GUNLP at SemEval-2026 Task 10: Psycholinguistic Conspiracy Marker Extraction and Detection (PsyCoMark)

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

The Georgetown University NLP (GUNLP) system was developed for SemEval 2026 Task 10, focusing on classifying conspiratorial beliefs in Reddit posts (Subtask 2). This system utilizes COVID-Twitter-BERT v2 (CT-BERT-v2) within a multi-task learning framework, employing a dual-head architecture to jointly optimize conspiracy classification and emotion label prediction. To counter data scarcity, the training set was augmented using paraphrasing and GPT-5-generated chain-of-thought emotion annotations, expanding the corpus to approximately 8,600 examples. Evaluation of two input configurations revealed that the emotion-aware setup achieved an F1 score of 0.87 on the official development set, surpassing the text-only baseline by five F1 points and highlighting the benefit of paraphrased samples and affective auxiliary supervision.

Key takeaway

For NLP Engineers developing robust social media content classification systems, especially for sensitive topics like misinformation, you should consider integrating multi-task learning with auxiliary emotion prediction. Leveraging data augmentation techniques, such as paraphrasing and large language model-generated annotations, can effectively mitigate data scarcity and significantly boost model performance, as demonstrated by the 0.87 F1 score achieved with emotion-aware inputs.

Key insights

Emotion-aware multi-task learning and data augmentation significantly enhance conspiracy belief detection in social media.

Principles

• Multi-task learning with auxiliary tasks improves primary classification performance.
• Data augmentation via paraphrasing and large language models can effectively address data scarcity.
• Incorporating affective features enhances social media text analysis.

Method

A multi-task learning framework with a dual-head architecture, using CT-BERT-v2, jointly optimizes conspiracy classification and emotion prediction, augmented by paraphrasing and GPT-5-generated emotion annotations.

In practice

• Use paraphrasing to double training data for scarce datasets.
• Employ GPT-5 for generating chain-of-thought emotion annotations.
• Integrate emotion labels as auxiliary supervision in multi-task learning for improved text classification.

Topics

• SemEval 2026
• Conspiracy Detection
• Multi-task Learning
• Data Augmentation
• Emotion Analysis
• COVID-Twitter-BERT v2

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning 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.
• Implicit and Indirect: Detecting Face-threatening and Paired Actions in Asynchronous Online Conversations — Detecting face-threatening actions in asynchronous conversations requires handling ambiguity and multiple interpretations.
• On GATE, Text and Social Media Analysis, and Detecting Misinformation Online — A multi-layer analytical pipeline effectively extracts and categorizes digital identity aspects of agrifood products from online discourse.
• Fuzzy Fingerprinting Encoder Pre-trained Language Models for Emotion Recognition in Conversations: Human Assessment and Validity Study — Fuzzy Fingerprints enhance PLMs for ERC, improving interpretability and reducing neutral class overclassification.
• A Multitask Transformer for Offensive Language Detection and Target Identification in HateBR — Multitask learning improves offensive language detection consistency but may reduce fine-grained target attribution.
• 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.