Collaborative Filtering Algorithm Based On Degree Of Trust And Preference Similarity

With the development of the Internet,information has grown at an unprecedented rate,and the information overload of network users has become increasingly serious.The recommendation system is an important tool that can try to solve the problem of information overload.It is used to recommend information that potential users on the Internet are interested in.Up to now,collaborative filtering technology is the most explored and more mature recommendation algorithm.However,it has some problems such as sparse data and cold start.Therefore,in point of how to solve these problems,it has become a collaborative filtering algorithm research hot-field.The main contents of this paper are as follows:First of all,through in-depth study of the recommendation system domestic and foreign development status,it finds that collaborative filtering technology is the most explored,and more mature recommendation algorithm.The key of collaborative filtering is to compute user similarity.First,it filters the nearest neighbour collection based on user similarity.Then,according to the neighbour user set rating,the target item rating value is inferred.The defect is that it has the problems of sparse data and cold starting.In the second place,in order to solve the difficulty of collaborative filtering algorithm,the collaborative filtering algorithm is improved.First,the collaborative filtering algorithm can not deeply explore the relationship between users.Therefore,the trust degree among users is added to the collaborative filtering algorithm to compute the comprehensive trust among users.Among them,the comprehensive trust among users can be divided into three aspects,such as direct,indirect and personal trust degree.Second,for different target items,the degree of preference among different users is introduced into traditional collaborative filtering algorithm to improved the similarity of users.It is helpful to show the users' real interest.Third,there are two situations in predicting the target item ratings.If the selected item is already evaluated by a neighbor user,the users' comprehensive trust and the improvement of users' similarity are introduced in the collaborative filtering predictive rating equation.It is helpful to solve the sparse data of collaborative filtering.If the selected item is a new project that is not evaluated by a neighbor user,the users' comprehensive trust degree and item similarity are introduced into the collaborative filtering predictive item rating equation.It has an advantage to solve the cold start of collaborative filtering.Last but not the least,the Movielens experimental dataset is used.The average absolute error(MAE)of the four sets of experiments is calculated respectively.The first group compares experimental data sparseness for improved user preference similarity,and the lattter three groups calculate their average error(MAE).Compared their value size,it verifies that the improved algorithm is effective and feasible.Three groups of comparative experiments is included such as improved algorithm contrasts with traditional recommendation similarity algorithm based on users collaborative filtering algorithm,and the improved algorithm in the simulation item cold start condition comtrasts with based on the user collaborative filtering algorithm.According to users' rating data,the results show that the improved algorithm has better recommended performance than the traditional collaborative filtering algorithm for the sparse and cold start.In a word,the basic idea of the new algorithm is to integrate the trust relationship between users into the collaborative filtering algorithm.It is easy to find the potential user relationship.The user's preference degree is used to improve the similarity of users.It is helpful to show the interest of real users.In the process of selecting the nearest neighbour set,the integration of comprehensive trust between users and the improveed of user similarity are enhanced the accuracy and expansubility of the collaborative filtering algorithm.Combined users' comprehensive trust degree,it is improved user similarity and item similarity,it predicts the users' interest for the new items.Moreover,it effectively solves the sparse data and cold start problem of collaborative filtering algorithm.