Knowledge Fusion Algorithm Based on Entity Relation Mining and Graph Neural Network

The changing application scenarios have resulted in an increased demand for data services that cannot be met by knowledge graphs from single sources alone. The rapid development of internet technology has led to an exponential increase in the variety and amount of available data, presenting a challenge during the graph construction stage. To address this challenge, multi-source knowledge graphs have emerged, allowing the integration of multiple data sources into one knowledge graph. However, different data sources may express the same objective thing differently, necessitating knowledge fusion to merge identical entities from various data sources and combine them into a single knowledge graph. This paper proposes a knowledge fusion algorithm based on entity relation mining and graph neural networks. Firstly, we construct a knowledge graph for each data source and bridge them using known common entities. Then, we mine the entity cluster features in the graph entities with modularity. Secondly, we use Graph Convolutional Networks (GCN) to extract vector representations of each entity’s structural features based on the obtained entity cluster features. Lastly, we calculate similarity between entities using their GCN-obtained vectors and merge identical ones.