The Application And Research Of RBM In The Recommendation System

Collaborative filtering is an important and popular technology for recommender systems. These methods are classified into user-based collaborative filtering and item-based collaborative filtering. The basic idea of user-based collaborative filtering approach is to find out a set of users who have similar favor patterns to a given user(neighbor of the user) and recommend to the user those items that other users in the same set like, while the item-based collaborative filtering approach aims to provide a user with the recommendation on an item based on the other items with high correlations(neighbor of the item). In all collaborative filtering methods, it is a significant step to find users’ or items’ neighbor, that is, a set of similar users or items. Currently, almost all collaborative filtering methods measure users’ similarity or items’ similarity based on co-rated items of users or common users of items. Although these recommendation methods are widely used in E-Commerce, a number of inadequacies have been identified, including:(1)Data Sparsity. The data sparsity problem is the problem of having too few ratings, and hence, it is difficult to find out correlations between users and items. It occurs when the available data are insufficient for identifying similar users or items. It is a major issue that limits the quality of collaborative filtering recommendation.(2)Recommendation accuracy. People require recommender systems to predict users’ preference or ratings as accurately as possible. However, some predictions provided by current systems may be very different from the actual preferences or ratings given by users. These inaccurate predictions, especially the big-error predictions, may reduce the trust of users on the recommender system.We note that using rated items to represent a user, as in conventional collaborative filtering, only captures the user’s preference at a low lever(item lever). Measuring users’ similarity based on such a low-lever representation of users can lead to inaccurate results in some cases. Thus derived latent factor model. The latent factor model can dig out the score at a deeper lever of the underlying information. This article will focuses on restricted boltzmann machines and matrix factorization to find the hidden information.In this paper, we propose the improved algorithm on the traditional RBM model, and combine SVD to construct the mixed factor model. And the nonlinear dimensionality reduction technology(AE) is introduced to carry out data preprocessing, and the effect is better through experiment.