Research On Tea Product Recommendation Of Graph Convolution Network Integrated With Comment Text

With the rise of online shopping,e-commerce has developed rapidly,and more and more agricultural and sideline products have moved from offline sales platforms to online e-commerce sales platforms.Among them,tea,as a traditional drinking product in my country,is sold on e-commerce platforms such as JD.com and Tmall.However,a large amount of tea product information makes users’ intentions ambiguous.The recommendation system mainly mines user preferences and actively recommends information that may be of interest to users,so that users do not have to search for themselves in massive amounts of information.It is a necessary tool to solve the problem of information overload.Since the development of recommendation systems,a variety of recommendation algorithms have been derived.Traditional recommendation algorithms mainly use collaborative filtering and other technologies to find products or information that users are interested in through complex calculations.The emergence of deep learning has brought about the effect of recommendation models.Significant improvement.The emergence of graph convolutional networks makes the graph structure data of non-Euclidean space better utilized.The user-item relationship in actual recommendation is most similar to the graph structure.Therefore,the use of graph convolutional networks to build recommendation models has become a new trend in the development of current recommendation technology.trend.Aiming at the problem that the graph convolutional network uses too a single form of information,this dissertation proposes a graph convolutional network tea product recommendation model BERT-LightGCN(BERT-Light Graph Convolution Network)that incorporates comment text.The user’s comment text on tea products is used to assist the graph convolutional network to learn the characteristics of users and tea products,and then to analyze user preferences to improve the recommendation effect of the tea product recommendation model.The specific work of this dissertation is as follows:(1)In view of the data sparseness and loss of structural information faced by traditional recommendation models,a graph convolutional network based on deep learning is used to process the interactive information in the user-tea product interaction graph.By using a simplified graph convolution method proposed in LightGCN,the feature information of users and tea product nodes is propagated on the interactive graph,and the features of users and tea products are more detailed and comprehensive,so as to improve the recommendation effect.(2)For most recommendation models,only the user-item interaction information is emphasized,and other effective and available information is ignored.By using the BERT model to process the user’s comment text on tea products,extract the semantic features of the target tea product and the target user from it,and use it to assist the graph neural network to learn node features to improve the robustness of the model.(3)Relevant experiments were carried out on the real data set of Jingdong to verify the effectiveness of the BERT-LightGCN model,and a large number of experiments were conducted to explore the final impact of different hyperparameters on the model.Finally,the advantages and disadvantages of traditional recommendation methods,recommendation methods based on deep learning and BERT-LightGCN are compared.The experimental results show that the model proposed in this dissertation has significant advantages in the recommendation effect of tea products.