[2Win5] Poster session 2

[2Win5-20] Framework for evaluating Knowldge Graph Completion based on the comparison with Knowledge Graph at different time slices

〇Ryoma Kobayashi¹, Masaharu Yoshioka¹ (1.Hokkaido University)

Keywords:Knowledge Graph, Knowledge Graph Completion

Knowledge Graph (KG) is a graph-structured knowledge base that stores real-world information and is used in various applications. Since KG are generally incomplete, KG Completion (KGC) has been extensively studied, and Dynamic KGC approaches have been proposed to account for the evolving nature of KG. However, when focusing on facts that ideally should have existed earlier but were missing, existing frameworks do not distinguish between those that will eventually be added in the future and those that will not. We hypothesize that making this distinction could provide a new perspective for interpreting system characteristics. To address this, we propose an evaluation framework that leverages multiple evaluation datasets, each created using different methods, to more properly assess missing facts. As a preliminary examination, we conducted entity–attribute link prediction and confirmed the potential to interpret system characteristics from alternative perspectives.