wangkongqiang at SemEval-2026 Task 10: PsyCoMark- Psycholinguistic Conspiracy Marker Extraction and Detection

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, medium

Summary

Wang Kongqiang and Tan Qingli developed a system for SemEval-2026 Task 10: PsyCoMark, focusing on Psycholinguistic Conspiracy Marker Extraction (Subtask 1) and Conspiracy Detection (Subtask 2). Their approach utilized four pre-trained language models for English text: distilbert-base uncased, distilbert-base-multilingual-cased, lxyuan-distilbert-base-multilingual-cased-sentiments-student, and microsoft-deberta-v3-base. The methodology involved visual analysis of training data, data augmentation for Subtask 2 using the gemma-3-27b-it generative model, and hyperparameter tuning for multiple single models. The system achieved a Macro F1 score of 0.1587 for Subtask 1, which evaluates Actor, Action, Effect, and Victim markers, and a Weighted F1 score of 0.7411 for Subtask 2, a binary text classification task, securing a good ranking on the leaderboard.

Key takeaway

For NLP engineers developing systems for identifying psycholinguistic conspiracy markers or detecting conspiracy theories, this work demonstrates the utility of fine-tuning pre-trained models like DeBERTa and DistilBERT. You should consider data augmentation with generative models such as gemma-3-27b-it to improve performance, especially for binary classification tasks. Rigorous hyperparameter tuning is essential to optimize single model predictions and achieve competitive results in semantic evaluation challenges.

Key insights

The system uses fine-tuned DistilBERT and DeBERTa models with data augmentation for conspiracy detection and marker extraction.

Principles

• Hyperparameter tuning is crucial for single model performance.
• Data augmentation can enhance performance in text classification tasks.

Method

Visually analyze training data, augment Subtask 2 data with gemma-3-27b-it, train multiple single models, then tune hyperparameters to select the best for prediction.

In practice

• Use gemma-3-27b-it for text data augmentation.
• Fine-tune DistilBERT/DeBERTa for classification.
• Evaluate marker extraction with Macro F1.

Topics

• SemEval-2026 Task 10
• Conspiracy Detection
• Psycholinguistic Markers
• Natural Language Processing
• Data Augmentation
• Pre-trained Language Models

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection — Data augmentation and self-training can effectively finetune LLMs for specialized text classification tasks with limited data.
• Muted: Multilingual Targeted Offensive Speech Identification and Visualization — Muted identifies targeted offensive speech using transformer models and spaCy, requiring no fine-tuning for toxic span detection.
• A Resource for Enthymeme Detection in Controversial Political Discourse — The study of annotator disagreement in enthymeme detection improves model performance and understanding of human inference.
• MentalMARBERT: Domain-Adaptive Pre-training and Two-Stage Fine-Tuning for Arabic Mental Health Disorders Detection — Domain-adaptive pre-training and hierarchical classification significantly improve Arabic mental health disorder detection.
• A Multimodal Framework for Financial Fake News Detection for Brazilian Portuguese — Multimodal analysis, integrating text and image-extracted text, significantly improves financial fake news detection.
• Robust Dual-Signal Fusion: Hybrid Neuro-Symbolic Gating with Compressed Chain-of-Thought Refinement for Irony Detection in Social Media Texts — Hybrid neuro-symbolic fusion with compressed Chain-of-Thought effectively overcomes LLM literalism for zero-shot irony detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.