Research Of Collaborative Filtering Recommendation Algorithms Based On Social Computing

With the development of E-Business, recommendation system, as one of the key techniques, has become the very focus of research on E-Business. Based on mining the personal features, historical behaviors and item features, recommendation system uses the techniques in data mining, machine learning and artificial intelligence to predict its users'desire for items.Among diverse algorithms of recommendation systems, collaborative filtering algorithms are the most wide-accepted ones. However, when facing the exponential incensement of information and users, collaborative filtering algorithms suffer two problems at least:(1) the problem how recommendation systems can utilize social features of users to improve the recommendation results; (2) the problem that users' interests may change as time passes by. To address the first problem, based on the traditional user-based collaborative filtering recommendation algorithms, this paper proposed three recommendation algorithms that utilized the trust computing including trust list mining, trust transfer and trust combination. They are:(1) recommendation algorithm based on trust mining; (2) hybrid recommendation algorithm based on users' similarity and trust mining; (3) hybrid recommendation algorithm based on users' similarity and trust transfer. The experimental results show that the proposed algorithms can provide higher global recommendation precision and coverage and can provide a better solution to solve the cold-start problems.To solve the second problem, based on the traditional item-based collaborative filtering recommendation algorithms, this paper proposed one recommendation algorithm based on interest computing. The algorithm assumed users'behaviors in any e-business platform are motivated by their inner interests or needs and they will change as time passed by. Consequently the proposed algorithm defined interest degree function to measure the change of users' interests caused by time. The experimental results show that the proposed algorithm can provide higher global recommendation precision and coverage. In conclusion, by introducing social computing (trust computing and interest computing), the four proposed algorithms are empirically proved that they can provide a better recommendation and coverage.