Research On Recommendation Algorithm Based On Knowledge Graph Ripple Propagation And Graph Neural Network

Recommendation systems have arisen as an efficient way for consumers to filter desired information from a variety of sources.In early research on recommendation algorithms,collaborative filtering-based recommendation systems were frequently employed in practice.As the subject of recommendation algorithms underwent further research,several academics discovered that this method often suffers from data sparsity and cold-start issues,which prevent better recommendations from being made.Many academics have improved the recommendation system by including some of the relevant data to fix this problem.This innovation has introduced a new stage of the recommendation algorithm,which will surely improve suggestion accuracy.As auxiliary information,knowledge graphs contain rich semantic properties that can be used to extract higher-order neighborhood information between entities and augment entity representation,hence overcoming the data sparsity problem and improving recommendation accuracy.The topological information in the knowledge network is strongly related to the information in the user-item interaction matrix in the recommendation algorithm.Deeper data associations are created by combining the two,making up for the sparseness of the initial data,and an efficient recommendation algorithm allows for the exploration of prospective user and item features.The research topic of integrating knowledge graphs into recommendation systems has been extensively used.Hybrid recommendation algorithms that combine knowledge graph embeddings and routes are currently effective and have emerged as the main trend in knowledge graph-based recommendation systems,which often use graph neural networks(GNNs)to simulate user preferences.Graph neural networks propagate higher-order information by pooling neighborhood information layer by layer to augment the user and item representations.Although the aggregation process of graph neural networks effectively addresses the drawbacks of embedding-based and path-based systems,there is still some potential for development,including not mining the knowledge graph’s data sufficiently deeply,not fully utilizing the entity and connection information there,and not being able to collect both higher-order user features and higher-order item features,enriching both the person representation and the object representation,but the current techniques only fully utilize the information at one end.They just consider the relationship between the user side or the item side and the entities in the knowledge graph and do not fully leverage the knowledge graph information.Thus,it presents a challenge to investigate the semantic linkage of user-side features and item-side features with entities in the knowledge graph as well as the mining of hidden information in the user-item history interaction records.Consequently,based on the aforementioned contemporary issues,this research performs an extensive analysis and suggests a recommendation model(KGCN-RN)that combines ripple propagation and graph neural networks based on the knowledge graph.The main contributions of this paper are:(1)In order to simultaneously get high-order user characteristics and high-order item features,enhance both the user representation and the object representation,The KGCN-RN model proposed in this paper automatically obtains potential knowledge graph relevant data by introducing knowledge graph information,enriches user representation and item representation at the same time,and uses the preference propagation of the Ripple Net model to enrich user feature representation,high-order modeling of user characteristics.It aggregates neighborhood information to augment item feature representation at the object side through the messaging mechanism of a graph neural network.Imagine the knowledge network as a power graph,in which the weighted relationship between entities represents the relationship’s influence on the level of user preference.The issue that the current recommendation approach does not completely extract features on the user side or the article side is fixed by this method.(2)In order to more precisely assemble very similar neighborhood data,the model employs important sampling to aggregate each layer of neighborhood content when sampling the weights between various associations associated to the entity.This is done according to the weight value acquired from high to low sampling a given number of neighborhood information,in order to generate a high-order item representation.Make the suggestion findings more precise in order to better integrate them into the recommendation system and increase its accuracy.By experimenting with the algorithm presented in this work on three distinct types of publicly available datasets and comparing the experimental findings with those of some existing models,the KGCN-RN model outperforms the current algorithms in click-through rate prediction,Top-K recommendation,and data sparsity experiments,demonstrating the usefulness of the KGCN-RN model proposed in this study.