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

Source: [LogSigma at SemEval-2026 Task 3: Uncertainty-Weighted Multitask Learning for Dimensional Aspect-Based Sentiment Analysis](<https://www.sota2.com/research/paper/69c60eea0636e5bf4afc83c3>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-28T05:00:04.914Z

Authors: Baraa Hikal, Jonas Becker, Bela Gipp

Published: 2026-03-26T00:21:13.000Z

## 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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Dimensional Aspect-Based Sentiment Analysis on eng\_laptop (test)](<https://www.sota2.com/research/sota/dimensional-aspect-based-sentiment-analysis-on-eng_laptop-test.md>): Dimensional Aspect-Based Sentiment Analysis; eng\_laptop (test); RMSE: 1.241.

- [Dimensional Aspect-Based Sentiment Analysis on eng restaurant blind (test)](<https://www.sota2.com/research/sota/dimensional-aspect-based-sentiment-analysis-on-eng-restaurant-blind-test.md>): Dimensional Aspect-Based Sentiment Analysis; eng restaurant blind (test); RMSE: 1.104.

- [Stance Detection on deu\_politics Blind (test)](<https://www.sota2.com/research/sota/stance-detection-on-deu_politics-blind-test.md>): Stance Detection; deu\_politics Blind (test); RMSE: 1.342.

- [Stance Detection on eng\_environmental (test)](<https://www.sota2.com/research/sota/stance-detection-on-eng_environmental-test.md>): Stance Detection; eng\_environmental (test); RMSE: 1.473.

- [Stance Detection on pcm politics Blind (test)](<https://www.sota2.com/research/sota/stance-detection-on-pcm-politics-blind-test.md>): Stance Detection; pcm politics Blind (test); RMSE: 1.127.

- [Stance Detection on swa\_politics (test)](<https://www.sota2.com/research/sota/stance-detection-on-swa_politics-test.md>): Stance Detection; swa\_politics (test); RMSE: 1.796.

- [Stance Detection on zho environmental (test)](<https://www.sota2.com/research/sota/stance-detection-on-zho-environmental-test.md>): Stance Detection; zho environmental (test); RMSE: 0.646.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2603.24896>)

[PDF](<https://arxiv.org/pdf/2603.24896.pdf>)
