Relation Also Knows: Rethinking the Recall and Editing of Factual Associations in Auto-Regressive Transformer Language Models

2024-08-27

arXiv.org

AI Summary

This work addresses the overgeneralization problem in autoregressive Transformer language models during factual editing, which arises from excessive reliance on subject-centric explanations. We propose a relation-focused knowledge editing paradigm. Methodologically, we first identify the dominant role of relational information in factual recall, introduce the Relation-Specificity (R-Specificity) metric for quantitative evaluation, and design a relation-localization mechanism leveraging attention patterns and MLP activation profiles. Building upon this, we develop a single-fact editing framework integrating targeted weight tuning with constraint-based optimization. Evaluated on our newly constructed R-Specificity benchmark, our approach significantly mitigates overgeneralization while preserving high edit fidelity and generalization capability. Our contribution lies in shifting beyond the conventional subject-centric paradigm, establishing a relation-driven, interpretable, fine-grained, and low-side-effect knowledge editing methodology.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Reasoning under Uncertainty: Relational Probabilistic Models
• Knowledge Representation and Reasoning: Reasoning with Beliefs

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The storage and recall of factual associations in auto-regressive transformer language models (LMs) have drawn a great deal of attention, inspiring knowledge editing by directly modifying the located model weights. Most editing works achieve knowledge editing under the guidance of existing interpretations of knowledge recall that mainly focus on subject knowledge. However, these interpretations are seriously flawed, neglecting relation information and leading to the over-generalizing problem for editing. In this work, we discover a novel relation-focused perspective to interpret the knowledge recall of transformer LMs during inference and apply it on single knowledge editing to avoid over-generalizing. Experimental results on the dataset supplemented with a new R-Specificity criterion demonstrate that our editing approach significantly alleviates over-generalizing while remaining competitive on other criteria, breaking the domination of subject-focused editing for future research.

Problem

• Interpreting knowledge recall in transformer LMs
• Avoiding over-generalizing in knowledge editing
• Incorporating relation information in model editing

Innovation

• Relation-focused knowledge editing
• Avoids over-generalizing problem
• Uses R-Specificity criterion