Recommendation Algorithm Research Based On Information Fusion

With the rapid development of Internet, multimedia resources in network increase quickly, and the ways by what people listen a song or see a movie are gradually changing to direct search on the Internet. However, the difficulty of making choices also increases. Facing the huge network library, users often don’t know how to choose resources conform to their interests. So, how to recommend media resources for users according to their requirements becomes the problem which a lot of scholars, websites and other related medias concern.Recommending methods mainly include two forms:TopN list recommending and score prediction. TopN list method needs to analyse the user preferences to recommend the top n items which is similar to the target user preferences; Score prediction needs to have a quantitative analysis on the user-item scores, to accurately predict the score to recommend items which have high grade; Both recommending methods need to mine the potential information of existing users and items in order to achieve accurate recommendation results. In view of the two methods of recommendation, we fuse the information of different spaces, and respectively propose two recommendation algorithms based on information fusion. Different spatial information reflects the user’s habits and preferences from different sides and can effectively improve the result of recommendation. More specific description is as follows:To the TopN list recommendation, we selects music to be recommended and social tags as the main data mining source. It understands tags’ semantic information and simultaneously maps them into three aspects of semantic spaces. In the method, it is analyzed in detail that tags are important to music recommendation. It is verified that the fusion of multi-information in recommendation method can effectively improves music recommendation accuracy.To the score prediction, we choose movies as the target items, and establishes three user-item score prediction model based on the bias matrix factorization algorithm with different information fusion, and then merge all the semantic spaces into the last model. Finally, we compare the prediction accuracy between the proposed model and other popular methods. Experiments show that information fusion can effectively improve the movie recommendation result.