Scholarlink

Due to the high complexity and diversity of writing, automated essay evaluation systems face significant challenges. Large language models (LLMs), representing the latest peak in NLP technology for semantic understanding, hold immense potential for advancing essay evaluation systems. In the NLPCC 2024 Shared Task 4 Chinese Essay Discourse Logic Evaluation and Integration, we investigated improving LLMs’ capabilities in evaluating essay logic, coherence, and quality. Considering the characteristics of different tasks, we adopted MRC-style instructions to optimize output formats and implemented undersampling to address data imbalance. To enhance efficiency and model performance, we explored LLM fine-tuning methods that decouple tasks and applied similarity comparison to refine model outputs. Additionally, we utilized noisy embedding fine-tuning to mitigate overfitting. Our approach achieved the top ranking in the NLPCC 2024 Shared Task 4.

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

Enhancing Chinese Essay Discourse Logic Evaluation Through Optimized Fine-Tuning of Large Language Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jinwang Song,
  • Yanxin Song,
  • Guangyu Zhou,
  • Wenhui Fu,
  • Kunli Zhang,
  • Hongying Zan

摘要

Due to the high complexity and diversity of writing, automated essay evaluation systems face significant challenges. Large language models (LLMs), representing the latest peak in NLP technology for semantic understanding, hold immense potential for advancing essay evaluation systems. In the NLPCC 2024 Shared Task 4 Chinese Essay Discourse Logic Evaluation and Integration, we investigated improving LLMs’ capabilities in evaluating essay logic, coherence, and quality. Considering the characteristics of different tasks, we adopted MRC-style instructions to optimize output formats and implemented undersampling to address data imbalance. To enhance efficiency and model performance, we explored LLM fine-tuning methods that decouple tasks and applied similarity comparison to refine model outputs. Additionally, we utilized noisy embedding fine-tuning to mitigate overfitting. Our approach achieved the top ranking in the NLPCC 2024 Shared Task 4.