Research On A Recommender System Based On Bayesian CBR

With the popularity of the Internet and the World Wide Web, web pages, as the carrier of Information, can be delivered at a rapid speed. Capturing information and purchasing products through the web has become the mainstream. However, the information resources on the web grow explosively, and people have to spend a lot of time and effort to obtain and filter relevant information before buying a suitable product or service. Researchers have addressed this issue by proposing many strategies and solutions. Recommendation system is one of them.However, for the existing e-commerce recommendation systems, the recommendation quality is poor because of data sparseness, single recommendation algorithm (based on either collaborative filtering or content-based recommendation), and the recommendation only consider the situation that the product property/attribute values are either completely accurate or inaccurate.This thesis addresses the above problems and examines e-commerce recommendation system in depth. More specifically,1. Applying Bayesian networks to the recommendation system, taking into account its uncertainty knowledge expression. Combining the recommendation system with the CBR, it provides powerful technical support to services and products resource discovery and recommendation. A new recommendation system architecture based on Bayesian CBR (BCRS) is proposed, and its function, structure and workflow are designed.2. By applying reasoning method based on Bayesian CBR to the BCRS, this thesis proposes a solution to overcome the problem caused by the retrieval difficulty of large-scale case base. The case representation method is designed, which can integratedly and effectively express customer's profile and their various needs while buying.3. For the algorithm, a hybrid data similarity measure algorithm based on distance is designed taking into account the property classification and interdependence. In this algorithm, the customers'needs information can be either accurate or inaccurate, and either complete or incomplete which is of more practical significance.4.Based on the above techniques and algorithm, a Web service textbook recommendation protocol system is designed and implemented, and experiments are designed to test the above-mentioned algorithms. The experiments show that the above-mentioned methods are effective for good recommendation.In short, this thesis integrates Bayesian networks, CBR and fuzzy logic and applies them to the e-commerce recommendation system. The hybrid data similarity measure algorithm for CBR combined with Bayesian networks is proposed. The proposed approaches in this thesis can facilitate the research and development of e-commerce recommendation systems.