LogSigma at SemEval-2026 Task 3: Uncertainty-Weighted Multitask Learning for Dimensional Aspect-Based Sentiment Analysis

2026-03-25

AI Summary

This work addresses the significant variation in difficulty of predicting continuous Valence and Arousal scores across languages and domains in dimensional aspect-based sentiment analysis. To tackle this challenge, the authors propose a language-dependent homoscedastic uncertainty weighting mechanism that automatically balances regression objectives within a multi-task learning framework. The approach integrates language-specific encoders with ensemble strategies of multiple sub-models to effectively capture cross-lingual differences in task difficulty. Evaluated on two tracks and five datasets from SemEval-2026 Task 3, the method achieves top performance across all settings. Experimental results reveal that optimal task weights vary substantially by language—e.g., 0.66× for German and 2.18× for English—demonstrating the necessity and effectiveness of the proposed weighting mechanism.

Technology Category

• Natural Language Processing: Lexical Semantics and Morphology
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

This paper describes LogSigma, our system for SemEval-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis (DimABSA). Unlike traditional Aspect-Based Sentiment Analysis (ABSA), which predicts discrete sentiment labels, DimABSA requires predicting continuous Valence and Arousal (VA) scores on a 1-9 scale. A central challenge is that Valence and Arousal differ in prediction difficulty across languages and domains. We address this using learned homoscedastic uncertainty, where the model learns task-specific log-variance parameters to automatically balance each regression objective during training. Combined with language-specific encoders and multi-seed ensembling, LogSigma achieves 1st place on five datasets across both tracks. The learned variance weights vary substantially across languages due to differing Valence-Arousal difficulty profiles-from 0.66x for German to 2.18x for English-demonstrating that optimal task balancing is language-dependent and cannot be determined a priori.

Problem

• Dimensional Aspect-Based Sentiment Analysis
• Valence-Arousal prediction
• cross-lingual sentiment analysis
• continuous sentiment scoring
• multitask learning

Innovation

• learned homoscedastic uncertainty
• dimensional aspect-based sentiment analysis
• multitask learning
• valence-arousal prediction
• language-dependent task balancing