ECPO: Evidence-Coupled Policy Optimization for Verifier-Audited Candidate Ranking
In brief
- ECPO generates Top-K rankings with verifiable evidence linked to candidates.
- Improves traceability and answer recovery on MAVEN-ERE and RAMS benchmarks.
- Maintains explicit trade-offs on standard metrics like NDCG.
This study introduces ECPO, a method to generate Top-K rankings with verifiable evidence linked to candidates. Results show improved traceability and answer recovery across multiple benchmarks while maintaining explicit trade-offs on standard metrics.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error