The Research Of Recommendation Algorithm Based On Complete Tripartite Graph Model

With the rapid development of the Internet, the phenomenon of information overload is becoming more and more drastic. Personalized recommendation technology came into being to solve the concentration that people could hardly find what they were interested in from enormous amount of information. It is one of the most effective and efficient ways to solve the information overload problem. Collaborative filtering is the most widely applied recommendation algorithm in industry. Graph-based personalized recommendation is an important issue in collaborative filtering technology researches.Graph-based personalized recommendation, is a filtration technology based on the graph theory. The graph model was established by the user behavior history data, and then a specific algorithm was executed on the graph model to gain recommendation results. Social tags could provide a wealth of information, which can be involved in recommender system to contribute better description and analysis of user interest. In this paper, we conducted a research on graph-based recommendation technologies, then aiming at the existing problems in recent researches on user-item-tag tripartite graphs such as user interest migration issue and accuracy and diversity dilemma, etc., a complete tripartite graph model with better reflection of user interest was proposed, meanwhile, both items recommendation and tags recommendation on this model were achieved. Main works in this paper is listed as follow:Firstly, in recent researches, since the consideration of user-item-tag relationship on user-item-tag tripartite graphs remained insufficient and required further investigation, complete tripartite graph model was proposed considering the relationship of user-item-tag by reflecting the relationships of user-item, item-tag and user-tag. The user interest migration was researched comprehensively, including time-weighted weight factor to measure connections of different nodes in complete tripartite graph model and to reflect the migration of user interest.Secondly, focusing on the dilemma of accuracy and diversity in recommender system, the mass diffusion algorithm and heat spreading algorithm on complete tripartite graph model were carried out. The former algorithm achieved better accuracy while the latter one achieved better diversity. The two algorithms were combined linearly to recommend items.Thirdly, from the perspective of improving confidence in recommender system, the item-tag joint recommendation mechanism was studied, and tag recommendation algorithm based on complete tripartite graph model was also proposed.Finally, experimental results on Movie Lens dataset showed the effectiveness of complete tripartite graph model and corresponding items recommendation algorithms.