This paper proposes a discourse-aware hierarchical framework leveraging Rhetorical Structure Theory (RST) to address the limitations of existing long-document question-answering systems, which fail to capture discourse structure that facilitates human comprehension. This framework transforms discourse trees into sentence-level representations and connects structural and semantic information using an LLM-based node representation. Key innovations include specialized discourse parsing for long documents, LLM-based discourse relationship node enrichment, and structure-based hierarchical retrieval. Experiments on QASPER, Quality, and NarrativeQA demonstrate consistent performance improvements over existing approaches, demonstrating that discourse structure integration significantly enhances question-answering performance across a variety of document types.
