Research On Cross-domain Recommendation Algorithm Based On Autoencoder

With the rapid development of Internet technology,the arrival of the information age makes the data generated by the Internet grow exponentially every year,leading to the explosion of information.It has become very difficult for people to obtain the information they need on the Internet.As an effective solution to the problem of information overload,recommendation systems have been widely used in various fields.Traditional single-domain recommendation systems have always had two problems,namely sparsity and cold start,leading to low accuracy and difficult algorithms to optimize and expand.With the increase in the fields of major ecommerce systems and the rapid development of social media,users’ interaction information in different fields can be complementary,which brings new research directions for recommendation systems.In recent years,the recommendation algorithm based on deep learning has set off a new wave of research.Compared with traditional recommendation methods,the recommendation algorithm based on deep learning forms a denser high-level semantic abstraction by combining low-level features,thereby automatically discovering distributed feature representations of data and better extracting user and item features.In order to improve the accuracy of the crossdomain recommendation algorithm and solve the problem of sparsity of the recommender system,this paper proposes a cross-domain recommendation algorithm based on an autoencoder.The main work is as follows:Aiming at the shortcomings of traditional recommendation algorithms,a cross-domain recommendation method based on stacked autoencoders,Cro-Auto Rec,is proposed.Introducing the cross-domain information into the autoencoder to jointly learn the deeper nonlinear network structure of users and items.Learn and migrate the shared rating mode of the same user in different fields to form an Ucro-Auto Rec model,transpose the rating matrix of the Ucro-Auto Rec model to form an Icro-Auto Rec model,and separately from the prediction accuracy and classification accuracy.Experimental results on three datasets of Amazon and one dataset of Movie Lens show that cross-domain recommendation combined with deep learning can improve the accuracy of rating prediction and classification,and effectively solve the sparsity problem.Aiming at the shortcomings of the Cro-Auto Rec algorithm,a cross-domain recommendation method Gcro-Auto Rec of the autoencoder group is proposed.By migrating the user preferences of the auxiliary domain,the problem of data sparsity in the target domain is reduced,thereby improving the quality of recommendation.First,the two stacked autoencoders respectively map the user and item rating matrices to a deep low-dimensional space to learn the deeper feature representation of users and items.Then,the two autoencoders respectively obtain an output vector about the user and an output vector about the item,and calculate the similarity of the two feature vectors.The higher the similarity,the higher the user’s interest in the item.The experimental results on Movie Lens dataset show that the model effectively solves the sparsity problem and achieves a good recommendation effect in ranking recommendation.