Knowledge-Aware Self-Correction in Language Models via Structured Memory Graphs

TL;DR

This paper introduces a lightweight, interpretable method for correcting factual errors in language model outputs using external structured memory graphs, without retraining the models.

Contribution

It proposes a novel knowledge-aware self-correction framework utilizing RDF-based memory graphs to improve factual accuracy of LLM outputs post-generation.

Findings

Effective correction of factual errors demonstrated on DistilGPT-2.

No retraining or fine-tuning required, enabling easy integration.

Promising results on simple factual prompts.

Abstract

Large Language Models (LLMs) are powerful yet prone to generating factual errors, commonly referred to as hallucinations. We present a lightweight, interpretable framework for knowledge-aware self-correction of LLM outputs using structured memory graphs based on RDF triples. Without retraining or fine-tuning, our method post-processes model outputs and corrects factual inconsistencies via external semantic memory. We demonstrate the approach using DistilGPT-2 and show promising results on simple factual prompts.