Improving Factuality for Dialogue Response Generation via Graph-Based Knowledge Augmentation
This addresses the issue of factual inaccuracies in dialogue systems, which is crucial for reliable AI assistants, though it is incremental as it builds on existing graph knowledge-augmentation methods.
The paper tackles the problem of hallucination in dialogue response generation by proposing two graph-based knowledge augmentation frameworks, which improve factuality scores by 3.47% on OpendialKG and 3.12% on HybriDialogue compared to state-of-the-art baselines.
Large Language Models (LLMs) succeed in many natural language processing tasks. However, their tendency to hallucinate - generate plausible but inconsistent or factually incorrect text - can cause significant problems in certain tasks, including response generation in dialogue. To mitigate this issue, we propose two novel graph knowledge-augmented frameworks, Dialogue Response Generation via Textualised Graphs (TG-DRG) and Graph-Aware Dialogue Response Generation (GA-DRG), which combine reasoning-guided dialogue reformulation, dialogue sense knowledge selection, and graph-enhanced response generation to improve the factuality of dialogue responses. To evaluate the factuality of generated responses, we propose a dialogue fact score that addresses the limitations of existing fact-score methods in dialogue settings, providing a more reliable assessment of factual consistency. We evaluate our methods using different baselines on the OpendialKG and HybriDialogue datasets. Our methods noticeably improve factuality compared to other graph knowledge-augmentation baselines, including the state-of-the-art G-retriever, achieving improvements of 3.47% on OpendialKG and 3.12% on HybriDialogue in terms of dialogue fact score. The code will be released on GitHub.