Scholarlink

Automated class utterance coding is a powerful tool for the evaluation of Chinese class utterances in the context of modern education, which can be seen as a multi-dimension single-label text classification task. However, due to the scarcity of data, the implicitness of Chinese language learning content, and the insufficient representation abilities of traditional machine learning methods, automated class utterance coding remains a challenging task. Therefore, we propose a framework based on a domain-specific large language model (LLM) trained on utterances from Chinese classes with well-designed prompts. The framework consists of several training strategies: data augmentation (DA) with external knowledge for background knowledge awareness, Chain-of-Thought (CoT) for reasoning ability improvement, and curriculum learning (CL) for improving coding ability from easy to hard dimensions progressively. Experiments show the effectiveness of the proposed method on the task of Chinese course core competence dimension coding for Chinese class utterances.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated Coding Utterances Toward Chinese Course Core Competence with Large Language Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mingyang Yue,
  • Tengda Qi,
  • Mingxuan Wang,
  • Wang Ruan,
  • Jun He,
  • Bo Sun,
  • Guomin Zheng

摘要

Automated class utterance coding is a powerful tool for the evaluation of Chinese class utterances in the context of modern education, which can be seen as a multi-dimension single-label text classification task. However, due to the scarcity of data, the implicitness of Chinese language learning content, and the insufficient representation abilities of traditional machine learning methods, automated class utterance coding remains a challenging task. Therefore, we propose a framework based on a domain-specific large language model (LLM) trained on utterances from Chinese classes with well-designed prompts. The framework consists of several training strategies: data augmentation (DA) with external knowledge for background knowledge awareness, Chain-of-Thought (CoT) for reasoning ability improvement, and curriculum learning (CL) for improving coding ability from easy to hard dimensions progressively. Experiments show the effectiveness of the proposed method on the task of Chinese course core competence dimension coding for Chinese class utterances.