Research On Recommender System Algorithm Based On Collaborative Filtering

With the rapid development of Internet technology,information technology applications have grown rapidly and also quickly penetrated into all aspects of people's life,such as ecommerce websites and the number of Internet information services,which makes people have to be faced with a huge amount of information.It is very difficult to filter out the required information for users,which causes a problem of information overload.As a technology that can provide users with personalized information recommendation services,the recommender system has become the primary solution for academics and industry.After developed in recent years,the recommender system has made a considerable progress.However,under the background of the further growth of network information and resources,some problems of the traditional recommendation technology are gradually exposed.These issues affect the performance of the recommender system from all aspects.This paper focuses on the recommendation diversity and matrix sparsity of the recommender system.This paper analyzes the specific impact of these two problems on the performance of the recommender system,proposing two optimized recommendation methods based on the model-based collaborative filtering recommendation algorithm.This article focuses on the following content:(1)Diversity recommendations help to increase user satisfaction.This paper researches on diversity,analyzing the necessity of diversity for recommender system,introducing the classification of recommended diversity: individual diversity,overall diversity and time series diversity.The individual diversity of the recommendation list is optimized by calculating the variance of the hidden semantic vector to be the variance minimization regularization term,adding it into the matrix decomposition algorithm to form the DBMF model.Experiments show that the method has superior performance in improving the individual diversity of the recommendation list,and the accuracy of the recommendation is improved to some extent by adjusting the diversity of the system.(2)Data sparsity is a long-term,typical problem that has a serious impact on system performance.To ease this problem,this paper proposes to use adding additional information sources to alleviate the impact of data sparsity.After comprehensively analyzing,we use the commodity description document as additional information,and use k-max pooling CNN to process text information.This CNN algorithm extracts the feature matrix of the commodity description document integrating the output result into the PMF model to form the k-CNNMF model.It is proved by experiments that k-CNNMF has good performance for improving recommendation accuracy on more sparse data sets,also can predict scores more accurately than other models.