Research On Data Sparsity Based On Collaborative Filtering Algorithm

With the development of the Internet,the recommendation algorithm has achieved good application in various fields.Among them,the collaborative filtering algorithm is a widely used and successful algorithm,but the collaborative filtering algorithm is also affected by some problems in solving the problem of “information overload”.The problem of data sparsity is the main problem that it faces.The collaborative filtering algorithm relies on the user's rating to calculate the similarity to find the nearest neighbor set,but too sparse scoring data will calculate the inaccurate similarity value,and it is difficult to find the real user neighbor set.Finally affect the accuracy of the recommendation.In this paper,the data sparsity of collaborative filtering algorithms is analyzed,and more in-depth analysis and research are conducted.The following three points are proposed to alleviate this problem and improve the recommendation accuracy.The first score prediction value for the traditional collaborative filtering algorithm is affected by the number of common items.Starting from reducing the sparseness of the nearest neighbor set common scoring items,the nearest neighbor that is similar to the target user but not scoring the target item is included in the score prediction calculation,improved scoring prediction algorithm.Second,from the perspective of reducing the sparsity of the user-item rating matrix,a hybrid filling algorithm based on predictive values and multivariate values is proposed.Calculate the predicted value of the unrated item in the user-item scoring matrix by using the traditional algorithm and save it.Combine the predicted value with the multivariate value according to the filling rule and mix the user-item rating matrix.When using the recommendation,use the filling value as the prediction value and Collaborative filtering algorithm experiments on the matrix after filling Two strategies verify that the filling rule can achieve better recommendation than a single value.Thirdly,starting from the optimization similarity calculation,a hybrid collaborative filtering algorithm based on user interest preference is proposed tomine the user-evaluated project attributes,and the user's preference for the project attributes is calculated,and the user-interest type matrix is??established,and then the user similarity degree is calculated.Then,the similarity of user ratings and similarity of user's interests are combined using a dynamic parameter to form a comprehensive similarity.Finally,experiments are performed on the Movielens data set.The experimental results show that the proposed method can effectively alleviate the data sparse problem and improve the recommendation accuracy.