Claim Networks Boost RAG by Typing Citations with Stance

Knowledge graphs of scientific literature typically treat citations as untyped edges, losing the stance and content of each reference. In 'Reading Between the Citations,' Ning Ding and colleagues propose the claim network, where each cross-document reference is reified as a typed claim carrying source, target, claim text, and a four-class stance label drawn from citation-intent literature (positive, negative, neutral, none). This transforms flat citation graphs into rich semantic networks that capture how works are actually received. The authors build a pipeline and instantiate it on 127 papers in the hot AI subfield of 3D point cloud semantic segmentation, producing 8,260 typed claims.

Three downstream task families demonstrate the network's value: retrieval signal augmentation (using claim stances to rerank RAG results), aggregated-stance summarization (e.g., automatic meta-reviews of a paper's reception), and topological analytics (e.g., identifying hubs of contradictory claims). Head-to-head against standard RAG baselines, the claim network consistently improves retrieval precision—the authors emphasize the gain comes from using the right intermediate representation. This approach enables previously impossible queries like 'Which papers are most negatively cited?' and opens scalable extraction of nuanced relationships from citation graphs for AI-assisted research tools.