Recommendation System Based On Graph Convolutional Network And Knowledge Graph

The recommendation system as a popular technology in the field of big data.In recent years,it has played an important role in e-commerce,film and television music,news,social networking and other platforms.It meets the user’s personalized needs and effectively solves today’s Internet era The problem of information generated can provide users with more suitable high-quality content.At present,in the research content of the recommendation algorithm,most researchers will focus on how to tap the association between users and projects,thereby ignoring the social associated information in the user group,and there are potential potential to excavate the project through the knowledge graph.Attributes,then modeling with user features,leads to the entity that enriches its own expression while incorporating redundant information,thereby increasing the noise of the model.Therefore,this article will reasonably use the knowledge graph,play the role of social networks,build a model that can realize social recommendations,and use it in the product recommendation system.The main tasks of this article are as follows:(1)In order to verify the effectiveness of using graph convolutional network GCN and knowledge graph KG in recommendation algorithms,this paper constructs a KGN model and conducts experiments on two public datasets,comparing it with the baseline model.It is verified that the graph convolutional network recommendation algorithm based on knowledge graph is more efficient than traditional recommendation algorithms,and has a significant improvement in prediction accuracy.(2)This article constructs a recommendation model SKGAN based on social information and knowledge graph attention networks.A method was proposed to construct two data processing channels for the model,separating the social recommendation propagation network KGSR and the project knowledge graph IKG,and outputting the processed user feature vectors and project feature vectors respectively,reducing model noise.At the same time,in order to leverage the similarity between users,the user project bipartite graph is combined with social networks in KGSR.Firstly,the neighborhood aggregation method in GCN is used to integrate user features into project features.Then,social relationships are used to propagate rich user information,allowing users to capture the interaction records between other social users and projects,thus achieving the goal of social recommendation.Finally,experiments were conducted on four public datasets to verify the model’s good feature extraction and model generalization capabilities.At the same time,the model structure was adjusted for experiments,and the most efficient model structure was tested.(3)This article incorporates an attention mechanism into the model to control the length of the network propagation path.When aggregating neighborhood entity features in the graph convolutional network in the model,the user’s preference for the relationship between project entities can be calculated,and the weight between users and social user entities can be calculated to determine whether the relationship between users is close.This method can better control the effectiveness of feature extraction during convolution and improve prediction accuracy.(4)Design and implement a product recommendation system based on the SKGAN model in this article.The system adopts the B/S architecture and is based on the mainstream technology framework,including background management,clients,and recommendation modules.The background management side can manage the contents of employees,product information,etc.The recommended module can train the recommended model of this article to predict the interaction probability of users and the project.After the user logs in,the system will call the recommended model for this user generate.The product recommendation list,finally users can add the product to the product to join the shopping cart,buy an order and other operations on the client.