Research And Application Of Collaborative Filtering Algorithm Based On Restricted Boltzmann Machine

With the rapid development of computer information, the Internet to provide users with more and more information and services, in order to solve the problem of information overload,recommended system came into being, recommend the system to provide users with personalized recommendations. At present, the most widely used is the collaborative filtering algorithm in the recommended technology, but the cooperative filtering algorithm also has the problem of sparseness, expansibility and cold start of data. Therefore, a hybrid recommendation algorithm is proposed to integrate different recommended algorithms according to the scene To achieve the recommended purpose. At present, deep learning has made significant progress in many fields.Restricted Boltzmann Machine (RBM) is a model of depth learning. Some scholars have proposed RBM-based collbaborative filtering algorithm and have achieved good results.However, the RBM model still has some defects, RBM model training process only uses the user's scoring data, but the user's scoring data there is a serious sparse problem, resulting in the recommended effect is not very accurate. Based on the RBM model, an Extra Restricted Boltzmann Machine (ERBM) is proposed, and the ERBM model is applied to the recommendation system. The feasibility of ERBM is verified by experiments.Deep learning as a branch of artificial intelligence is applied in many fields, with the depth of learning as a model applied in the personalized recommendation of the field gradually attention.The Restricted Boltzmann Machine (RBM) is one of the most important models in the field of deep learning. RBM,as an "unsupervised" model,can be used to classify and characterize unknown structures. RBM model of collaborative filtering algorithm in the data sparse problems on the existence of precision is not enough accuracy is not high and so on. In this paper, the basic structure of the RBM model is studied, and the improved RBM model is combined with the collaborative filtering algorithm to the recommendation system. The details are as follows:(1) Based on the basic research method of the proposed algorithm, the classical collaborative filtering algorithm is analyzed, including model-based collaborative filtering and neighborhood-based co-filtering, and contrasting the advantages and disadvantages of various model-based algorithms. This paper analyzes the structure of RBM and RBM training algorithm - contrast divergence algorithm, and points out the shortcomings of RBM in collaborative filtering by constructing RBM and user - based collaborative filtering algorithm.(2) RBM collaborative filtering only consider the user's single score, the user's sparse score will directly affect the RBM effect, this paper based on the RBM model proposed additional limited Boltzmann machine (Extra Restricted Boltzmann Machine, ERBM ), The ERBM model adds the additional layer as the condition input on the basis of the RBM model. The input condition of the condition input and the original input is taken as the input condition of the model.The training model is used to compare the divergence algorithm to get the characteristics of the input data. Based on the classical collaborative filtering algorithm, a new user similarity calculation method is proposed. The similarity between the similarity of the project feature and the similarity of the user is used to calculate the similarity of the user. Calculate the similarity of the user, take the highest degree of k users as the input of the ERBM additional layer, with a similar user's score to assist the user's single score, using ERBM to learn the characteristics of the user to predict unknown user ratings. In this paper, the feasibility of the algorithm is verified by the open source dataset MovieLens. The results show that the cooperative filtering algorithm based on ERBM has better accuracy.(3) This paper describes the recommended function module in the online course recommendation system, and analyzes the recommendation scheme in the course recommendation system, analyzes the concrete realization way of the collaborative filtering based on ERBM, and solves the problem of the online course recommendation system Cold start problem.