A Framework for Evaluating Entity Alignment Impact on Downstream Knowledge Discovery

Entity alignment (EA) is a crucial process in integrating data from multiple sources, facilitating Knowledge Discovery (KD). Despite advances in EA techniques, selecting the appropriate algorithm for downstream KD tasks remains challenging due to several issues. These issues include domain entities alignment difficulties, the impact on KD tasks, and bias in data distribution. This paper presents a framework to address these challenges by providing a systematic approach to evaluate the impact of different EA algorithms based on three critical aspects: quality of alignment, information retrieved through alignment, and information imbalance or bias introduced through alignment. Our framework enables users to make informed decisions about algorithm selection, ensuring reliable, effective, and balanced KD. We demonstrate the application of the framework using a digital humanities case study, where the KD task involves enriching information about colonial collections. The choice of such a sensitive and historically imbalanced use-case allows us to highlight how the proposed framework helps identify suitable algorithms and to emphasis the importance of understanding the propagated information biases introduced through data alignment.