mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection

TL;DR

This paper presents a system that fine-tunes a large language model with data augmentation and self-training to detect conspiracy beliefs in Reddit comments, achieving competitive results in SemEval-2026 Task 10.

Contribution

It introduces a novel application of machine-generated text detection techniques to conspiracy detection, demonstrating effectiveness with limited training data.

Findings

Achieved 85th percentile ranking in the competition.

Utilized data augmentation and self-training for fine-tuning.

Proved the approach's applicability to conspiracy detection.

Abstract

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.