Research On User Profile Modeling Of Personalized Recommendation In Mobile Environment

The personalized recommendation in mobile environment has become one of the hot research topics. User profile modeling is the core of personalized recommendation. However, there are few researches on user profile modeling in mobile environment. Therefore, this paper studies on user profile modeling and provides of the contribution as below:Firstly, context is integrated into user profile model, and the U-I-C user profile model with context is proposed.The U-I-C user profile model is based on user-item matrix. Context is added to the matrix as another dimension, the user's preferences to items in different context are recorded, and the U-I-C user profile model with context is constructed. The model combines context with user's preference and categorizes context into time, location, terminal and service channel which influences user's using mobile value-added services. Similar research has not yet existed in previous studies. This paper provides a theoretical framework for the user profile modeling in mobile environment.Secondly, the calculation methods for user's preferences of browsing service and downloading service are given. Context similarity algorithm and the recommendation mechanism based on content and context are proposed.The page documents browsed by user are denoted by Vector Space Model, which consists of feature terms and its weights. The calculation of the feature term weight is based on TF-IDF. With the contents user browses, user's browsing behaviors and word numbers that user's terminal screen can display, the user's preferences to page documents are finally calculated. Using the ratio of user's downloading different type of services and user's setting of the services, the user's preferences to downloading services are calculated.Through matching user's current context and historical context, the proportion of current context to historical context is calculated, and the context similarity is proposed.While recommending items to user, context similarity is calculated firstly to get historical context that is similar to user's current context. Then the items that are similar to those which are top user's preference in the historical context, are recommended to user.Compared with previous research, this dissertation improves the calculation method of feature term weight, and proposes new calculation method of user's preference to downloading service, context similarity algorithm and recommendation mechanism. This research result can effectively find out user's preference in mobile environment.Thirdly, an experiment is carried out using Color Ring Back Tone users' data of a mobile operator.Using 200 Color Ring Back Tone users' download data from June to August in 2008, the user profile model, the calculation method of user preference and context similarity, and the recommendation mechanism proposed in this paper are experimented. The experimental results show that the U-I-C user profile model can effectively reflect user's preference in different context, and users also have preference to context. The results validate that user profile model with context is completely feasible.