Uncovering Entity Identity Confusion in Multimodal Knowledge Editing

TL;DR

This paper investigates a systemic failure in multimodal knowledge editing models called Entity Identity Confusion, revealing how models confuse original and new entities after editing, and proposes strategies to mitigate this issue.

Contribution

The paper identifies Entity Identity Confusion as a key failure in multimodal knowledge editing and introduces EC-Bench for diagnosis, offering insights and mitigation strategies.

Findings

EIC causes models to confuse original and new entities after editing.

Existing methods fail to distinguish between I-E binding and E-E relations.

Constraining edits at the I-E stage reduces EIC significantly.

Abstract

Multimodal knowledge editing (MKE) aims to correct the internal knowledge of large vision-language models after deployment, yet the behavioral patterns of post-edit models remain underexplored. In this paper, we identify a systemic failure mode in edited models, termed Entity Identity Confusion (EIC): edited models exhibit an absurd behavior where text-only queries about the original entity's identity unexpectedly return information about the new entity. To rigorously investigate EIC, we construct EC-Bench, a diagnostic benchmark that directly probes how image-entity bindings shift before and after editing. Our analysis reveals that EIC stems from existing methods failing to distinguish between Image-Entity (I-E) binding and Entity-Entity (E-E) relational knowledge in the model, causing models to overfit E-E associations as a shortcut: the image is still perceived as the original entity, with the new entity's name serving only as a spurious identity label. We further explore potential mitigation strategies, showing that constraining edits to the model's I-E processing stage encourages edits to act more faithfully on I-E binding, thereby substantially reducing EIC. Based on these findings, we discuss principled desiderata for faithful MKE and provide methodological guidance for future research.