Claim Networks: Typed Citations for Scientific Literature
A new representational pattern called the claim network reifies cross-document references as typed claims, each carrying source, target, claim text, and a four-class stance label. The construction pipeline is applicable to any corpus of scholarly inter-referencing documents and was instantiated on 127 papers in 3D point cloud semantic segmentation, producing 8,260 typed claims. Three downstream tasks—retrieval signal augmentation, aggregated-stance summarization, and citation graph analysis—demonstrate the network's capabilities. The work addresses the limitation of standard knowledge graphs that encode topology but not stance, losing evaluative content for community-level queries.
Key facts
- Claim network reifies cross-document references as typed claims.
- Each claim has source, target, claim text, and four-class stance label.
- Pipeline applicable to any corpus of scholarly inter-referencing documents.
- Instantiated on 127 papers in 3D point cloud semantic segmentation.
- Produced 8,260 typed claims.
- Three downstream tasks: retrieval signal augmentation, aggregated-stance summarization, citation graph analysis.
- Standard representation collapses evaluative relation into untyped edge.
- Addresses loss of stance content in knowledge graphs.
Entities
Institutions
- arXiv