A Novel Collaborative Filtering Recommendation Method Based on Weight Determination

Collaborative filtering generates recommendations for target users by summarizing the preferences of neighbor users, so it does not need to analyze the content of digital resources in the Internet, which makes it a more successful recommendation technology. However, with the rapid growth of the number of users in recommendation systems, collaborative filtering suffers from serious scalability and sparsity problems. In this study, a novel collaborative filtering recommendation algorithm is proposed. Firstly, non-negative matrix factorization is adopted to factor the original rating matrix, and the nearest neighbors are searched on the low-rank user feature matrix, which improves the speed of the collaborative filtering algorithm. Then a weight determination method is designed and applied to collaborative filtering. Compared with the previous collaborative filtering algorithms that only consider rating values, the proposed algorithm takes both rating values and feature weights into account when calculating user similarities, which can more accurately express user relationships, thereby resulting in high quality recommendations. The MAE and F1 values of different collaborative filtering algorithms are compared based on two actual datasets. The proposed algorithm performs better than the popular collaborative filtering algorithms. The complexity values of different collaborative filtering algorithms are also compared. The computational complexity of the proposed algorithm MFFW-CF is O ( m ), which is the lowest among all the algorithms compared.