Benchmark RadarGraphInfer-Bench
GraphInfer-Bench evaluates LLMs on graph inference tasks where answers reside in no single node or path, covering five task types over six real-world graphs with 42,000 samples. Tasks include masked-node prediction, edge inference, theme description, outlier detection, and community partition.
- Released
- 2026-06-10
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Targets an open capability gap in graph understanding: inference over joint neighborhood structure. Useful for diagnosing weaknesses in LLMs and GNNs for tasks like fraud detection, drug repurposing, and recommendation.
Motivation
Graph analysis underlies many applications whose answers cannot be looked up in a single record or retrieved along a path: laundering rings, drug repurposing, user preference, and scientific theme are all inferred from a node together with its neighbourhood.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.