Research On Collaborative Filtering Recommendation Algorithms Based On Restricted Boltzmann Machine

With the rapid development of mobile Internet technology,network information is growing explosively.When users face a large number of complex and redundant data,people can't quickly find useful information for themselves,which will lead to the problem of "Information Overload".In order to solve the problem of information overload,a recommendation system came into being.The recommendation system collected historical behavior information of users and extracted item features,and then combined related recommendation algorithms to recommend items of interest to users.The performance of the recommendation system is mainly determined by the quality of the recommendation algorithm.The collaborative filtering algorithm is one of the most widely used recommendation algorithms.Collaborative filtering algorithms also have some deficiencies,such as processing highly sparse data sets,which will lead to a decrease in the accuracy of the recommendation system and data sparsity problems.In addition,when processing large-scale data sets,the calculation efficiency is too low,which leads to a decrease in recommendation efficiency and scalability problems.Therefore,solving the problems of data sparseness and scalability is of great significance to the research of collaborative filtering algorithms.This thesis is based on collaborative filtering based on Restricted Boltzmann Machine,combining user trust,project time weight,multi-source information clustering algorithm and Hadoop platform for research.This thesis obtains the following contents and innovations:1.Aiming at the problem of data sparsity in the collaborative filtering recommendation system.This thesis first improves the similarity calculation between users based on user trust and project time weights,and then uses multi-source information clustering algorithms to perform score prediction;At the same time,this thesis also improved the visible layer of the Restricted Boltzmann Machine model,replacing the binary unit with a Gaussian distribution unit.After the improvement,it can represent real values and reduce model complexity;Finally,this thesis uses linear weighting to fuse the scoring data generated by the improved Restricted Boltzmann Machine model with the scoring data of the clustering algorithm.Experiments show that the hybrid recommendation algorithm improves the recommendation accuracy and alleviates the problem of data sparsity.2.The algorithm scalability problem of collaborative filtering recommendation system.In this thesis,a hybrid recommendation algorithm based on multi-source information clustering and a Restricted Boltzmann machine model is used as a framework,combined with the Hadoop distributed platform,increasing the impulse factor,and improving the training method of the Restricted Boltzmann machine model.Experiments show that the model training time is reduced,the recommendation efficiency is improved,and the scalability problem of the algorithm is alleviated.