Recommendation Algorithm Based on Restricted Boltzmann Machine and Item Type

Abstract

Because of the sparsity of the ratings in the recommendation system, the calculation of the neighbors will be affected.The common method is to predict the missing ratings and calculate the neighbors with the prediction ratings.However, due to the deviation between prediction ratings and true ratings, it will also lead to the inaccuracy of nearest neighbors.In order to solve this problem, we use RBM to predict the missing ratings.Considering that the type or label of the item has certain influence on the rating, we introduce the type similarity of the item to modify the original neighbors.So that we get the neighbors which is closer to the target user.In this paper, the new model is applied to the MovieLens data set.The result shows that the results of the new model are better than collaborative filtering based on RBM and collaborative filtering based on SVD.