is an improvement on traditional vector space model.Firstly,getting user's context set by calculating the context similarity,and reducing the dimension of the “user-interest item-context” 3D model through the method of context pre-filtering.Then,extracting the user interest categories from websites by rough classification.After that,extracting user interest keywords and weight by mining page content of each interest category deeply.Finally,a hierarchical vector space model based on user context is established.The experimental result shows that the prediction error of user interest degree is controlled within 9%,which optimizes the traditional user interest model.The second model is the user interest topic model based on context(UTM),which is an improvement on topic model.In UTM,the user context and text information are represented by independent hidden variables.After training the model,the topic feature vector and keyword sequence under different contexts are obtained.The experimental result shows that the introduction of context information can provide a more accurate description of user's interest than traditional topic model,as well as reduce the model's perplexity.