TeamLasse at SemEval-2026 Task 3: A Hybrid Generative-Discriminative Framework for Dimensional Aspect-Based Sentiment Analysis

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

Summary

TeamLasse presented a hybrid generative-discriminative framework for SemEval-2026 Task 3 Track A, focusing on Dimensional Aspect-Based Sentiment Analysis (DimABSA). This system aims to extract structural sentiment elements, including aspects, opinions, and categories, from text and predict their continuous Valence-Arousal (VA) scores. The primary challenge addressed is the simultaneous execution of structural extraction and continuous numerical regression across highly imbalanced datasets spanning multiple languages and domains. Their proposed solution involves a decoupled, two-stage approach: a generative Large Language Model first extracts structured sentiment tuples, followed by an encoder-based language model that performs the continuous VA regression. To enhance cross-lingual and cross-domain generalization, the models are trained using a targeted data balancing mechanism.

Key takeaway

For NLP Engineers developing advanced sentiment analysis systems, consider adopting a decoupled, two-stage architecture like TeamLasse's. This approach, separating structural element extraction via a generative LLM from continuous VA regression using an encoder, can enhance performance on imbalanced, multi-lingual datasets. You should also implement targeted data balancing to improve cross-domain and cross-lingual generalization in your models.

Key insights

TeamLasse's system decouples sentiment element extraction and VA regression using a hybrid generative-discriminative framework for DimABSA.

Principles

• Decouple complex tasks into specialized stages.
• Address data imbalance for generalization.
• Combine generative and discriminative models.

Method

A generative LLM first extracts sentiment tuples (aspects, opinions, categories). Then, an encoder-based language model performs continuous Valence-Arousal (VA) regression. Targeted data balancing improves cross-lingual and cross-domain generalization.

In practice

• Apply two-stage models for complex NLP.
• Use data balancing for imbalanced datasets.
• Integrate LLMs for structured extraction.

Topics

• Dimensional Sentiment Analysis
• Aspect-Based Sentiment Analysis
• Generative Language Models
• Encoder-Based Models
• Data Balancing
• SemEval-2026

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Automatic Evaluation of Enem Essays: A Comparative Analysis between Feature Engineering and Transformers — Regression formulations are generally more suitable for ordinal essay scoring than multiclass classification.
• Beyond Sentiment Classification: A Generative Framework for Emotion Intensity Evaluation in Text — Generative LLMs fine-tuned on continuous emotion intensity scores outperform classification and generalize to unseen affective dimensions.
• AspectRAG: An Architecture of Retrieval and Generation for Aspect-Based Sentiment Analysis — AspectRAG leverages aspect-guided retrieval and generation for unsupervised, high-performance sentiment analysis in Portuguese.
• Telenor Nordics Customer Service self-help corpus — A new, ethically sourced, multilingual customer service self-help corpus for Nordic languages addresses a critical data scarcity for NLP research.
• 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.
• DiffuSent: Towards a Unified Diffusion Framework for Aspect-Based Sentiment Analysis — DiffuSent unifies ABSA subtasks via non-auto-regressive boundary denoising, improving accuracy and speed for multi-word sentiment extraction.

Open in AIssential →

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