Research On Social Recommendation Combining Document Contex-Aware

With the rapid development of Internet technology and the exploding growth of information in cyberspace,it is difficult for user to choice effective information with a lot of redundant information.Therefore,it will be a great challenge to analyze large amounts of data and establish an effective information filtering mechanism to find effective information.At present,the recommendation system is widely used to solve the problem of information overload,and social network information is often additionally used to improve the performance of recommender systems.The rating prediction accuracy and the Top-N ranking prediction accuracy are the most popular indicators for measuring recommendation system performance.Aiming at Optimization performance,the specific content and innovation of this thesis are as follows:1.Aiming at the problem of sparse score data and subtle semantic difference in document context,a document context.aware model based on decomposition of convolution matrix is proposed in this thesis.It integrates convolutional neural network(CNN)into social probabilistic matrix factorization to improve performance.First,the model which can get the document contextual potential characteristics of the project.Then,it is used to calculate the potential feature vectors of the project,and combines with the potential feature vectors of users in SocialMF to build a scoring prediction model.Final,it optimizes the loss function between predicted and observed values.The results of the study is that the model can improve results in a very sparsely scored data set.2.Aiming at the problem of insufficient information in sorting model,a multi.dimensional trust Top-N recommendation for context.aware document is proposed in this thesis.First,it use PL model to model project potential features and user potential features.Then,multi.dimensional trust model is constructed by using the user as a truster and trustee.Final,the unified model is optimized to find the best sorting list.The results of the study is that the model can improve the recommendation performance in asymmetric relational data sets.