Research On Personalized Recommendation Systems Based On Deep Learning In Social Networks

The rapid development of Internet technology has brought a lot of pressure on the information processing of social networks,and a recommendation system is an effective technique to solve the phenomenon of information overload in big data,which has deep applications in the field of social networks.In the field of the Internet,almost all web products use recommendation technology.How people can choose what they need among a large number of goods and services is the first concern of consumers,and that is the real meaning of recommendation systems.Since each person’s needs are unique,making choices based on individual needs is also an important task of recommendation technology.The recommendation algorithm of collaborative filtering is one of the most widely used recommendation algorithms,but the accuracy and personalization precision of the recommendation still needs to be improved due to the problems of data sparsity,the cold start of users,and the strength of social relationships affecting the recommendation effect in social networks.Therefore,in this thesis,a dynamic collaborative filtering recommendation algorithm based on MLP and LSTM and a deep learning-based recommendation algorithm for social network relationships are studied for the above problems,and the detailed research work is as follows.(1)To address the problem of low accuracy due to data sparsity,this thesis studies and designs a collaborative filtering algorithm model based on deep learning techniques.The model first extracts user and item feature vectors through CNN to obtain deeper features;then goes through MLP to train the feature vectors to obtain higher-order feature interaction vectors;followed by adding the capture of explicit and implicit user information.Meanwhile,to cater to the users’ movie-viewing experience,LSTM is added to analyze each user’s historical movieviewing data and reasonably provide each user with the number of movies worth watching in that month.Experimental results show that the proposed algorithm improves the accuracy of movie rating prediction,has a good recommendation effect,and alleviates the data sparsity and cold start problems in traditional collaborative filtering.(2)To address the problem that users trust the recommendations of their friends in social networks but the preferences of their friends do not necessarily match,this thesis introduces a deep learning approach to learn the preferences of users and the social competence of their friends in generating recommendations.We design a deep learning architecture by stacking multiple marginalized denoising autoencoders and also design a joint objective function to enhance the potential representation of social relationships in the hidden layer of the autoencoder so that the potential representation of social relationships is as close as possible to the potential representation of users.Our experimental results on four benchmark datasets validate the effectiveness of the proposed approach.(3)A personalized movie recommendation system based on the LSTM-CMLP recommendation algorithm is designed,which crawls the data of users and movies on the Douban network through crawler technology,and then calculates the data of users’ movies,and the recommendation results are displayed to users through a GUI interface made by Py Qt5.The cold start problem is handled by combining the item-based collaborative filtering algorithm,which can also satisfy the movie recommendation for new users.