This paper details DS@GT ARC's participation in the CLEF 2026 LongEval Task 4, focusing on Retrieval-Augmented Generation (RAG) systems. The research highlights a discrepancy between standard natural language evaluation metrics and the crucial aspect of citation integrity in RAG QA. By employing a corrective pipeline with Corrective RAG (CRAG) and CiteFix, the study found that while frontier models excel in answer relevance and fluency, they do not always strictly adhere to cited sources. The proposed pipeline enforces strict entailment of generated claims to cited material, marginally improving citation faithfulness and answer grounding, suggesting a need for evaluation metrics that prioritize strict answer grounding for trustworthy RAG QA. AI
IMPACT This research suggests a need for improved evaluation metrics in RAG systems to ensure factual grounding and citation integrity, potentially influencing future development of trustworthy AI.
RANK_REASON The cluster contains a research paper detailing a new method for evaluating and improving RAG QA systems.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
