What it is
Generates academic surveys from a methods-entity knowledge graph instead of flat paper retrieval, improving structure, retrieval, and citation support. Generate literature surveys that trace method evolution, competition, and shared resources by building a methods-entity knowledge graph from papers and using it to retrieve evidence, plan outlines, and write grounded sections.
Key results
- Graph-aware retrieval nearly doubles recall vs flat lexical on core diagnostic
- 80.6% of 160 generated sentences had full or partial citation support
- Semantic evidence gate reached AUC 0.75
Shows survey generation can be made more evolution-aware and better grounded by retrieving over method-level relations instead of flat document sets.