Researchers have introduced Lineage, a new benchmark designed to evaluate AI scientific literature retrieval systems. Unlike existing benchmarks that focus on topical relevance or citation links, Lineage specifically assesses a system's ability to identify papers that use a prior method as an empirical comparison baseline. This is crucial for research agents that synthesize literature and plan experiments. The benchmark was constructed using data from 17 top AI venues between 2018 and 2025, containing over 67,000 paper nodes and 164,000 lineage edges, and includes 80,000 multi-facet retrieval queries. Initial experiments indicate that empirical lineage retrieval remains a challenging task, with performance significantly decreasing as the lineage distance increases. AI
IMPACT This benchmark could improve the ability of AI research agents to synthesize information and plan experiments by better understanding the empirical relationships between scientific papers.
RANK_REASON The item describes a new academic paper introducing a benchmark for AI scientific literature retrieval. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
- alphaXiv
- arXiv
- CatalyzeX Code Finder for Papers
- Connected Papers
- CORE Recommender
- DagsHub
- Gotit.pub
- Hugging Face
- Influence Flower
- Lineage
- Lineage-Constructor
- Litmaps
- ScienceCast
- scite Smart Citations
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →