We are delighted to announce SemEval-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis on Customer Reviews and Stance Datasets.
Aspect-Based Sentiment Analysis (ABSA) is a widely used technique for analyzing people’s opinions and sentiments at the aspect level. However, current ABSA research predominantly adopts a coarse-grained, categorical sentiment representation (e.g., positive, negative, or neutral). This approach stands in contrast to long-established theories in psychology and affective science, where sentiment is represented along fine-grained, real-valued dimensions of valence (ranging from negative to positive) and arousal (from sluggish to excited). This valence-arousal (VA) representation has inspired the rise of dimensional sentiment analysis as an emerging research paradigm, enabling more nuanced distinctions in emotional expression and supporting a broader range of applications.
To bridge this gap, we propose Dimensional ABSA (DimABSA), a shared task that integrates dimensional sentiment analysis into the traditional ABSA framework. Furthermore, there is a conceptual similarity between stance detection and ABSA when the stance target is treated as an aspect. Building on this, we introduce Dimensional Stance Analysis (DimStance), a Stance-as-DimABSA task that reformulates stance detection under the ABSA schema in the VA space. This new formulation extends ABSA beyond consumer reviews to public-issue discourse (e.g., social, political, energy, climate) and also generalizes stance analysis from categorical labels to continuous VA scores.
———————
Languages
———————
We provide data in 11 languages, including: German (deu), English (eng), Hausa (hau), Japan (jpn), Kinyarwanda (kin), Russian (rus), Swahili (swa), Tatar (tat), Twi（twi), Ukrainian (ukr), and Chinese (zho)
———————
Subtasks
———————
Track A – Dimensional Aspect-Based Sentiment Analysis (DimABSA): Predict real-valued valence–arousal (VA) scores for aspects and extract their associated information from text. Its subtasks include:
- Subtask 1: DimASR – Dimensional Aspect Sentiment Regression
- Subtask 2: DimASTE – Dimensional Aspect Sentiment Triplet Extraction
- Subtask 3: DimASQP – Dimensional Aspect Sentiment Quad Prediction
Track B – Dimensional Stance Analysis (DimStance): A Stance-as-DimABSA task, where the target in stance detection is treated as an aspect. Its subtasks include:
- Subtask 1: DimASR for stance analysis
———————
Evaluation
———————
For both tracks, RMSE is used for Subtask 1, and a new metric (continuous F1) for Subtasks 2 & 3.
———————
Participation
———————
Website (checkout details):
https://github.com/DimABSA/DimABSA2026
Codabench (register and submit results)
- Track A: https://www.codabench.org/competitions/10918/
- Track B: To be announced soon.
Discord (community and discussion)
https://discord.gg/xWXDWtkMzu
Google Group (official updates):
https://groups.google.com/g/dimabsa-participants
———————
Important Dates
———————
- Sample Data Ready: 15 July 2025
- Training Data Ready: 30 September 2025
- Evaluation Start: 10 January 2026
- Evaluation End 31 January 2026
- System Description Paper Due: February 2026
- Notification to Authors: March 2026
- Camera Ready Due: April 2026
- SemEval Workshop 2026: co-located with ACL 2026 (San Diego, CA, USA)
We warmly invite the community to participate in this exciting shared task
and contribute to advancing NLP research.
Best regards,
SemEval-2026 Task 3 Organizers