Research On Recommendation Algorithm Based On Nonnegative Matrix Factorization

Collaborative filtering systems are vulnerable to shilling attacks or profile injection attacks in which malicious users can manipulate the systems’ recommendation output by inserting fake profiles. While some robust collaborative filtering methods based on matrix factorization have been proposed, they suffer from low robustness and recommendation accuracy in the presence of shilling attacks. To solve these problems, on the basis of existing researches, make some depth study of robust recommendation from the construction of loss function, the feature extraction of users, the detection of outliers.Firstly, we propose a robust collaborative filtering recommendation algorithm based on nonnegative matrix factorization. We introduce 1R-norm to construct a robust regularized loss function which can reduce the influence of shilling attacks on parameter estimation. Then we propose an iterative optimization method of feature matrices based on the iterative updating algorithm of nonnegative matrix factorization, which guarantees that the predicted ratings are accurate and nonnegative. Finally, we devise a robust collaborative filtering algorithm, and make an analysis on the stability of the algorithm, which proved the stability of the recommendation algorithm in theory.Secondly, we propose a robust collaborative recommendation algorithm based on entropy of information and nonnegative matrix factorization. Based on the theory of that entropy of information can measure the value of information, we construct the filtering method of outliers. Then, we construct the regularization item by 1R-norm, so that the robust regularized loss is obtained and the robustness of the recommendation algorithm is ensured. Finally, applying the principle of NMF to manipulate the learning-rate, we propose an efficient iterative stochastic gradient descent strategy, thus an efficient and accurate recommendation algorithm using which the rating is nonnegative.Finally, we develop the robust collaborative recommendation algorithms and conduct experiments on the MovieLens dataset to demonstrate its effectiveness, and compared with the traditional robust collaborative recommendation algorithm based on matrix factorization.