【作 者】：
尹 珺 瑶
【关 键 词 】：
高等教育； 学位论文； 写作辅助能力； 大模型测评； 人—技关系理论
【栏 目】：
课程与教学
【中文摘要】：
大规模语言模型引发了高等教育领域的深层变革。研究基于人—技关系理论，聚焦人文社会学科研究生学位论文写作辅助场景，构建包含知识、逻辑和应用能力三大维度15项任务的能力框架，及包含2,990道题项的测试数据集，以ChatGPT、Qwen-72B-Chat和Qwen-14B-Chat为样本模型进行写作辅助能力测评。研究发现：大模型在人文社会学科学位论文写作辅助场景下表现出较好的知识型能力与应用型能力，逻辑型能力也存在较强的发展潜力，Qwen-72B-Chat在高等教育与科研垂类表现较好。研究针对大模型在科研方法、社会关系和因果关系三方面的不足，探索了有效的补充指令改进策略，从学习者、研发者、管理者和研究者视角提出了建议，以科学规范与指引学生的技术行为，推动高等教育与科研垂类大模型的建设。
【英文摘要】：
Large-scale language models(LLMs) have triggered profound transformations in the field of higher education. Based on the theory of human-technology relationship, this study focused on the scenario of writing assistance for postgraduate students' dissertations in humanities and social sciences, constructed a competency framework containing 15 tasks in three dimensions of knowledge, logic and application ability, and a test dataset containing 2,990 items. ChatGPT, Qwen-72B-Chat and Qwen-14B-Chat were used as sample models to assess the writing assistance ability. It is found that the LLMs demonstrate better knowledge-based and application-based abilities in humanities and social sciences dissertation writing assistance scenarios, and there was a strong development potential for logic-based abilities. Qwen-72B-Chat performed better in higher education and scientific research vertical category. This study explores effective strategies to improve the supplementary instruction for the large model with regard to its shortcomings in three aspects of scientific research methods, social relations and causality, and puts forward suggestions from the perspectives of learners, developers, administrators and researcher to scientifically regulate and guide students' technical behaviors, and to promote the construction of large models in higher education and scientific research vertical category.