Towards a Principled Evaluation of Knowledge Editors

TL;DR

This paper critically examines the evaluation methods for knowledge editors, revealing that different metrics and methodologies can significantly alter editor rankings and highlighting issues with current string matching evaluation approaches.

Contribution

It provides a comprehensive analysis of evaluation robustness for knowledge editors and demonstrates the impact of methodological choices on editor ranking and assessment.

Findings

Different evaluation metrics lead to different editor rankings.

Evaluation methodologies can bias the perceived effectiveness of editors.

String matching methods may produce false positives in knowledge editing evaluation.

Abstract

Model editing has been gaining increasing attention over the past few years. For Knowledge Editing in particular, more challenging evaluation datasets have recently been released. These datasets use different methodologies to score the success of editors. Yet, it remains under-explored how robust these methodologies are and whether they unfairly favor some editors. Moreover, the disruptive impact of these editors on overall model capabilities remains a constant blind spot. We address both of these problems and show that choosing different metrics and evaluation methodologies as well as different edit batch sizes can lead to a different ranking of knowledge editors. Crucially we demonstrate this effect also on general language understanding tasks evaluated alongside the knowledge editing tasks. Further we include a manual assessment of the string matching based evaluation method for knowledge editing that is favored by recently released datasets, revealing a tendency to produce false positive matches.