Research On Collaborative Filtering Recommendation Algorithm Fusion Multi Context

In the time of information explosion, it becomes increasingly difficult to accurately and effectively provide users with the required information. The emergence of search engine ease the information burden of users in a certain extent, but present to the user search results, not for the interest of users take the initiative to provide personalized service. Under this background, the recommendation system came into being. Specifically, the recommendation system to the user as the center, through the study of user behavior, preferences, and environment and other factors, filter the content that is not related to users' preference, so recommend more personalized information to users.Among the recommendation methods, collaborative filtering is the most widely used. The basic idea of collaborative filtering algorithm is to analyze user behavior, to find similar users. We use the similar user preferences to infer the preferences of target user. In the field of electronic commerce, this algorithm can promote the conversion of users from browsing to buyers, so as to improve the ability of cross selling.Although it has achieved great success, the traditional collaborative filtering algorithm only uses score data to find similar users(items), while the recommended effect is not ideal. Many scholars consider the time information and tag information in the CF recommendation algorithm, in order to improve the quality of personalized recommendation. Based on reading a lot of literature, we summarize the key technology and put forward the innovation points and experimental validation. The concrete results of this paper are as follows:(1)Add the time context information to the CF. The time relationship between different users buy the same goods time are used to measure the similarity between users, get the users feature vector; The time relationship between different items that are bought by user are used to measure the similarity between items, get the items feature vector; Finally, the fusion probability matrix decomposition model. Through continuous optimization model, reduces the error rate.(2)Add the tag context information to the collaborative filtering algorithm. The use of label information to enrich the user information(articles), proposed a modeling method based on user(items) tag feature vector. Calculate the similarity of users and items by two similarity map. Consider the user rating time context. Optimization of the nearest neighbor model, dynamic discovery the greatest impact neighbor set of the current user(items).(3) Put forward a kind of fusion time context and context label collaborative filtering algorithm. The time context is used to mine the influence relationship between users and the tag context is used to measure the relationship between items at the same time.Finally we fuse user relation vector and item relation vector to probabilistic matrix factorization model.(4) Put forward a framework of recommendation system integration of a variety of contexts. Given the context data acquisition, user preference extraction, context aware recommendation methods generated.