A Recommendation Algorithm Based On Trust Conditional Transitivity And Merging

At present, the widely used recommender system is based on collaborative filtering algorithm. The principle of the algorithm is to find the recommenders who are similar to the source user and then generate prediction according to these recommenders' preferences. The advantage of collaborative filtering recommender system is that it only depends on the ratings of users, which required less input information. But the collaborative filtering recommendation system also has some problems such as the new users problem, the new items problem and the data sparseness problem. Recommender system based on trust network can effectively solve these problems, Trust recommendation system introduces the users' trust relationships, the algorithm searches trustees through trust relationships between users and generates trustors' preferences based on those trustees' preferences. The current trust recommendation systems assume that trust relationships can transfer in any case, so the system will search all of the trustees who are trusted by trustor. A recommendation system often has millions of users,if the system searches all of the trustees who are trusted by trustor, it is bound to reduce the performance of the algorithm.In order to improve the performance, this paper puts forward two algorithms based on trust network. Firstly, we put forward an algorithm based on trust conditional transitivity and merging, which is called trust stream merging algorithm. Conditional trust transitivity was introduced into the algorithm to filter the search paths, so as to improve the search accuracy and improve the efficiency of the algorithm. At the same time, the rewarding and punishment mechanism was introduced into the algorithm. The algorithm divides recommenders into good recommenders and bad recommenders. The algorithm will reward the good recommenders and punish the bad recommenders according to the recommendation effect to improve the accuracy of the algorithm. We tested the accuracy and coverage,and compared with collaborative filtering recommendation algorithm and other algorithm based on trust. The results show that thealgorithm has a further improvement on the two indicators.Secondly, We try to introduce trust network into matrix factorization algorithm based on well-known matrix factorization algorithm and propose a recommendation algorithm based on trust network and matrix factorization. The algorithm tested the accuracy indicator, and compared with other recommendation algorithms. The results show that the algorithm has a further improvement on the accuracy indicator.