Research On Recommendation Of Fruit Commodity Based On Knowledge Graph

With the rapid development of information technology,online fruit purchase has become a new trend.At present,fruit has become the agricultural product category with the highest sales on the mainstream e-commerce platform.In the existing ecommerce recommendation systems,fruits are mostly regarded as ordinary commodities,ignoring the impact of attribute information such as origin,shelf life and price on users’ purchase.The final recommendation effect is not ideal and the user experience is poor.In order to make full use of the relevant attribute information of fruit commodities and improve the recommendation effect,aiming at the problems of incomplete utilization of information in the existing knowledge graph recommendation methods,this dissertation uses the relevant technology of knowledge graph to construct the knowledge graph of fruit commodities;On this basis,a fruit commodity recommendation model based on knowledge graph(FR-KG)is proposed.The main research work of this dissertation is as follows:(1)The Neo4 j graph database is used to construct the knowledge graph of fruit commodities.Obtain real data related to fruit commodities through the fruit column on Jing Dong website,and then clean and preprocess them;Then,using Neo4j’s definition of many different entity types and relationship types,construct the knowledge graph of fruit commodities;On this basis,using the Trans R knowledge representation learning model,the constructed fruit commodity knowledge graph is vectorized and mapped into a low dimensional dense vector space.(2)In order to alleviate the problem of incomplete information utilization in the existing knowledge graph recommendation methods,a FR-KG recommendation model is proposed.In the knowledge graph,the representation of each node is updated by aggregating neighbor nodes,and the attention mechanism is used to assign different weights to neighbor nodes with different importance.It integrates the processing of the connection path between the user and the fruit commodity in the knowledge graph,uses the multi-layer perceptron to predict the possibility of interaction between the user and the target fruit commodity,and recommends it to the user according to the possibility.(3)In order to improve the effect of aggregating neighborhood information,this dissertation proposes a new neighborhood aggregation method Double-End Aggregator based on GCN Aggregator and Graph Sage Aggregator.For the connection path between users and fruit commodities,the path is obtained by random walk,processed by bidirectional cyclic network,and the embedding representation of the path is obtained.The connection path is helpful to reveal the user’s preference for fruit commodities,which not only helps to improve the recommendation performance of the whole model,but also provides an explanation behind the recommendation.(4)Verification and evaluation,this dissertation compares FR-KG recommendation model with traditional recommendation model and several knowledge graph recommendation models to verify the effectiveness of FR-KG recommendation model.This dissertation explores the attention mechanism in FR-KG recommendation model and the impact of the number of layers of aggregation neighbor nodes on the recommendation performance of the model,compares three aggregation methods: GCN Aggregator,Graph Sage Aggregator and Double-End Aggregator,and verifies the effect of the aggregation method proposed in this dissertation.The shortcomings of this dissertation are pointed out,and the future research on recommendation based on knowledge graph is prospected.