Collaborative Filtering Algorithm Research On The Integration Of User Relationship Strength

With the rapid development of the Internet,network information is also growing rapidly,a recommended system(RS)represented by a collaborative filtering system,helps users find content of interest in a short period of time.Collaborative filtering algorithm plays an important role in the recommendation system,and its application is most extensive and effective,but there are still a series of problems,such as historical annotation information is too sparse especially new users(project)issues,urgent need to be resolved.In recent years,the rapid development of social networks,social networks are quickly integrated into everyone's life,greatly changed the pattern of people to get along,with the microblogging as the representative of the application by a large number of users sought after,which generate the massive user-centered and interest-related social information.Combining the user's historical behavior and annotating data,how to use this information to improve the performance of collaborative filtering algorithm to alleviate the evaluation matrix data sparse prediction accuracy is not high,is the focus of this paper.In order to solve the above problems,this paper has carried out the following two aspects.First,the study of user similarity model based on positive and negative binary labeling evaluation matrix.The most critical step in the collaborative filtering algorithm is the selection of similar neighbors,and the neighbor selection is based on the user similarity calculation model.In the mode of binary scale scoring matrix,there are some shortcomings in the traditional similarity model.At the same time,because the evaluation information is very sparse and the number of users is too small,the similarity calculation is seriously affected,which affects the prediction accuracy of the algorithm The In the positive and negative binary labeling mode,we compared the three typical similarity calculation methods.It is found that the Jekard's similarity model is more suitable for the scoring model of binary labeling.And the Jekard similarity model based on smoothing function is proposed.The experimental results show that the improved algorithm can effectively alleviate the sparseness of data,thus improving the system prediction performance.Second,the collaborative filtering algorithm for the fusion of user relationship strength.In the social network,the rapid growth of the social information of the users provides us with a lot of information other than the evaluation matrix information,which can be provided to the researchers for mining,which can be used to alleviate the sparseness of data.The focus of this paper is how to make full use of these additional social information to improve the accuracy of the algorithm.Based on the real user information of Tencent microblogging and the similarity calculation method based on the user history evaluation matrix information,the reliability calculation model of the social information is established.And the fusion algorithm is used to evaluate the missing value of the scoring matrix.Compared with the method,the fusion algorithm is used to increase the amount of information.It can be seen from the experimental results that the cooperative filtering algorithm combining the strength of user relationship effectively reduces the influence of the evaluation matrix too sparse,which makes the prediction performance of the algorithm significantly improved.