How Much Structure Do LLMs Need? Evaluating LLMs for Bibliometric Cluster Description

In three linesStudy evaluating 6 LLM pipelines for generating bibliometric cluster descriptions. On 100 published analyses, LLMs produce semantically close descriptions to human versions but hallucinate references and fail to infer bibliometric structure alone. Optimal performance in hybrid workflow: algorithms define clusters, LLMs generate readable descriptions.

Summary generated by Claude — human-verified