用数据增强和自训练提升小样本下的阴谋论检测效果
mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection
- 基于数据增强与自训练，微调Qwen3-32B模型进行二分类
- 在52个提交中排名前8（85百分位），表现强劲
- 源自机器生成文本检测的方法首次有效应用于阴谋论识别
SemEval-2026 Task 10聚焦于阴谋论检测，目标是判断Reddit评论是否表达阴谋信念。我们提交的mdok-style系统采用数据增强与自训练策略，应对训练数据量较小的问题，对Qwen3-32B模型进行微调，完成二分类任务。该系统表现优异，在52个参赛方案中位列第8，处于85百分位。结果表明，最初用于机器生成文本检测的方法同样适用于阴谋论识别。
原文摘要 · Abstract (English)
SemEval-2026 Task 10 is focused on conspiracy detection. Specifically, the goal is to detect whether a Reddit comment expresses a conspiracy belief. Our submitted mdok-style system utilizes data augmentation and self-training (to cope with a rather small amount of training data) to finetune the Qwen3-32B model for a binary text-classification task. The submitted system is very competitive, ranking in the 85th percentile (8th out of 52 submissions). The results shown that our approach, which originated in machine-generated text detection, can be used for conspiracy detection as well.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。