Research And Application Of Knowledge Graph Based On Multi-Source Heterogeneous Data

As an important branch of artificial intelligence,knowledge graphs are a powerful factual representation that connects all different types of information together in a graph-based data structure to obtain a relational network,which is often regarded as a special kind of heterogeneous information network because it contains more than two types of nodes and relationships.Knowledge graphs use relationships as an important tool for analyzing problems and have been widely used in recent years in various fields such as recommendation systems,fraud detection,and social networks.Knowledge graphs,which can provide structural information and enable inference based on available graph data,have been a hot research topic in terms of how they can be better applied to various specific tasks.In this paper,we focus on knowledge graphs based on multi-source heterogeneous data and their applications.On the one hand,we explore how knowledge graphs can be used as auxiliary information to combine structural and semantic information with the interaction information of users and items in bipartite graphs to construct paths for inference,improve recommendation accuracy and enhance the interpretability of recommendation results,and use graph neural networks(GNN)to mine knowledge based on related research Based on the related research,a knowledge graph recommendation model based on user perception vectors is designed to explore the potential relationship between user vectors and higher-order information in the knowledge graph,to discover user interests at a deeper level,and to provide more useful information for the recommendation task.On the other hand,for graphs composed in fraud scenarios such as financial transactions and citation networks,whose data are characterized by multi-source heterogeneity,and inspired by the effectiveness of graph neural networks in learning graph-structured data,this paper extends the research work to the field of fraud detection.The main work done in this paper is as follows:(1)Recommendation methods that use only user item interaction information for recommendation suffer from data sparsity and cold start problems.In this paper,we propose a path-aware recommendation model based on knowledge graphs,PAKG,which introduces knowledge graphs as auxiliary information in the recommendation system to map items and their attributes into KG to understand the interrelationships between items.In addition,user information can also be integrated into the KG,which allows the relationships between users and items as well as user preferences to be captured more accurately.Not only does it generate a representation of paths by considering entity vectors and relationship vectors,but it also performs path-based reasoning to infer user preferences and sets up a path-following mechanism to measure the importance of different paths.In addition,the path is optimised for the noise problems that can be caused by too long a path.Experimental results on publicly available datasets demonstrate that the introduction of knowledge graphs in recommendation can achieve better recommendation results.(2)Among the existing recommendation methods based on knowledge graph,they are characterized by historical items that users have clicked on when capturing user vectors.Most of them focus more on exploring user-item and the relationship between items and entities in the knowledge graph.Although the knowledge graph is used as auxiliary information,the role of relevant higher-order information in the knowledge graph in characterizing user vectors is rarely considered.It also ignores some user’s potential preference information.This paper proposes a new user-aware vector-based knowledge graph recommendation method KGPA,which is a framework based on user-item and item-entity interactions in the knowledge graph,combined with graph neural networks and attention mechanisms to consider the higher-order information related to user nodes in the knowledge graph and discover user interests at a deep level.In addition,this paper uses a path-aware sampling strategy to reduce the impact of noisy entities to reduce computational costs.Experiments show that the method has significant advantages over other baseline methods.(3)Networks such as financial transaction scenarios,citation networks,and social networks also belong to heterogeneous information networks,and graph neural networks have also achieved good results on multi-node,multi-relationship type multi-source heterogeneous graph data classification tasks,therefore,this paper proposes a heterogeneous graph fraud detection model based on knowledge graph embedding to extend part of the work to the field of fraud detection to identify nodes with fraud potential and detection.Many existing fraud detection models based on meta-path approaches have the problem of only focusing on the nodes at the end of the meta-path and ignoring the information of the nodes in the middle of the meta-path when capturing node feature information,which will lead to the problem of information loss.Therefore,the model introduces the knowledge graph embedding method Rotat E as a meta-path internal aggregation encoder to obtain more comprehensive meta-path information by focusing on the intermediate nodes on the meta-path at the same time when capturing node information on the meta-path.Experimental results on different types of datasets show that the method achieves relatively good results when compared with a variety of existing fraud detection methods.