@inproceedings{bohan-etal-2024-ji,
    title = "基于指令微调与数据增强的儿童故事常识推理与寓意理解研究",
    author = "Bohan, Yu  and
      Yunlong, Li  and
      Tao, Liu  and
      Aoze, Zheng  and
      Kunli, Zhang  and
      Hongying, Zan",
    editor = "Lin, Hongfei  and
      Tan, Hongye  and
      Li, Bin",
    booktitle = "Proceedings of the 23rd Chinese National Conference on Computational Linguistics (Volume 3: Evaluations)",
    month = jul,
    year = "2024",
    address = "Taiyuan, China",
    publisher = "Chinese Information Processing Society of China",
    url = "https://preview.aclanthology.org/fix-sig-urls/2024.ccl-3.36/",
    pages = "320--326",
    language = "zho",
    abstract = "``尽管现有语言模型在自然语言处理任务上表现出色,但在深层次语义理解和常识推理方面仍有提升空间。本研究通过测试模型在儿童故事常识推理与寓意理解数据集(CRMUS)上的性能,探究如何增强模型在复杂任务中的能力。在本次任务的赛道二中,本研究使用多个7B以内的开源大模型(如Qwen、InternLM等)进行零样本推理,并选择表现最优的模型基于LoRA进行指令微调来提高其表现。除此之外,本研究还对数据集进行了分析与增强。研究结果显示,通过设计有效的指令格式和调整LoRA微调参数,模型在常识推理和寓意理解上的准确率显著提高。最终在本次任务的赛道二中取得第一名的成绩,该任务的评价指标Acc值为74.38,达到了较为先进的水准。''"
}