The Research Of Personalized Recommender Algorithms Based On Neighbor Expansion And Semantic Tree

The rapid development of the Internet has given us an extremely popular and convenient living. At the same time, it has brought us into an era of information overload. Growing masses of information confuses the users; therefore they can't find the part they really need quickly. Moreover, some unpopular information becomes dark, for it can't be gotten by general users. To solve the problems above, the personalized recommender system appears. It provides personalized recommendations for each user by building a two-dimension relation of users and items, and digging the potential tastes of users according to their historic behaviors and similarities. The research of personalized recommender techniques has been paid great attention by more and more researchers from kinds of fields.At present, information filtering is the most widely used recommender technique, including collaborative filtering and content-based filtering. However, the traditional collaborative filtering algorithm loses some important information because of data sparsity, and content-based filtering algorithm leaves the semantic information out during the procedure of modeling. In this paper, we propose the corresponding improved algorithms to solve the problems above of information missing. We also verify the effectiveness of our new algorithms through specific experiments.For the data sparsity of collaborative filtering algorithm, we propose two algorithms based on neighbor expansion, introducing two conceptions of potential neighbors and backup neighbors. More useful information can be obtained to make up the data scarcity by expanding the neighbors of the target user. Experimental results show that our algorithms can solve the data sparsity effectively and generate more accurate prediction.For the information loss in content-based filtering algorithm, we propose a new algorithm based on semantic trees. In this algorithm, users and items are represented with semantic trees respectively and semantic information is taken into the similarity computation. Experimental results show that our algorithms can improve the recommendation accuracy.Finally, we present an appliance of recommender algorithms. We construct an expert recommender system to select the appropriate experts to review the new proposals instead of traditional hand operation, which will lead to higher work efficiency.