ScholarSum: Student-Teacher Abstractive Summarization via Knowledge Graph Reasoning and Reflective Refinement
Researchers have introduced ScholarSum, a novel framework designed to improve abstractive summarization of scientific literature. This system employs a student-teacher approach, utilizing a hierarchical knowledge graph to capture the document's global logic and themes. A student model generates an initial draft, which is then refined by a teacher-like reviewer that identifies and corrects unsupported content through iterative retrieval and rewriting. Experiments indicate that ScholarSum surpasses existing methods in both completeness and factual consistency. AI
IMPACT This framework could significantly improve the efficiency and accuracy of understanding scientific literature for researchers.