Year: 2026
Type: conference-paper
Abstract: This paper describes our system designed for SemEval-2026 Task 10: PsyCoMark-Subtask 2: Conspiracy Detection.We proposed a twostage approach that leverages large-scale pretrained models and a fine-tuned smaller model to detect conspiracy theories in text.In the first stage, we utilize a large model to test all the test samples and filter out those that are clearly unrelated to conspiracy theories.For the remaining samples, we apply a retrievalenhanced custom prompt strategy combined with the Roberta-Large model in the second stage.This allows us to fine-tune the model with weighted predictions based on relevant retrieved information, enhancing detection accuracy.Our system achieved first place on the leaderboard, with an impressive F1 Score of 0.8874.We also present a brief analysis of the effectiveness of the methods used, including the advantages and limitations of large model-based filtering and retrieval-augmented fine-tuning.
Authors: Yuhan Zheng, Yang Yang
Institution: Nanjing University of Science and Technology
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7166824089