LATE-iimas at SemEval-2026 Task 10: Conspiracy Detection via DeBERTa-v3 Ensemble and Weighted Loss Optimization

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

Summary

The LATE-iimas team presented a system for SemEval-2026 Task 10, Subtask 2, specifically addressing conspiracy detection. Their methodology centered on fine-tuning the DeBERTa-v3-Large pre-trained language model. To effectively manage inherent dataset challenges, such as significant class imbalance and the nuanced linguistic ambiguity associated with the "Can't tell" label, the team employed a robust 5-Fold Stratified Cross-Validation technique. This was further enhanced by integrating a Weighted Cross-Entropy Loss function during model optimization. The ultimate system leveraged an ensemble of the individually trained models, which collectively achieved a Weighted F1-Score of 0.75. This strong performance positioned the LATE-iimas team within the top 10 of the competition's overall ranking.

Key takeaway

For NLP engineers developing text classification systems, especially for sensitive tasks like conspiracy detection, consider the benefits of advanced fine-tuning. Your team should explore combining DeBERTa-v3-Large with techniques like 5-Fold Stratified Cross-Validation and Weighted Cross-Entropy Loss to mitigate class imbalance and linguistic ambiguity. Implementing an ensemble of these optimized models can significantly boost performance, as demonstrated by a 0.75 Weighted F1-Score.

Key insights

DeBERTa-v3 ensembles with weighted loss effectively detect conspiracies, addressing class imbalance and ambiguity.

Principles

• Address class imbalance with weighted loss.
• Stratified cross-validation improves robustness.
• Ensembling models enhances performance.

Method

Fine-tune DeBERTa-v3-Large using 5-Fold Stratified Cross-Validation and Weighted Cross-Entropy Loss, then combine models into an ensemble for final prediction.

In practice

• Apply DeBERTa-v3 for text classification tasks.
• Use weighted loss for imbalanced datasets.
• Implement model ensembling for higher scores.

Topics

• Conspiracy Detection
• DeBERTa-v3-Large
• SemEval-2026
• Class Imbalance
• Weighted Cross-Entropy Loss
• Ensemble Learning
• Text Classification

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, 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.
• Ideology Prediction of German Political Texts — Transformer models can effectively map German political texts to a continuous ideological spectrum.
• How I Fine-Tuned DistilBERT to Classify Complaints — And What I Learned Along the Way — Deeply understanding transformer fine-tuning components and their interconnections is crucial for effective model development.
• Shapley Value-Guided Adaptive Ensemble Learning for Explainable Financial Fraud Detection with U.S. Regulatory Compliance Validation — Explainable AI models can achieve high fraud detection accuracy while meeting stringent U.S. financial regulations.
• Toward Trustworthy AI: Multi-Target Adversarial Attacks and Robust Defenses for Continuous Data Summarization — Adversarial attacks on data summarization can compromise trustworthy AI, necessitating robust defenses formulated via DR-submodular optimization.
• I fine-tuned DeBERTa-v3 into a prompt injection pre-filter — 184M, runs on CPU, catches compound attacks by splitting input into fragments — A fine-tuned DeBERTa-v3 model, SPID, pre-filters prompt injections locally using fragment splitting to catch compound attacks, reducing LLM API costs.

Open in AIssential →

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