The Research And Implementation Of Collaborative Filtering Recommendation Based On Genres And Scoring Matrix Filling

The rapid development of technologies,such as Web2.0,mobile Internet,e-commerce and so on,has brought a great convenience to people,which also brings out information overload due to the explosive growth of network resources.How to solve the problem of information overload is one of the most important problems that people face now.Recommender system is an effective technology in handling information overload.The quality of recommender system is largely dependent upon the recommender algorithm adopted.Collaborative filtering algorithm is one of the most successful and widely used algorithms currently.However,it is facing some issues such as sparseness,extensibility and so on.As the number of users and items in the recommender system increases rapidly,these problems become more and more severe.In order to alleviate the data sparsity problem on the collaborative filtering algorithm,In this paper we proposed improved scheme from two aspects.Main works are as follows:Firstly,the sparseness of scoring data will seriously affect the accuracy of similarity calculation.We propose an improved genre-based collaborative filtering algorithm,which can reduce the dimension of data matrix by constructing the user-genre matrix and effectively alleviate influence of sparseness.Aim at the deficiency of user-genre matrix element calculation proposed by Yue Wu,in this paper,we propose an improved method to calculate user-genre matrix and a new method to calculate the similarity degree,which can improve the accuracy of similarity calculation,thus provide a more accurate user interest neighbor.Secondly,when the neighbor user did not score the target item,usually we can use some prediction filling algorithm to fill the blank ratings.The Slope One algorithm is widely used to predict and fill the blank ratings in user-item rating matrix because of its simple principle,easy updating and high efficiency.However,the Slope One algorithm ignores the important factor of the user's rating scale difference.In this paper,we present an improved Slope One algorithm,which can improve the accuracy of the forecast ratings by dealing with the factor of the user scale difference during the prediction.Finally,we propose a cooperative filtering recommendation algorithm based on genre and prediction of rating matrix by combining the advanced user-genre matrix calculation and optimized Slope One algorithm mentioned above.In order to verify the effectiveness of the algorithm proposed in the paper,we carry out the multi-group contrast experiments on the ml-100k data set in the MovieLens.in this experiments,we compare algorithm proposed in this paper with the traditional user-based cooperative filtering algorithm and the cooperative filtering algorithm based on the original Slope One.We use the average absolute error and the root mean square error to evaluate algorithm quality.The experimental results show that the two improved points proposed in this paper are effective to improve the accuracy of the cooperative filtering algorithm.In addition,it is verified that the algorithm proposed in this paper has good recommendation performance and can effectively mitigate the influence caused by sparseness.At the end of this paper,we summarize the whole paper and prospect the follow-up work.