Research On Personalized Recommendation Algorithm Enhanced By Deep Learning

With the rapid development of Internet technology and the sharply increase of data volume,the recommendation system which can effectively solve the problem of information overload comes into being.Collaborative filtering recommendation algorithm is the most widely used and fastest developing algorithm in the recommendation system.But it is seriously influenced by the Data sparsity and cold boot problems,because it runs by the mutual information between users and project.As a result,the accuracy of the algorithm is limited.Research shows that bringing in the auxiliary information can effectively alleviate the cold start and data sparse problems of collaborative filtering algorithm,but the shallow model has the disadvantage of poor feature extraction effect.In recent years,the application of deep learning model to recommendation system is emerging,but the above problems still exist.To solve these problems,to complete the scoring predictions more accurately,we make use of the coder correlation model in deep learning,and integrate the auxiliary information into the traditional recommendation algorithm to enhance the feature extraction ability of the algorithm.The main research contents of this paper are as follows:1.We propose “a latent factor recommendation algorithm based on user classification to enhance the Stacked Sparse Denoising Auto-encoder”.The algorithm adopts the combination mode of “Use deep model to extract project features + use shallow model ”to extract user features.User feature extraction: To make the criteria for indicating functions more reasonable,we improve it using the average of the system scoring system,integrate the user classification matrix into the predictive score of the implicit factor model,determine the category of new users by user classification,effectively solved the problem of cold startup for new users.Cold start of the project,sparse data and the weak ability to extract features from shallow models: Use the "deep learning + implicit factor model" framework,and incorporate noise reduction and sparsity into the model,form a cascade sparse noise reduction self-encoder model.It improves the robustness and generalization ability of the model.Learn the deep features of the project from its basic information,incorporate it into the implicit factor model,form personalized recommendation algorithm based on deep learning enhancement,so the problems of feature extraction and cold start are solved.2.We propose “a latent factor recommendation algorithm incorporating auxiliary information based on the Stacked Sparse Denoising Auto-encoder”.Use Stacked Sparse Denoising Auto-encoder to extract user characteristics,and use the deep learning model to extract project features,so the deep feature representation of users and projects is fully integrated with the scoring matrix.3.We compared the proposed algorithm with five widely used algorithms in the open source data set MovieLens-1M,the results show that,the proposed algorithm is superior to other control algorithms in accuracy,It shows that the new algorithm has effectively improved the accuracy of the recommendation system.And “the latent factor recommendation algorithm incorporating auxiliary information based on the Stacked Sparse Denoising Auto-encoder” works better than “the latent factor recommendation algorithm based on user classification to enhance the Stacked Sparse Denoising Auto-encoder”,it turns out that the characterization ability of the deep learning model is better than the shallow model on enhancement of the recommendation algorithm.