Abstract
"随着社交媒体的广泛普及,模因(meme)已成为信息传播与舆论引导的重要载体,其中蕴含的仇恨内容对网络生态与公共安全构成威胁,尤其是通过图像暗示、文化隐喻或社会符号等方式表达的隐性仇恨模因,具有更强的隐蔽性与误导性,给仇恨模因检测任务带来显著挑战。针对上述问题,本文提出了一种仇恨模因理解模型(Hateful Meme Understanding Model,HMUM),在Qwen2.5-VL-72B-Instruct模型基础上引入LoRA微调,并设计了一种多模态多阶段的提示学习框架。该框架通过阶段性引导模型依次完成文本识别、情绪建模与仇恨性推理,逐步增强其对模因语义与情感的理解能力,从而有效提升模型在中文语境下检测语义隐晦、情绪复杂仇恨模因的准确性。在公开数据集ToxiCN MM上的实验结果表明,HMUM(Qwen)在整体任务中取得了显著性能提升,在隐性仇恨模因子集检测方面,相较于基线模型表现出更强的优势。为评估其在更广泛隐性场景中的检测能力,本文构建了以隐性仇恨模因为主的数据集ITTD-220,实验结果显示,HMUM(Qwen)在该数据集上的检测性能同样优于现有模型,验证了其出色的泛化能力。"
- Anthology ID:
- 2025.ccl-1.59
- Volume:
- Proceedings of the 24th China National Conference on Computational Linguistics (CCL 2025)
- Month:
- August
- Year:
- 2025
- Address:
- Jinan, China
- Editors:
- Maosong Sun, Peiyong Duan, Zhiyuan Liu, Ruifeng Xu, Weiwei Sun
- Venue:
- CCL
- SIG:
- Publisher:
- Chinese Information Processing Society of China
- Note:
- Pages:
- 783–791
- Language:
- URL:
- https://aclanthology.org/2025.ccl-1.59/
- DOI:
- Bibkey:
- Cite (ACL):
- 裴淑娟 裴淑娟, Jiali Zuo, Le He, Jianyi Wan, and Mingwen Wang. 2025. HMUM:面向仇恨模因检测的多阶段多模态理解模型. In Proceedings of the 24th China National Conference on Computational Linguistics (CCL 2025), pages 783–791, Jinan, China. Chinese Information Processing Society of China.
- Cite (Informal):
- HMUM:面向仇恨模因检测的多阶段多模态理解模型 (裴淑娟 et al., CCL 2025)
- Copy Citation:
- PDF:
- https://aclanthology.org/2025.ccl-1.59.pdf