Tag
The paper proposes SF-Re2G, a method that improves document-grounded dialogue systems by leveraging document structure to enhance retrieval, reranking, and generation. It validates on Chinese and English datasets.