Part-of-Speech Enhanced Prompt Learning (PEPL) introduces grammatical features into state-of-the-art pre-trained language models, elevating clickbait detection to new levels of accuracy and efficiency. By converting classification into a prompt-driven cloze task and selectively highlighting part-of-speech cues, the method preserves rich contextual information while requiring significantly fewer labeled examples. This approach stands apart from conventional fine-tuning by systematically activating the latent knowledge within large language models, ultimately delivering a highly scalable solution for content moderation and user experience enhancement in Chinese social media.

本项目提出了**词性增强提示学习（PEPL）**框架，将语法特征巧妙地融入到预训练语言模型中，以更高效、更精准地检测标题党。通过将分类任务转化为“填空式”提示，PEPL在保留上下文信息的同时，大幅减少对标注数据的依赖。与传统微调方式相比，此方法通过深度挖掘大型语言模型的潜在知识库，实现了在中文社交媒体内容审核和用户体验提升方面的高度可扩展性与精确性。