Research On Collaborative Filtering Recommendation Algorithm On Stacked Denoising Autoencoder With Auxiliary Information

With the rapid popularization of mobile portable devices and the rapid development of Internet technology,the amount of information in the network is constantly increasing,and the problem of information overload is constantly emerging.As an important way to solve the problem of information overload,recommendation systems are used by scholars in the field of information.Traditional recommendation algorithms use scoring information to learn user interests.In practice,data sparseness may occur,resulting in inaccurate recommendation results.In the traditional recommendation algorithm,only the shallow features of the user's interest are extracted,and the user's auxiliary information is not used.In this paper,the deep features of the project are used to obtain the implicit features of the project,and the auxiliary tag information is used to recommend the target user,effectively improving the recommendation system.Recommended accuracy,research work has focused on the following three areas:1.For the traditional recommendation algorithm,only the shallow features are extracted for prediction,and the prediction results do not dig deep feature information with high accuracy.This paper combines a deep learning model with a recommendation system,extracts feature vectors through the deep learning model,and inputs the obtained implicit feature vectors into a collaborative filtering recommendation model for prediction,thereby alleviating the problem of low recommendation accuracy of traditional recommendation algorithms.2.Aiming at the problem that a single autoencoder extracting data features can easily cause over-fitting and inaccurate recommendation results,a recommendation algorithm based on a denoising autoencoder is proposed.Use the denoising autoencoder to extract theuser's deep features,and improve the activation function of the denoising autoencoder so that it incorporates regularization coefficients during the training process,alleviating the sparse data and preventing the overfitting.Experiments on the MovieLens dataset prove that the score prediction model can effectively improve the accuracy of the score prediction.3.Aiming at the problem that the denoising autoencoder recommendation algorithm does not effectively use the project auxiliary information,the stack denoising autoencoder recommendation algorithm incorporating auxiliary information is proposed.The algorithm uses the stack denoising autoencoder to extract item feature vectors combined with tag auxiliary information,calculates the item similarity weight using the interaction between the user and the item,and calculates the similarity between users based on the weight.Make predictions based on calculation results.The model is experimentally verified and analyzed on different data sets.The experimental results show that the model can improve the accuracy of prediction and has a good recommendation effect.