Research On Recommendation System Based On Binary Network

The rapid development of Internet resources has enabled users to freely access a large amount of media,document information and services,etc.At the same time,the rapid growth of resources will also lead to resource overload.While users have huge information and data resources,they are also deeply immersed in the dilemma of finding content that they are really interested in.In such cases,the recommendation system evolved from information filtering is particularly important to help people discover items of interest,such as movies,books,news,pictures or web pages.However,with the increasing demand of people,people are increasingly relying on the recommendation systems which will take their unique preferences into consideration.Such a recommendation system usually needs to record and analyze a large amount of data of users'behaviors.However,as the number of users increases,more and more data needs to be recorded The traditional recommendation algorithm not only faces the great pressure of how to store the data and cannot process a large amounts of data efficiently in a short time,but also performs poorly on the datasets which has sparse information of users.Recently,the researchers found that the recommendation system based on who rated what to record user behaviors,rather than the recommendation system based on the actual rating,can not only relieve storage pressure to a certain extent,but also gets more accurate recommendation results.What's more,it is more robust So it is of great significance to study and implement a binary recommendation systemThis thesis not only deeply studies and analyzes the current status of the recommendation system,but also gets insight on the concept,theory,current mainstream recommendation algorithm and the commonly used evaluation indicators.Moreover,it summarizes the applicable fields of each recommendation algorithm and evaluates them based on the pros and cons.On top of that,in order to solve the real-time efficiency of the recommendation system and to solve the cold start problem,this thesis has done some works listed as followsFirst,in order to solve the problem of low real-time performance of existing recommendation systems,this thesis introduces a binary recommendation system(BCNN)to improve the efficiency of recommendation.BCNN simplifies the convolution operation in the convolutional neural network by using binary data,converting the multiplication operation in the convolution into an exclusive OR operation which saves a lot of calculation time and improves the efficiency of recommendation system.In this thesis,the BCNN recommendation model is trained on Movielens dataset and compared with the recommendation models based on support vector machine(SVM)and logistic regression(LR).The experimental results show that BCNN model has obvious improvement in efficiency when it guarantees certain accuracy of recommendation system.Secondly,recommendation system usually performs poorly on data sets which has few behavior data of users.It is called the cold start problem.In order to solve this problem,this thesis combines the users' preference information to improve the recommendation quality.Based on the BCNN model,this thesis introduces the RBF,Radial Basis Network,and builds the binary network recommendation system integrated with RBF(RBFCNN)to realize this operation.The main idea of the model is to collect and analyze the user's behavior data so that they can predict the user's preference and use the RBF network to cluster users with similar preferences and finally generate a user-trusted kinship group.In this way,we can make a recommendation between the group.In this thesis,the recommended effects of the RBFCNN model are verified by experiments,and compared with the recommended models based on BCNN,SVM and LR.The experimental results finally show that the RBFCNN model not only ensures the efficiency of recommendation system,but also solve the cold start problem of the recommendation system to a certain extent.When it is used on the different datasets,the recommendation effects are better and stable.