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

arXiv cs.AI / 3/27/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

The LogSigma paper introduces a system for SemEval-2026 Task 3 focused on dimensional aspect-based sentiment analysis, predicting continuous Valence and Arousal scores rather than discrete sentiment labels.
It tackles cross-language and cross-domain differences in prediction difficulty by learning task-specific homoscedastic uncertainty (log-variance) to automatically weight multiple regression objectives during training.
The approach uses language-specific encoders and multi-seed ensembling, and reports 1st-place results on five datasets across both tracks.
The authors find that learned variance weights differ substantially by language, indicating that optimal balancing is language-dependent and cannot be reliably set in advance.

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.