Research And Design Of Recommendation Algorithm For E-commerce Platform Of Agriculturalproducts

The recommender system is widely used in modern e-commerce platform.The recommender system can not only enhance user loyalty and adhesion,but also pull consumption and create benefits for economy.It is that the existing conventional product recommendation technology of e-commerce is not fully applicable to the recommendation of agricultural products,which often leads to the problem of low recommendation accuracy and low user satisfaction.Therefore,this thesis studies and improves the existing agricultural product recommendation algorithm,and proposes new recommendation algorithm for agricultural products.The new recommendation algorithm is applied to the "SouTeChan" e-commerce platform of agricultural products to verify its effectiveness.The research work mainly includes:Firstly,the data set of agricultural products is constructed.In view of the fact that there is no general data set of agricultural product,this thesis crawls the raw agricultural product data with the frame of WebCollector.After filtering and correcting the raw data,we construct the raw agricultural product data which include 8239 users' 55231 ratings and reviews on 1216 agricultural products.Secondly,according to the properties of the agricultural products,this thesis improves the recommendation algorithm for agricultural products.To improve the effect of recommendation for agricultural products,this thesis proposes a novel recommendation model named HGAPR(Hierarchical Graph based Approach for Agricultural Product Recommendation)with graph neural networks.We start from the view of the graph to construct the user-agricultural product bipartite graph,which fully leverages text information about the name and description of products and the reviews of the user for the product.In this way,this thesis alleviate the problems of data sparseness and over-fitting in the collaborative recommendation algorithm.Meanwhile,the hierarchical graph structures in our model can learn hierarchical representations for users and agricultural products,which is more suitable for agricultural product's properties and scenes of agricultural products recommendation.In the dataset of ours,HGAPR proposed in this thesis beats all baselines models and achieves the best recommendation effect.Finally,this thesis verifies HGAPR model in "SouTeChan" e-commerce platform of agricultural products.To realize the recommendation of agriculture products and verify the feasibility of the our model,we build a recommendation module with B/S pattern for "SouTeChan" e-commerce platform of agricultural products.Hope to complement the research of recommendation algorithm,the next step is to analyse and study in combination with the individual needs of users.