Design And Implementation Of Agricultural Product Supply And Demand Docking Recommendation System Based Network Representation Learning

As the source of people’s food and the foundation of survival,agriculture ensures the construction and vigorous development of country’s national economy,it is the foundation of all industries.With the rapid development of modern Internet technology,new requirements are put forward for the innovation of agricultural technology information service.Traditional agriculture exists many problems,such as the lack of close connection of resource supply and the poor information dissemination,it must to be solved urgently.Compared with other ecommerce products,agricultural products have geographical and shelf-life restrictions,the traditional e-commerce recommendation model has poor recommendation effect on agricultural products.Therefore,the sales recommendation service for agriculture has great research value.Taking the supply side farmers and demand side sellers in the agricultural sales scenario as the target users,this thesis designs and implements an agricultural product supply and demand docking system including popular recommendation,online recommendation,offline recommendation,and publish-subscribe services by using the technologies of knowledge graph and graph neural network,so as to break the information barrier and provide convenient and practical services for farmers and sellers’ supply and demand docking.The specific contents of this thesis are as follows:(1)According to the characteristics and related elements of agricultural products,this thesis constructs Crop knowledge graph under the scenario of agricultural product docking recommendation,and puts forward a translation and graph convolution based knowledge graph embedding(TGC-KGE)model to learn the entity embedding representation in the Crop graph,and then mine the correlation between entities.Through comparative experiments,the effectiveness of TGC-KGE for node embedding learning on graph is verified.(2)Aiming at the agricultural sales recommendation scenario,the User-Crop dataset is constructed by using the constructed Crop knowledge graph and user interaction data,and a multi-task recommendation based network representation learning(KG-MTNR)model is proposed.The model introduces the agricultural product knowledge graph as external auxiliary information to alleviate the problems of sparse data and cold start in recommendation.combined with the attention mechanism to learn the weight coefficient,the neighborhood information is distinguished.Multi-task learning used to jointly optimize the node embedding.Comparative experiments verify the effectiveness of KG-MTNR model in recommendation task and node classification task.(3)Based on TGC-KGE and KG-MTNR models,the agricultural product supply and demand docking recommendation system is designed and implemented.The system is built on the Spring Boot framework.According to the characteristics of user groups,a convenient and practical agricultural product supply and demand docking recommendation system is designed and implemented.It is divided into different recommendation modules,such as popular recommendation,online dynamic recommendation,offline recommendation,publishsubscribe models.The system test shows that the system has complete functions,good performance and interactive experience.