Research On Knowledge Graph Recommendation Algorithm Based On Graph Neural Network

As one of the main means of providing personalized information to users,recommendation systems often rely on multiple aspects such as accuracy,effectiveness,and interpretability to determine their performance.Therefore,we construct a knowledge graph for the objects to be recommended,and combine the powerful understanding and cognitive abilities of graph neural networks to improve the performance of recommendation systems.Although existing recommendation models have achieved good results,they have not yet provided detailed classification and modeling of the recommended items,and have not fully utilized the unity and specificity of different types of edges in the knowledge graph,leading to poor interpretability of the recommendations.Therefore,this article mainly updates and improves existing recommendation algorithms from the perspectives of interpretability and effectiveness,and proposes the KHANrec recommendation model based on the heterogeneous graph neural network HAN.The main contents of this article include:1.Building a knowledge graph based on the moivelens-2k and goodbooks-10k datasets.The recommended project information was selected and preprocessed to build a heterogeneous knowledge graph.Suitable nodes were selected to construct isomorphic graphs and feature vectors as meta-paths.2.Implementing and optimizing the KHANrec recommendation algo rithm model.This paper combines the heterogeneous classification graph neural network HAN with the recommendation algorithm,and trains the s hared and unique features separately at different levels for item similarity recommendations.The feature types are refined for different scenarios,i mproving the interpretability and accuracy of the recommendation algorit hm to a certain extent,and optimizing system performance.3.Implementing the recommendation of rating and predicting the click-through rate of recommended items.Feature analysis of users and items is carried out,and a filtering strategy is used to provide personalized top 5,10,and 20 item recommendations and hot recommendations for each user.4.Conduct multiple comparative experiments based on the recomme ndation model proposed in this article,including experiments comparing between meta-paths,different datasets,and other models.Through multid imensional comparisons,explore the recommendation effectiveness of the model proposed in this article.The final experimental results show that the KHANrec model has an MSE of only 15%for rating prediction on the movielens-2k dataset,and the accuracy of TOP-5 recommendations reaches 87%.On the goodbooks-10k dataset,the MSE is only 7.8%,and the accuracy of TOP-5 recomme ndations can reach 95%.This indicates that the proposed model has high accuracy and effectiveness.The introduction of multiple meta-paths has a lso improved the interpretability of the model to some extent.