Research On Trust Propagation And User Cluster Construction Based Personalized Recommendation Method

With the development of information technology and mobile terminals,the use of the Internet has become more and more popular,and the amount of information generated has also been explosive increasing,and the information explosion has become prominent.In recent years,various types of recommendation systems have been widely used in various fields.Among them,the most widely used recommendation system is based on collaborative filtering recommendation algorithm,but it is still flawed to some degree.For example,with the explosive growth of users and projects,the sparsity of user ratings will become higher and higher,and the sparsity of data will affect the quality of the collaborative filtering algorithm.When new users are entering the system,it is difficult to effectively recommend their preferences,because there is no history preference records.Recently,some scholars have introduced clustering methods into the recommendation model to solve some problems.Although the recommendation model based on cluster reduces the size and complexity of nearest neighbor search,the recommendation model based on cluster tends to ignore the impact of social relationships,and such models are affected by data sparsity as well.Based on the above problems,this paper proposes a personalized recommendation model that integrates trust propagation and user cluster construction by extracting user similarity,building trust model and related clustering algorithms.The model calculates the attribute similarity of the user by using multiple attributes of the user and the user's rating information,extracting the user's direct trust and indirect trust value from the user rating matrix,and using the clustering method of similarity-trust fusion matrix to cluster.To some extent,it can alleviate the user's cold start problem and data sparsity problem,and make the model recommendation swifter,which can better recommend the project to the user's interest.The main contents and conclusions of this paper are as follows:First of all,this paper conducts a detailed review and analysis of the domestic and international researches of trust propagation and recommendation algorithms,and puts forward some problems existing in the current researches.In addition,the paper collates and analyzes the relevant theories and techniques involved in the recommendation model,including similarity measurement methods,personalized recommendation methods and clustering methods.The third chapter is to build a user trust model.The trust model in the recommendation system utilizes the trust information of users in the social network to compensate for the sparsity of the rating data,and extracts implicitly the direct trust value between users through the user-item rating matrix.Combining the weak transitivity of trust and the method of average shortest path of the multipath trust which is proposed in this paper,the indirect trust value can be calculated.Then the direct trust value and the indirect trust value are combined to form the user's trust matrix.It can support effective data for the overall recommendation model.In the fourth chapter,through the analysis of the existing recommendation model,a personalized recommendation method integrating trust propagation and user cluster construction is proposed.Firstly,the similarity of each attribute of different users is calculated by the user's age,gender,occupation and geographical attributes.Then,the user's rating of the item in the rating matrix is used to calculate the similarity degree of different users.Next,these weighted summation of similarities are to obtain the attribute similarity between different users,and the user similarity matrix is formed.Afterwards,the two adjacency matrices are combined to form an adjacency matrix of two-dimensional weights.After that,the clustering algorithm of similarity trust fusion matrix is used to divide the higher similarity users into one cluster,by using the shortest sum of distances between users as the evaluation index.In the online recommendation stage,the comprehensive similarity between the target user and other users in the same cluster is calculated,and the rating of the target users are predicted in combination with other users' ratings on the target items.Through the experimental analysis of the public data set,the evaluation indexes of different cluster recommendation algorithms and the advantages and disadvantages of different recommendation algorithms are compared.The personalized recommendation method proposed in this paper,which integrates trust propagation and user cluster construction,has improved the recommendation effect in data sparsity and cold start.It can be widely applied to business customer cluster analysis,personalized recommendation with rating and other related fields.