Recommended Algorithm of Latent Factor Model Fused with User Clustering

摘要

To solve the problems of partial implicit feature information loss and long-time model training caused by matrix factorization on recommended algorithm of latent factor model (LFM), a recommended algorithm of user clustering fused with latent factor model is put forward. Firstly, the users' preference information is used to cluster them, and then the similarity calculation method is used to find the cluster and the nearest neighbor users that are most similar to the target user. Next, training similar clusters with the improved LFM to obtain the user's implicit features Matrix p and item's implicit feature Matrix q, and then generating the predictive score matrix of similar clusters. Finally, the predictive score of similar clusters are weighted and summed to gain the final user score. Compared with the traditional collaborative filtering and LFM, the improved model effectively reduces the training time and the root-mean-square error of predictive score, also improves the accuracy of predictive recommendation based on the experiments on Movielens datasets.