Research On Social Recommendation Method Combining Clustering And Matrix Decomposition

With the rapid rise and vigorous development of Internet information technology,as well as the widespread application of a new generation of mobile smart terminals in our country,these have brought huge amounts of information and data to people's daily lives.Faced with this amount of information,consumers often hope to find the information and data they need quickly and accurately,while manufacturers want to use these data to improve the quality of their products and services.There are several solutions at this stage.Search engines obtain search results by comparing keywords,but it is often difficult for people to provide effective keywords.The recommendation system can get more reasonable results in some aspects,helping users to quickly and effectively filter massive amounts of data to get the information they need.With the continuous in-depth study of recommendation systems,new recommendation algorithms are constantly being proposed.Although traditional collaborative filtering algorithms are most commonly used in scientific discussions and real-world business,they still have shortcomings.The collaborative filtering algorithm only calculates the predicted score data based on the user's scoring information,but the scope involved is too single,and does not take into account the user trust information generated by the user in social communication activities and some unknowable factors.Preference for a certain project.In response to the above problems,this paper proposes an improved hybrid recommendation algorithm,the main contents are as follows:1.Collaborative filtering has the defect of data sparseness.Therefore,the user trust value is defined in the matrix completion process,and the influence of friend familiarity and friend interest on the matrix completion is considered,so that the completion value of the default item can be closer to the user's In reality,the collaborative filtering algorithm based on this can reduce the algorithm error and the recommendation accuracy is higher;then use the improved K-means algorithm to cluster all items to obtain multiple clusters with relatively high internal similarity,So that the efficiency of the algorithm when calculating the collaborative filtering algorithm is improved,and the accuracy of the algorithm is further improved.2.When scoring users,other users may be affected due to uncertainties and other factors.Based on the results of user trust value proposed in this paper,a hybrid algorithm combining user trust value and matrix factorization algorithm is proposed.Before the process of matrix decomposition starts,the user trust value is used as the original scoring item of the filling item,and then the processed matrix is ??decomposed into a user's predicted unknown feature matrix P and an item's predicted unknown feature matrix Q,and then through matrix multiplication The resulting prediction score matrix.Finally,considering the advantages of the two algorithms,a hybrid algorithm using the above-mentioned technology is integrated.This algorithm fully takes into account the limitations of a single algorithm and effectively improves the accuracy of prediction scoring.Finally,this paper applies the algorithm to the FilmTrust dataset to verify the effectiveness of the social recommendation algorithm that combines clustering and matrix factorization.Experimental results show that the algorithm proposed in this paper effectively improves the accuracy of recommendation.