Search papers, labs, and topics across Lattice.
This paper identifies and analyzes "entity interference" as a significant, previously overlooked source of catastrophic forgetting in Continual Knowledge Graph Embedding (CKGE). They demonstrate that the introduction of new entities can negatively impact the performance on previously learned facts, leading to an overestimation of CKGE methods' performance by up to 25% under standard evaluation protocols. To address this, the authors propose a corrected CKGE evaluation protocol and a new catastrophic forgetting metric tailored to CKGE, providing a more accurate assessment of model performance.
CKGE benchmarks overestimate performance by up to 25% because they fail to account for "entity interference," a newly identified phenomenon where embeddings of new entities disrupt previously learned relationships.
Knowledge Graph Embeddings (KGEs) support a wide range of downstream tasks over Knowledge Graphs (KGs). In practice, KGs evolve as new entities and facts are added, motivating Continual Knowledge Graph Embedding (CKGE) methods that update embeddings over time. Current CKGE approaches address catastrophic forgetting (i.e., the performance degradation on previously learned tasks) primarily by limiting changes to existing embeddings. However, we show that this view is incomplete. When new entities are introduced, their embeddings can interfere with previously learned ones, causing the model to predict them in place of previously correct answers. This phenomenon, which we call entity interference, has been largely overlooked and is not accounted for in current CKGE evaluation protocols. As a result, the assessment of catastrophic forgetting becomes misleading, and CKGE methods performance is systematically overestimated. To address this issue, we introduce a corrected CKGE evaluation protocol that accounts for entity interference. Through experiments on multiple benchmarks, we show that ignoring this effect can lead to performance overestimation of up to 25%, particularly in scenarios with significant entity growth. We further analyze how different CKGE methods and KGE models are affected by the different sources of forgetting, and introduce a catastrophic forgetting metric tailored to CKGE.
