Research On Sparse Generative Adversarial Nets And Its Application In Recommendation System

In the era of big data,a large number of content with different themes are generated on the Internet,which causes users to be unable to quickly obtain the interested content,which seriously affects the user experience.In order to solve this problem,the recommendation system has achieved successful research and application.Since the rating of users on items can reflect the user's preference for the item,a large number of recommendation extract valuable information from the user's historical rating data on items,and provide recommended content accordingly.Usually,user's historical ratings are represented as a High-Dimensional and Sparse(Hi DS)matrix in existing recommendation algorithms,and extract numerical features from the Hi DS rating matrix.These algorithms ignore the distribution characteristic of user preferences hidden in the Hi DS rating matrix,thereby affecting the performance and accuracy of the recommendation model.In order to solve the above problems,this dissertation proposes two new prediction models.1.Proposing a prediction model based on generative adversarial networksSparse GAN,which can simultaneously extract rating characteristics and distribution characteristics from an Hi DS rating matrix.First,adopting the ability of GAN to learn the complex distribution,the distribution characteristics of ratings are extracted from the Hi DS rating matrix.Secondly,the mean square error between the observed ratings and the corresponding predicted ratings in the Hi DS matrix is integrated to the objective function of GAN.Therefore,the model can learn the rating features in the local area while learning the distribution characteristics of the ratings in an Hi DS matrix.Finally,a large number of experiments have been conducted on real datasets,and the results showed that the prediction accuracy of the model was improved compared with the baseline models.2.Proposing a Siamese Generative Adversarial Predicting Network-SGAPN.The model adopts two GAN models,which are used to learn the distribution characteristics of the ratings and the numerical characteristics of the ratings in an Hi DS matrix.Secondly,the distribution characteristics learner is used to predict the user's preference distribution characteristics,and the distribution characteristics and the ratings characteristics are combined to predict unknown ratings.The model eliminates the learning bias caused by the combination of the mean square error and Wasserstein distance in the Sparse GAN model.Finally,a large number of experiments on four real data sets show that the model has improved prediction accuracy compared with the baseline models.3.Using the proposed SGAPN model as a recommendation engine,a movie recommendation system based on rating data is designed and implemented.The system mainly includes functional modules such as data processing,recommendation service and recommendation result display.