KEditVis: A Visual Analytics System for Knowledge Editing of Large Language Models

2026-03-31

AI Summary

This work addresses the challenge that large language models frequently generate factual errors in question answering, and existing knowledge editing methods struggle to identify optimal editing layers and lack transparency. To bridge this gap, the paper introduces visualization-driven analysis into knowledge editing for the first time, presenting an interactive system that enables users to explore the editing process at fine-grained levels, trace cross-layer influences, diagnose failure cases, and select optimal editing strategies. By integrating multidimensional evaluation metrics with mechanisms for analyzing layer-wise impact, the system substantially enhances the transparency, controllability, and efficacy of knowledge editing. Through scenario-based analyses, expert interviews, and user studies, the authors demonstrate the system’s effectiveness in improving editing performance, usability, and its utility in supporting algorithm development.

Abstract

Large Language Models (LLMs) demonstrate exceptional capabilities in factual question answering, yet they sometimes provide incorrect responses. To address this issue, knowledge editing techniques have emerged as effective methods for correcting factual information in LLMs. However, typical knowledge editing workflows struggle with identifying the optimal set of model layers for editing and rely on summary indicators that provide insufficient guidance. This lack of transparency hinders effective comparison and identification of optimal editing strategies. In this paper, we present KEditVis, a novel visual analytics system designed to assist users in gaining a deeper understanding of knowledge editing through interactive visualizations, improving editing outcomes, and discovering valuable insights for the future development of knowledge editing algorithms. With KEditVis, users can select appropriate layers as the editing target, explore the reasons behind ineffective edits, and perform more targeted and effective edits. Our evaluation, including usage scenarios, expert interviews, and a user study, validates the effectiveness and usability of the system.

Problem

• Large Language Models
• Knowledge Editing
• Model Transparency
• Editing Strategies
• Factual Accuracy

Innovation

• visual analytics
• knowledge editing
• large language models
• interactive visualization
• model interpretability