Research On The Design Of Business Recommender Systems

Living in the Internet age, people often suffer from the information-overload problem. Under this circumstance, researchers introduced recommender systems in early 1990s. In their early stage, recommender systems only focused on pure information filtering field. They filtered out the information that users might be interested in and recommended it to users, in order to alleviate the information-overload problem. It didn't take long for the recommender systems to branch their footsteps into various applications, e.g. music, movie, books or any other items. By now, various recommender systems, whether elaborate or not, have become one of the essential components of a Web site or an e-business systems. Recommender systems have exceeded the basic information filtering function for users, and played an important role in sales promotion, brand forging of business process. The research on recommender systems, have also become an active theoretical researching area. Researchers in this area proposed diversity of algorithms to make better recommendations, which including collaborative-filtering, content-based, hybrid ways or algorithms transplanted from Data Mining, Machine Learning. Some leading research groups in this area developed some experimental systems, e.g. MovieLens, as the platform for algorithm's refining.Most current research on recommender systems focused on the improvements of recommend algorithms, while few researchers shed eyesight on the design and building of recommender systems. What's more, different researchers always introduce different criteria for the evaluation of algorithms. This makes it very complicated and hard to compare different algorithms.To issue above problems, this dissertation devoted in the design of a Business Recommender Systems, which aim at giving recommendations to users to promote product sale.We started from discussing the definition of Recommendation Problems, and finished the follow works:1) Provide a more clear definition of Recommendation Problem. Then analyzed the tradition methods of Recommendation Problem solving had the inherited flaw. Thus a new way is proposed, which basic idea is distinguish the Recommender Systems and the Recommendation Algorithms. An accessory effects of this way is it can give the Recommendation Algorithms some objective, unified evaluating metrics, this allowed the compare between different algorithms.2) Detailed discussed the different evaluating metrics of Recommender Systems and Recommendation Algorithms. Moreover, we find that it's not necessary for a recommender system to consider user satisfaction metrics in its design. Because a "Improvement based on Interaction" methods was proposed to guide the system refining its performance.3) A Intelligent Business Recommender System Architecture was designed in the dissertation, which relied on the methods we found before, and can fulfill the goal of Business Recommender Systems.4) Discussion some common rules of Business Recommender Systems design, from different aspects of Data, Interface, Process, and Algorithm, which was suggested by the Whitten Information Systems Architecture.5) Introduced the Support Vector Machine and Bayesian Network to improve the performance of recommendation algorithms. And then tested the effect of using Information Fusion theory to guide the fusion of different recommendation algorithms.