Comparative Analysis of Large Language Models in Interpreting Publication Research Ethics: Insights from COPE Forum Case Studies.

Abstract

BackgroundCommittee on Publication Ethics (COPE) serves as one of the key resources for journal editors navigating ethical challenges. This study evaluated two large language models (LLMs) in interpreting real-world publication ethics cases compared to COPE member responses.MethodsEleven anonymized COPE cases were selected. Gemini 2.5 Flash and DeepSeek-V3.2 provided editorial advice for each case. Responses were scored using a structured rubric supplemented by descriptive case-wise comparisons relative to COPE responses.ResultsBoth LLMs generated detailed responses for all cases, with similar domain scores. LLMs largely aligned with COPE, often adding actionable steps like contacting secondary authors or updating policies. However, they occasionally omitted specific COPE suggestions or contradicted guidance.ConclusionLLMs demonstrate competence in interpreting publication research ethics and, when paired with appropriate prompt-engineering training, can offer editors structured, actionable advice. However, their occasional contradictions and inability to fully replicate nuanced human judgment necessitate careful human oversight.