clt-international

ISSN Number

 2708-9517

Abstracting/Indexing/Listing

MLA Directory of Periodicals

REAO: East Asian Studies Journals

EBSCO Education

DOAJ

ProQuest

Google Scholar

Semantic Scholar

ROAD

BASE

Helka Helsinki Library

Baidu Scholar

Ex Libris

Jouroscope

US Department of Commerce Research Library

Home Journal Index 2026-3

大语言模型的教育伦理评估研究——以海外国际中文教学机构评测为例

Download Full PDF

肖 锐

刘 玲

杨 蓉

云南大学,中国

 

张邝弋

北京语言大学,中国

 

摘要

为探究大语言模型( LLM)在国际中文教育场景中的内容输出倾向及其潜在伦理风险线索,本研究选取四类代表性 LLM——包括通用型模型( Copilot、 Claude、 LLaMa-2)与教育垂直型模型( Educhat),依托海外中文教学机构及学员展开情感分析与学段倾向性测评。实验结果显示: LLM 对海外中文教学机构的情感态度总体呈积极倾向(积极词 55%,中性词 41%,消极词 4%);不同 LLM 在对待不同学段学员时存在显著的内容输出差异,其中中学学段获得最高关注度( 143次),小学学段关注度最低( 103 次)。上述差异可为后续伦理风险识别与算法优化提供经验依据。建议加强舆情监测与训练数据优化,提升算法公平性与内容多样性,以增强 LLM 在国际中文教育场景中的适应性。

 

关键词

大语言模型,海外国际中文教学机构,教育伦理

 

Research on Educational Ethics Evaluation of the Large Language Models: A Case Study of Overseas International Chinese Language Teaching Institutions' Evaluation

 

Rui Xiao

Ling Liu

Rong Yang

Yunnan University, China

 

Kuangyi Zhang

Beijing Language and Culture University, China

 

Abstract

To explore the content output tendencies of large language models (LLMs) in the context of international Chinese education and their potential ethical risk cues, this study selected four representative LLMs—including general-purpose models (Copilot, Claude, LLaMa-2) and education-specific models (Educhat)—and conducted sentiment analysis and grade-level inclination assessments based on overseas Chinese teaching institutions and learners. The experimental results show that: LLMs generally exhibit a positive sentiment attitude toward overseas Chinese teaching institutions (positive words 55%, neutral words 41%, negative words 4%); different LLMs show significant differences in content output when addressing learners of different educational stages, with the highest attention given to secondary school students (143 instances) and the lowest to primary school students (103 instances). These differences can provide practical guidance for subsequent ethical risk identification and algorithm optimization. It is recommended to strengthen public opinion monitoring and training data optimization to enhance algorithmic fairness and content diversity, thereby improving the adaptability of LLMs in the context of international Chinese education.

 

Keywords

Large language model, overseas international Chinese teaching institutions, educational ethics