mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection
For researchers in conspiracy detection, this work demonstrates that a method originally designed for machine-generated text detection can be effectively transferred to this domain, though the improvement is incremental.
The authors applied data augmentation and self-training to finetune Qwen3-32B for binary conspiracy detection in Reddit comments, achieving 8th place out of 52 submissions (85th percentile).
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.