Research On Recommendation Of Tea Products Based On Graph Autoencoder

With the development of the Internet,more and more kinds of tea products can be purchased on the e-commerce platform,and the information is more abundant.At the same time,there is also a large amount of interaction information between tea products and users.More information often means more complex decisions.Faced with this kind of information overload,users are likely to face the dilemma of making difficult choices.The recommendation system can use user behavior data to actively recommend appropriate tea products for users,which not only improves users’ shopping experience,but also improves the sales volume of the e-commerce platform.In the face of massive information of users and tea products,how to efficiently assist consumers to find the tea products they are interested in in time is a subject worth studying.In the current tea product recommendation model,the characteristics of tea product review text are not fully utilized,and the user characteristics and tea product characteristics based on interactive information cannot be fully explored.Graph neural network has the advantage of deep mining data features in interactive information.As an important part of graph-neural network,graph auto-encoder not only has the advantages of graph-neural network,but also can select different neural networks in the encoding and decoding parts.In view of the existing problems in tea product recommendation and the advantages of graph auto-encoder,the main researches are as follows:(1)A semantic feature extraction model of users and tea products based on second-order neighborhood information aggregation was proposed in view of the fact that the review text features of tea products were not fully utilized and the interaction information between users and tea products could not be fully mined in the current tea product recommendation.The feature extraction model takes comment text as interactive information,and extracts semantic features of tea products and users by aggregating second-order neighborhood information through graph attention network.The effectiveness of the feature extraction model is proved by experiments.(2)A tea product recommendation model based on graph auto-encoder was proposed.The encoder of the recommended model is divided into three parts.In the first part,the graph convolutional neural network is used to extract the interaction features of tea products and users based on the scoring information.In the second part,the semantic feature extraction model of users and tea products based on second-order neighborhood information aggregation was used to extract semantic features.On the basis of the first two parts,the third part spliced the characteristics of extracted tea products and users respectively,and integrated them with a fully connected network.Finally,the decoder part of the model uses bilinear decoder to reconstruct users’ ratings of tea products.By real data sets related experiments show that the proposed model of tea product recommendations based on figure since the encoder,make full use of the score information tea products and the comment text information,not only the user and the depth of mining,the characteristics of the tea products and from a certain extent,improve the performance of the tea products recommended,compared with other existing models,has a certain advantage.