Restricted Boltzmann Machine Based on Item Category for Collaborative Filtering

Abstract

To improve the accuracy of recommendation, an new model is proposed in this paper, which solves the problem of data sparseness of collaborative filtering to a certain extent. On the base of Real_value Restricted Boltzmann Machine (R_RBM), the impact of item category on recommendation results are considered. The item category, as a new layer, is added to R_RBM, which aims to predict the missing rating in the matrix. Then the recommendation is produced by combining ICR_RBM and collaborative filtering. The experimental results on MovieLens show that recommended effect of new model is more efficient than traditional RBM and singular value decomposition.