Abstract: China’s foreign educational aid not only contributes to the educational development of recipient countries but also helps improve bilateral diplomatic relations and enhance China’s international image. To investigate whether such aid projects effectively improve national image, this paper proposes a human–machine collaborative framework and constructs EduAid-ASent, an Aspect-level Sentiment dataset specifically designed for evaluating the image effects of China’s foreign educational aid. Specifically, the proposed method first employs Large Language Model (LLM) to perform preliminary sentiment analysis on news texts from the Global Database of Events, Language, and Tone (GDELT). Low-confidence outputs are subsequently reviewed by human experts, with annotation quality ensured through iterative refinement. Sentiment analysis is conducted across six aspects corresponding to the main types of educational aid, including the construction and renovation of school buildings. Experimental results demonstrate that the dataset constructed under this framework achieves an average accuracy of 88.02% in aspect-level sentiment annotatio, while reducing manual annotation costs by approximately 97% compared with fully manual methods. Furthermore, temporal trend analysis based on the dataset reveals the differential effectiveness of China’s educational aid projects over various periods, offering valuable insights for future policy adjustment and implementation.

Key words: foreign educational aid, aspect-level sentiment polarity, human–machine collaborative framework, Large Language Model (LLM), news text

摘要： 中国对外教育援助不仅有助于增进受援国的教育发展，还能改善我国与受援国的外交关系，进而提升我国的国际形象。为探究中国对外教育援助项目能否有效提升国家形象，提出了一种人机协同框架，并构建了一个面向对外教育援助形象评估的方面级情感倾向数据集（EduAid-ASent）。首先，利用大语言模型（LLM）对“全球事件、语言和语调数据库（GDELT）”中的新闻文本进行初步情感分析，随后由专家对低置信度样本进行复核，并通过迭代优化确保标注质量。根据教育援助类型，从学校建筑的建设与维修等6个方面，进行情感倾向分析。采用该框架构建的数据集，在方面级情感标注任务中，平均准确率达到了88.02%；且相比纯人工标注，节省了约97%的人工成本。进一步对数据集进行时间趋势分析，揭示出中国对外教育援助项目在各时期的成效，为后续政策的调整和实施提供了参考。

关键词: 对外教育援助, 方面级情感倾向, 人机协同框架, 大语言模型, 新闻文本