Improved collaborative filtering algorithm of restricted boltzmann machine
Zihua Li, Gang Li · 2019
To overcome the matter of the poor generalization ability resulted by characteristics of homogeneity and the problem that visible layer and hidden layer units only receive 0 and 1 binary data in the unsupervised training of Restricted Boltzmann Machine (RBM) ,the paper introduces the implementation of Collaborative filtering algorithm of RBM with category condition based on real value (R_CCRBM) that can be directly dealing with real value rating data and use category features as additional layer information. For the training of RBM, the category feature information is taken as the training condition for RBM and involved in the activation probability calculation of hidden layer element. Using the Movielens dataset to train and test the model,the experimental results show that this model can improve the efficiency of training speed and feature extraction, and improve the performance of the recommendation system.