Design And Implementation Of Cold Start Recommendation Algorithm Based On Social Network

With the development of science and the innovation of the internet, human gradually enter the information society. Facing the problem of information overlo ad, recommender system emerges as the times require. The cold start problem is a classical problem of recommender system. Social network has not only chang ed the way people communicate, but also changed the way information is transm itted. Cold start recommendation algorithm researching based on social network h as become popular. In this paper, we study the cold start recommendation proble ms based on social network, mainly include the following aspects:Firstly, we study the cold start algorithm based on user’s similarity and social network. MF algorithm is only suitable for warm start user, witch performs worse for cold start users. SocialMF algorithm is incorporating social network into the MF algorithm, to solve the cold start problem. In this paper, we propose SibSocialMF algorithm, witch combines SocialMF algorithm and user’s similarity. Experimental results show that SibSocialMF algorithm is superior to SocialMF algorithm for both warm start users and cold start users.Secondly, we study the cold start problem based on social network and decision tree model. The original fMF algorithm just uses user’s history rating data to build a decision tree model to solve the cold start problem. Based on the fMF algorithm, we propose SocialfMF algorithm witch incorporates social network; further, the SibSocialfMF algorithm is proposed based on SocialfMF algorithm and user’s similarity. Experiments on real data set shows that both SibSocialfMF algorithm and SocialfMF algorithm perform better than fMF for cold start users, however, SibSocialfMF and SocialfMF are worser than MF method for warm start users.Finally, we compare the performance for the three proposed algorithms SibS ocialMF, SocialfMF and SibSocialfMF on the real data set for both cold start us ers and warm start users. Experimental results show that both SocialfMF algorith m and SibSocialfMF algorithm perform better than SibSocialMF for cold start users, and SibSocialfMF algorithm performs better than SocialfMF algorithm; SibSo cialMF algorithm performs better than both SocialfMF algorithm and SibSocialfM F algorithm for warm start users.