Research On User Interests Modeling

With the rapid growth of Network Information Resources, Internet has become an indispensable source of information. However, compared to the huge, chaotic Internet information space, the information that each user really interested in is extremely finite, merely a drop in the bucket. To solve the contradiction between the large information resources supply and the limited target resources requirements, we must find a way to quickly and accurately find the required information from the vast information resources. In such demand-driven, the personalized information service technology has made great progress, and became a focus research in the domain of information service.User interest modeling is the core of the personalized information service technology. Whether user model can accurately reflect the user's interest decides the quality of the personalize information service system. The key issues involved in on the process of user modeling were studied in this paper. The following four tasks are completed in the thesis:(1) Proposed three different strategies of collecting the user interest inform ation: user interest information collecting based on the Sogou text classification data, user interest information collecting based on the pseudo relevance feedback data, user interest information collecting based on the Sogou search system query logs. The study provided a wealth of experimental data.(2) Proposed two-layer model structure to repressent the user interest model, proposed two kinds of evaluation indicators for the model structure, developed a standard answer sets and labeling standards.(3) Realized the category-mapping user modeling system and carried out multiple experiments on it. Experimental results show that the model can better complete the user interest modeling task. Different construction method for the user interest information in the model experimental results obtained by different depending on the quality of experimental data.(4) Proposed a user interest model based on the memory model. The modeling method can complete the update work while studying the model. Finally, the category-mapping based modeling program and the memory- based model program were compared with experimental. Experimental results show that memory-based model method is superior in performance than category-mapping based method.