生成式人工智能如何辅助研究生知识生产 ——基于23万篇硕士学位论文的大模型语义识别研究-清华大学教育研究

文章信息/Info

Title:: How Does AIGC Empower Graduate-Level Knowledge Production ——A Semantic Analysis Based on 230,000 Master’s Theses Using Large Language Models

Author(s):: WANG Chuan-yi¹ QIANG Hao² SHEN Dong-ling² LIU Chang² ZHOU Gang³; 1.School of Education, Tsinghua University;2.Tongfang Knowledge Network Technology Co., Ltd.;3.College of Electronic Engineering, Naval University of Engineering

摘要:: 本研究基于23万篇硕士学位论文进行大模型语义识别研究,发现硕士研究生使用AIGC的程度在逐年上升,2022年至2024年间的使用比例从不足2%增长至超过5%。论文不同部分中AIGC的使用情况,可分为初步探索型、功能偏好型、论证支持型和融合使用型四种类型。不同类型学科硕士研究生的AIGC使用方式具有显著不同。在各类学科中“功能偏好型”均为使用最多的方式,初步探索型均占10%左右的比例。此外,软学科学生更多采用论证支持型的方式,应用硬学科更多采用融合使用型的方式。

Abstract:: This study conducts a semantic recognition analysis using large-scale language models on a corpus of 230,000 master’s theses to explore the role of AIGC (Artificial Intelligence Generated Content) in graduate-level knowledge production. The findings indicate a steady increase in AIGC usage among master’s students, with the proportion rising from less than 2% in 2022 to over 5% by 2024. Patterns of AIGC application within theses can be categorized into four types: exploratory use, function-preference use, argument-support use, and integrated use. Significant disciplinary differences are observed in AIGC usage styles. Across all disciplines, the function-preference type is the most prevalent, while exploratory use accounts for approximately 10% consistently. Students in soft disciplines are more likely to adopt argument-support approaches, whereas those in applied hard disciplines tend to employ integrated usage patterns.