The Research Of Collaborative Recommendation Algorithm Based On Semi-Supervised Learning

The popularity of the Internet and the rapid development of information technology,resulting in exponential growth of network resources,further strained information overload.The recommender system is the most effective tool for solving this problem,and the recommendation algorithm is the core of the recommender system.Collaborative filtering is one of the most widely used algorithms for the current application.It recommends items through calculating the similarity of interest preferences among users,thereby recommending items that neighborhoods interested to the target user.However,collaborative filtering algorithm existed data sparseness in the real application.As the number of users and items in the recommender system increases rapidly,the problem become more and more severe.In order to alleviate the influence of sparseness problem on the collaborative filtering algorithm,main works are as follows:1.To solve the inaccurate similarity calculated problem in the case of data sparseness,this paper proposes a novel weighted similarity algorithm of fusion user different degree(WDPCC algorithm).The proposed algorithm considers the influence of the number of common rating items and the difference scale on the similarity calculation.It introduces the relevant weighting factor and the correction factor to modify the similarity.At the same time,the proposed algorithm is applied to collaborative filtering and compared with the traditional similarity algorithm,the experimental results show that the proposed similarity algorithm can effectively alleviate the inaccurate problem of traditional similarity calculation and improve the recommendation quality of the recommendation system.2.To solve data sparseness and few label information in recommendation algorithm,we propose a novel algorithm named collaborative recommendation algorithm based on semi-supervised learning(SSLCF algorithm).First,build heterogeneous information networks through combine the neighbor relationship between users or items with user's rating information.Then we can use the similarity of the improved similarity algorithm(level of interest)as weight between isomorphism(heterogeneous)networks.Second,we use regularization framework algorithm to discriminate label information for unlabeled users and items,we predict rating according to the preferences category of the target users.Experimental results show that,given the small number of labels,the proposed algorithm can solve the few label data issue and helps to improve the quality of recommendation.The proposed method is experimented on the International Standard Data Set MovieLens.MAE and RMSE are used as the indexes to evaluate.The experimental results show that the proposed algorithm has good performance and can and improve the accuracy of the recommendation.