GRAG: Generic Response-Augmented Generation Framework for Personalized Conversational Systems

The Generic Response-Augmented Generation (GRAG) framework has been introduced to enhance personalized conversational agents by decoupling content grounding from personalization, addressing computational challenges in resource-constrained settings. GRAG utilizes offline, generic responses from large language models (LLMs) to guide the fine-tuning of smaller, task-specific models, resulting in significant performance improvements on benchmark datasets, achieving up to 47% higher ROUGE-2 and 36% higher BLEU scores compared to existing methods. This framework provides a scalable approach for developing grounding-aware conversational systems that maintain contextual relevance while ensuring personalized interactions.