UNED at SemEval-2026 Task 9: Sentiment-Aware Transformer Models with Back-Translation Augmentation for Online polarisation Detection

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

Summary

UNED's submission to SemEval-2026 Task 9 (Subtask 1) investigates Spanish online polarization detection using sentiment-adapted transformer models. Researchers compared a base XLM-RoBERTa model, an emotion-adapted model, and a sentiment-adapted XLM-R model, the latter trained on Twitter data. To counter overfitting on a small dataset, back-translation was employed for data augmentation. Experimental results demonstrated that the sentiment-adapted checkpoint consistently outperformed other pretrained models under identical conditions. When combined with back-translation, the final system achieved a macro-averaged F1 score of 0.743 on the preliminary competition leaderboard. These findings indicate that prior adaptation to social media's affective signals offers a beneficial inductive bias for effective polarization detection.

Key takeaway

For NLP Engineers developing social media analysis tools, consider integrating sentiment-adapted transformer models. Your systems can achieve higher accuracy in tasks like online polarization detection by leveraging models pre-trained on affective signals. Additionally, employ back-translation data augmentation to improve model robustness, especially when working with limited training datasets, potentially boosting performance to scores like 0.743 F1.

Key insights

Sentiment-adapted language models, augmented with back-translation, enhance online polarization detection.

Principles

• Sentiment adaptation improves polarization detection.
• Affective signals provide inductive bias.
• Back-translation mitigates overfitting.

Method

Compare base, emotion-adapted, and sentiment-adapted XLM-R models for binary classification. Apply back-translation for data augmentation to small datasets.

In practice

• Pre-train models on sentiment-rich social media.
• Use back-translation for small dataset augmentation.
• Evaluate sentiment-adapted models for social media tasks.

Topics

• Online Polarization Detection
• Sentiment Analysis
• Transformer Models
• XLM-RoBERTa
• Data Augmentation
• Back-Translation
• SemEval-2026

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• Model in Distress: Sentiment Analysis on French Synthetic Social Media — Synthetic data generation with backtranslation can overcome data scarcity and privacy issues in multilingual sentiment analysis.
• Learning Word Vectors for Sentiment Analysis: A Python Reproduction — Word vectors can effectively capture both semantic similarity and sentiment polarity by combining unsupervised and supervised learning objectives.
• Turkish Sentiment Analysis (Keras LSTM + BERT) — A hybrid NLP pipeline combines custom LSTMs and pre-trained BERT for robust Turkish sentiment analysis with real-time data acquisition.
• From Prediction to Justification: Aligning Sentiment Reasoning with Human Rationale via Reinforcement Learning — Integrating explicit reasoning into sentiment analysis models enhances interpretability and performance by mimicking human cognitive processes.
• Consistency Analysis of Sentiment Predictions using Syntactic & Semantic Context Assessment Summarization (SSAS) — SSAS improves LLM sentiment prediction consistency by pre-processing noisy data into structured, high-signal prompts.

Open in AIssential →

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