Research On Hybrid-Mode Recommendation Algorithm

With the rapid development of Internet, the online information overload is becoming a serious issue. Traditional resource search technology requires users to search information based on one’s knowledge and experience, which is more and more inefficient with the large amount of information. Based on the idea of "making information to seek users", Recommendation system technology was suggested, and soon became the hot topicAs the most basic and important type of recommendation algorithm, Collaborative filtering algorithm has important research value and application prospects. Research indicates that the current collaborative filtering algorithms still face three major challenges:(1) to reduce sparsity in recommendation systems,(2) to improve scalability in a real-time environment,(3) to resolve the "cold start" problem.In response to these challenges,this paper starts with researches on several collaborative filtering algorithms, focuses on detailed analyses of user-based collaborative filtering algorithm(UBCF) and item-based collaborative filtering algorithm(IBCF).Aiming on problems of these two algorithms,this paper present a hybrid-mode collaborative filtering recommendation algorithm,which blends both user-based and item-based recommendation mechanisms.details are as follows.First,this hybrid-mode algorithm builds the rule of recommendation mode switching between User-Based and Item-based mechanism.Considering the Subjective fuzziness of user rating,this paper proposes a method to convert User historical ratings into fuzzy vector.Based on this fuzzy vector,this paper defines the concept of "Uncertainty of user’s history collection" using Shapley entropy,with this concept,hybrid-mode algorithm switches between UBCF and IBCF.Second,hybrid-mode algorithm proposes an advanced user-based collaborative filtering algorithm.Traditional UBCF does not consider the concept of "Uncertainty of user’s history collection",this paper modifies the formal UBCF with this concept,then shows the experiments results for hybrid-mode recommendation algorithm accuracy on Movielen’s database,as to verify the validity of hybrid-mode algorithm and prove its values.