Research On Personalized Recommendation Algorithm Based On Social Information

Due to the vigorous development of social media,users can easily share their consumption experience with friends on the Internet,making users not only recipients of Internet content,but also producers of Internet content,causing serious information overload problems.The social recommendation algorithm integrates social information into the recommendation algorithm,can provide users with personalized services,recommend the most suitable products for users,and effectively solve the problem of information overload.Through a lot of academic research,it is found that the existing social recommendation algorithm models have the following two problems: 1.The existing social recommendation algorithm models often fail to solve the problem of sparse social information.2.The existing social recommendation algorithm model usually only considers the homogeneity effect in the social relationship without considering the social influence effect.In response to the abovementioned problems,two social recommendation algorithm models are proposed to solve them.The main work content is summarized as follows:(1)Aiming at the problem of sparse social information in existing social recommendation algorithm models,a recommendation algorithm model based on implicit social information and rating bias is proposed.The model is specifically composed of three parts.The first is matrix decomposition,the rating bias is added to the matrix decomposition to better fit the user's rating preferences;the second is the social regularization that integrates social information into the model to make the user vector is similar to the user vector of social friends;the last is collaborative user network embedding,which uses graph embedding technology to generate implicit social information to alleviate the problem of sparse social information.Six groups of comparative models are set up,and comparative experiments are carried out on two datasets respectively.The experimental results show that this model has better rating prediction ability than other mainstream recommendation algorithm models,and effectively solve the problem of sparse social information.(2)Since the recommendation algorithm model based on implicit social information and rating bias does not consider the social influence effect in social scenarios,a recommendation algorithm model based on homogeneity factors and influence factors is proposed.The model improves the weight distribution method in the traditional graph attention network,and adds a bias term in the weight calculation process,so that the model can better aggregate the information of social neighbors.The improved graph attention network was used to model the homogeneity factor and influence factor in the user domain and the item domain respectively.Comparative experiments,ablation experiments and parameter sensitivity experiments were carried out in two datasets with a large amount of data.Experimental results show that the model can better simulate real-life social scenes,has a better performance in the case of large amounts of data,and the rating prediction error has been further reduced.