【摘要】 【目的/意义】大语言模型（LLMs）对学术研究中的文献自动摘要具有很大的帮助，但其生成结果的质量存在较大争议。本研究提出了评价模型生成摘要质量的评价体系，作为改善大语言模型用于学术研究中的重要依据。【方法/过程】首先，基于访谈内容构建初步评价模型，通过专家函询的方法修订后，形成涵盖语言连贯性、内容准确性以及符合伦理原则等方面的评价体系。最后，就提出的评价体系对主流大语言模型用于科学研究生成摘要结果进行实证分析。【结果/结论】研究结果显示，所选用的大语言模型生成的摘要都存在不足，尤其在语言连贯性和内容准确性上有较大的出入，在伦理原则的满足上也仍有较大的改进空间。【创新/局限】本研究提出了双重评价框架，结合了过程评价与结果评价，为提升大语言模型在学术文献方面的生成能力提供了实际性建议。更多还原

【Abstract】 【Purpose/significance】Large Language Models(LLMs) offer promising support for automatic summarization of academic literature, yet the quality of their outputs remains highly debated. This study proposes a comprehensive evaluation system for assessing the quality of LLM-generated abstracts, aiming to provide a theoretical and practical basis for improving their application in academic research.【Method/process】A preliminary evaluation framework was constructed based on expert interviews, and subsequently refined using the Delphi method. The final framework encompasses dimensions such as linguistic coherence, content accuracy, and adherence to ethical principles. Empirical analysis was then conducted to assess the performance of mainstream LLMs in generating scientific abstracts using the proposed system.【Result/conclusion】The findings indicate that while LLMs demonstrate potential, their generated abstracts exhibit notable deficiencies, particularly in linguistic coherence and content accuracy. Ethical compliance also remains an area requiring substantial improvement.【Innovation/limitation】This study proposes a dual evaluation framework integrating both process-based and outcome-based assessments, offering practical insights into enhancing the generative performance of LLMs in the academic domain.更多还原