Intelligent Recommendation Algorithm Based on Nonnegative Matrix Factorization and Bipartite Graph

In the era of data information, Internet data overload makes it difficult for users to quickly obtain the information they need. Intelligent recommendation algorithms are widely used. The traditional collaborative filtering and bipartite graph algorithms have problems such as data sparsity and unreasonable resource initialization. This paper proposed a hybrid recommendation algorithm based on matrix factorization and weighted bipartite graph algorithm. Firstly, the dimension of user-item matrix is reduced by non-negative matrix factorization, and the missing values in the original matrix are completed by using the inner product of the lowdimensional matrix obtained by decomposition, and the completed matrix is clustered. Based on the clustering results, a bipartite graph based on feature weights is constructed for differentiated recommendation to achieve the purpose of optimal recommendation. The experimental results show that the optimized algorithm shows high accuracy on different data sets and different numbers of neighbors. Compared with the traditional collaborative filtering algorithm CF and the weighted bipartite graph recommendation algorithm WBG, the hybrid recommendation algorithm can achieve higher accuracy with a smaller number of neighbors through dimension reduction. The data sparsity and cold start problems are alleviated by filling missing values. By adding feature weights, the hybrid recommendation algorithms of recommendation results maintain high accuracy and interpretability, and improve the performance of the recommendation system.