Abstract:
Entity Alignment (EA) involves identifying entities across two knowledge bases that represent the same real-world entity. This task is crucial for the automated integration of multiple Knowledge Graphs (KG) thus enriching the knowledge. Recently, embedding methods based on KG have become predominant in EA techniques. These methods project entities into a lower-dimensional space and align them by evaluating their similarities. However, the classification and alignment of entities
between two KG remain complex. This article evaluates the performance of various classifiers across multiple aspects of entity embedding features, applicable to both source and target data in binary classification processes for EA. Our experiments indicate a consistent range in the F1-score and accuracy, particularly when dealing with imbalanced data and changes in the dimensions of embeddings. This observation suggests that future research may need to focus on developing more robust classification algorithms.