Research On Collaborative Filtering Recommendation Algorithm Based On Denoising Autoencoder

With the rapid development of internet technology,the information resources on the network have shown an exponential growth trend,and the massive information resources have caused the problem of information overload.The recommendation system is an effective method to solve the issue of information overload.Over the past two decades,researchers have continuously proposed new recommendation algorithms and related technologies,which have made the recommendation system develop rapidly.However,there are still many problems and challenges need to be solved in the research of recommendation system and related technologies.This thesis addresses the problems of insufficient use of auxiliary information,inefficient extraction of auxiliary information features,and multi-source auxiliary information fusion in existing recommendation system methods.The main contributions of this work are as follows:(1)In order to efficiently extract auxiliary information features,this thesis proposes a collaborative filtering recommendation algorithm based on semi-autoencoder.Using an efficient and novel deep learning structure semi-autoencoder to extract auxiliary information features of users and items,and then map the extracted features to the latent factors of users and items of the matrix decomposition model through a mapping matrix to achieve joint training of feature extraction and matrix decomposition model and give full play to the supplementary function of auxiliary information to the matrix decomposition model.The experimental results show that the performance of our proposed model has outperformed other baselines,verifying that the feature extraction of auxiliary information by the semi-autoencoder is effective.(2)Aiming at solving the problem of multi-source auxiliary information feature fusion,this thesis proposes a deep collaborative filtering method that combines interactive historical and auxiliary information.Taking the interaction history of users and items as a kind of auxiliary information,we utilize deep denoising autoencoders to extract auxiliary information features of users and items,and use multiple extracted auxiliary information features as the input of the feature fusion network.The purpose of designing this fusion network is to map multiple auxiliary information features to the same feature space and to solve the fusion problem of multi-source auxiliary information.The experimental results show that the proposed model with fusion network has better recommendation accuracy than without,which shows the effectiveness of the proposed method.