NYCU Speech Lab at SemEval-2026 Task 3: Heterogeneous Model Ensemble with Adaptive Weighted Voting for Dimensional Aspect Sentiment Quadruplet Extraction

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

Summary

NYCU Speech Lab's submission to SemEval-2026 Task 3 (DimABSA) for Dimensional Aspect Sentiment Quadruplet Extraction (DimASQP) achieved first place in both Chinese Restaurant and Laptop domains on the tentative test leaderboard. Their system extracts structured tuples—aspect term, aspect category, and opinion term—along with continuous valence-arousal (VA) values from reviews. The approach employs a post-processing ensemble of heterogeneous architectures, including LoRA/QLoRA fine-tuned decoder-only LLMs, a fine-tuned encoder-only model, and optionally, prompted API-based LLMs. To enhance robustness for the continuous F1 (cF1) metric, the system utilizes validation-calibrated weighted voting for tuple selection, weighted VA fusion for numerical aggregation, and strict output validation to enforce task constraints. Experiments demonstrated consistent performance gains over individual models.

Key takeaway

For NLP engineers developing advanced sentiment analysis systems, especially those tackling complex structured extraction tasks like DimASQP, consider implementing heterogeneous model ensembles. Integrating fine-tuned LLMs with encoder-only models, combined with validation-calibrated weighted voting and strict output validation, can significantly boost performance and robustness. This strategy, proven by NYCU Speech Lab's top-ranking system, offers a clear path to improving continuous F1 scores in challenging multi-aspect sentiment extraction.

Key insights

A heterogeneous model ensemble with adaptive weighted voting significantly improves dimensional aspect sentiment quadruplet extraction.

Principles

• Ensemble diverse model architectures for robustness.
• Calibrate voting weights using validation data.
• Enforce task constraints with strict output validation.

Method

The system uses a post-processing ensemble combining LoRA/QLoRA fine-tuned decoder-only LLMs, a fine-tuned encoder-only model, and optional API-based LLMs. It applies validation-calibrated weighted voting for tuple selection and weighted VA fusion for numerical aggregation.

In practice

• Combine LLMs with encoder-only models for sentiment tasks.
• Implement weighted voting for ensemble decision-making.
• Use strict validation to ensure output format compliance.

Topics

• SemEval-2026 Task 3
• Aspect-Based Sentiment Analysis
• Large Language Models
• Model Ensembling
• Weighted Voting
• Valence-Arousal Extraction

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

Related on AIssential

• From TF-IDF to Transformers: A Comparative and Ensemble Approach to Sentiment Classification — A soft voting ensemble of diverse machine learning and transformer models significantly enhances sentiment classification accuracy, particularly for nuanced text.
• Comparing LLM and Fine-Tuned Model Performance on NVDRS Circumstance Extraction with Varying Prompt Complexity — LLMs, guided by a context-aware "Complexity Score" for prompt selection, significantly outperform fine-tuned models on inferentially complex, low-prevalence data extraction.
• 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.
• Consistency Analysis of Sentiment Predictions using Syntactic & Semantic Context Assessment Summarization (SSAS) — SSAS improves LLM sentiment prediction consistency by pre-processing noisy data into structured, high-signal prompts.
• How to Fine-Tune an SLM for Emotion Recognition — Fine-tuning SLMs for fine-grained emotion recognition on imbalanced datasets requires a multi-technique approach to data balancing and loss weighting.
• Multi-Stage Training for Abusive Comment Detection in Indic Languages — A multi-stage pipeline combining language preprocessing and model ensembles effectively detects abusive Indic language comments while minimizing false positives.

Open in AIssential →

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