The Solution Of Candidate Set Generation And Cold Start Problem For SVD++ Based Recommender System

With the explosive growth of information,the user also presents a personalized information needs.The recommendation system obtains the user's interest through user preference data,in order to provide users personalized recommendation service.Because of its good accuracy and scalability,SVD++ algorithm is very suitable for large-scale Internet system,which has become a hot research topic.However,cold start problem and candidate set generation have been the bottleneck of SVD++ algorithm.Recommender systems need to predict the user's preference for a subset of items,i.e.,a candidate set,to form a recommendation.The random sampling based candidate set generation method of SVD++ algorithm can not meet the user's interest.On the other hand,new user new item introduced from system expansion can not provide rating data,SVD++ algorithm can not train features for new user and new item,so new user and new item cannot be added to the system,this is the cold start problem.In order to solve the candidate set generation problem,consider that the user's historical behavior has relation to the candidate set,this paper uses the association mining method,and proposes a FP-Tree based candidate set generation method.In this paper,FP-Tree is used to store the association between items,and to find the candidate set.In order to solve the cold start problem,this paper uses the tag model to describe the features of the user and the item,the label model is incorporated into the SVD++ algorithm,and the Tag-Based SVD++ algorithm is proposed.In this algorithm,the feature parameters of the tag are trained by the existing rating data,and the system generates the feature parameters by using the labels provided by the new user's new commodity,thus solving the cold start problem.The experimental results show that,compared with the SVD++ algorithm,the FP-Tree based candidate set generation method meets user's interest more.The Tag-Based SVD++ algorithm can effectively solve the problem of cold start of the system,and can provide high prediction accuracy.