Research On The Recommendation Method Based On Collaborative Filtering

In the face of the various problems and challenges brought by information overload, human beings are constantly studying and developing new technologies, where the personalized active information service technology have the sole advantage, including search engine and recommendation system. Although the search engine can make targeted search services, based on the specific needs information provided by users, but due to the restrictions of the internal mechanism, the user participation is high and service content is limited. And the generation of the recommendation system can make up for the deficiency of the search engine, it can find the user’s interest preference by analyzing the user’s historical behavior data, so as to generate the information content to meet the needs of users. So far, the recommendation system has not only become an essential part of the development of e-commerce, but also is widely used in other fields.The core of the recommender system is the personalized recommendation method, a variety of different recommendation technologies are gradually being studied and practiced widely by the industry at home and abroad, in which collaborative filtering recommendation technology is the most popular. On the basis of analyzing the present situation of domestic and foreign research, this paper discusses the relevant theories and technologies of personalized recommendation, and does further research on collaborative filtering recommendation technology.In this paper, according to the collaborative recommendation method based on memory, an optimized collaborative filtering algorithm is proposed firstly, which is based on the view of data sparsity and user bias. To overcome the influence of data sparsity on the accuracy of user similarity computation, a user preference vector is established to obtain the similarity between users. Also based on the consideration of user bias, it introduced the Bayesian reranking algorithm and established trust subgroup of projects, to gain the user’s local bias for trust subgroup. And it produced the final recommendation by weighting the goal of the projects generated by similar user. Finally experiments are carried out to verify the feasibility of this method.Then, according to the collaborative recommendation method based on model, a recommendation method based on user potential temporal preference is presented, which is considered from the point of view the potentiality of user interest as well as it’s instability and time-mobility on high real-time business in recommender systems. By deeply analyzing the relationship between the user’s historical behavior and the potential interest of the user, a user interest mining method based on the probabilistic topic model is adopted to avoid the ignorance of user potential interest in traditional way. At the same time, it combined with hidden markov model to capture the user preference in real time and found the drift of interests sequence. And repeated experiments verify that the proposed method improved the accuracy of recommendation algorithm in high efficiency business.Finally, aiming at the problem of concept drift existing in recommendation model on the data set of user explicit rating, this paper tried to extend and also made further study for the dynamic and diversity of user interest and sensitivity of the learning process in potential scenario. By synthesizing the global and local influence of time factor on user potential interest, it proposed a two-stage interest learning method based on concept drift. Through the analysis of close relationship between the time factor and the concept drift problem and two-stage study about user interest model on the data set of user-item rating, the effectiveness of this method in the recommendation system to solve the concept drift problem, and the improvement of the quality about the whole recommendation system are proved.