Research On Dual Information Source Model-based Collaborative Filtering Algorithms

In today's highly competitive e-commerce environment, the personalized recommendation has emerged as a critical application which is essential to a Web site to retain visitors and turn casual browsers into potential customers. One of the most successful recommendation techniques is collaborative filtering, whose performance has been proved in various e-commerce applications.However, conventional CF methods suffer from a few fundamental limitations such as the cold-start problem, data sparsity problem, and recommender reliability problem. Thus, they have trouble dealing with high-involvement knowledge-intensive domains. To overcome these problems, researchers have proposed recommendation techniques such as a hybrid approach combining CF with content-based filtering. Because e-commerce Web sites often have various product categories, extracting the many attributes of these categories for content-based filtering is extremely burdensome. Under this background, this thesis developed Dual Information Source Model-Based Collaborative Filtering Algorithms (DISCF) to overcome data sparsity problem and in view of knowledge-intensive project recommendation.First of all, the personalized recommendation in e-commerce is discussed and large associated data from domestic and abroad is searched. this thesis Outlined function of the personalized recommendation system in e-commerce and the application example, gave the personalized recommendation system's model in e-commerce, introduced types and characteristics of the input data in the recommendation system and has done more thorough research to recommendation system's classification and typical technology.Secondly, Collaborative Filtering Algorithms is researched, explaned principle and I/O of Collaborative Filtering Algorithms; two kinds of widely applied Collaborative Filtering Algorithms—User-Based and Item-Based recommendation algorithms are introduced. Then it is appointed that conventional CF methods suffer from a few fundamental limitations, analyzed current proposed solutions to these questions and pointed out their superiority and the deficiency.Thirdly, Dual Information Source Model-Based Collaborative Filtering algorithms (DISCF) is discussed detaily. the CF method forms dual recommender groups—a similar-users'group and an expert-users'group—as credible information sources. Then, it analyzes each group's influence on the target customers for the target product categories. The DISCF method fully considerated the personalized service's actual situation in e-commerce, caused the recommendation system established in the more reliable information source foundation, the simulation testing proved that this method has the better recommendation quality.