HUS@NLP-VNU at SemEval-2026 Task 3: Dual-Stream Syntax-Aware Modeling and Direct Preference Optimization for Dimensional ABSA

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

Summary

HUS@NLP-VNU presented a Dual-Stream Syntax-Aware architecture for SemEval-2026 Task 3, focusing on Dimensional Aspect-Based Sentiment Analysis (DimABSA). This task involves predicting continuous sentiment intensity within the Valence-Arousal space. Their model addresses regression subtasks (DimASR and DimStance) by integrating contextual semantics with a Deep Syntax-Guided Graph Convolutional Network (GCN), featuring a Context-Aware Anchor for semantic filtering and post-norm residuals to prevent oversmoothing. For generative extraction, the team applied Direct Preference Optimization (DPO) using a resource-efficient, heuristic-based data perturbation strategy to create preference pairs without costly LLMs. The regression model achieved top-5 rankings across nine domains and secured the best result on the Chinese-Finance dataset. Empirical analysis confirmed that explicit syntactic modeling consistently improves continuous sentiment regression, while DPO provided modest but stable gains for boundary-constrained extraction.

Key takeaway

For NLP Engineers developing advanced sentiment analysis models, especially for continuous intensity prediction or multilingual contexts, you should consider integrating syntax-aware Graph Convolutional Networks. This approach, combined with Direct Preference Optimization using resource-efficient data perturbation, offers a robust framework. It can significantly improve both continuous sentiment regression and boundary-constrained extraction tasks, as demonstrated by top-5 rankings in SemEval-2026 Task 3.

Key insights

A dual-stream, syntax-aware model with efficient DPO excels in Dimensional ABSA's continuous sentiment prediction and extraction.

Principles

• Explicit syntactic modeling enhances continuous sentiment regression.
• Post-norm residuals prevent oversmoothing in GCNs.
• Heuristic data perturbation can efficiently generate DPO preference pairs.

Method

The Dual-Stream Syntax-Aware architecture combines contextual semantics with a Deep Syntax-Guided GCN, using a Context-Aware Anchor and post-norm residuals. Direct Preference Optimization (DPO) is applied for generative extraction via a heuristic-based data perturbation strategy.

In practice

• Integrate GCNs with post-norm residuals for syntax-aware NLP tasks.
• Explore heuristic data perturbation for DPO without large language models.
• Consider DimABSA for nuanced, continuous sentiment intensity prediction.

Topics

• Dimensional Aspect-Based Sentiment Analysis
• Graph Convolutional Networks
• Direct Preference Optimization
• Sentiment Regression
• SemEval
• Multilingual NLP

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• 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.
• GHI: Graphormer over Conditioned Hypergraph Incidence for Aspect-Based Sentiment Analysis — GHI unifies diverse linguistic and semantic evidence into token–hyperedge incidence relations for robust ABSA.
• Turkish Sentiment Analysis (Keras LSTM + BERT) — A hybrid NLP pipeline combines custom LSTMs and pre-trained BERT for robust Turkish sentiment analysis with real-time data acquisition.
• 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.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• Rethinking Negative Sampling for Knowledge Distillation in Retrieval, A Distillation Recipe for Small Language Model Search Agents, and More! — Effective information retrieval advancements often stem from rethinking fundamental assumptions and optimizing data representation or model behavior.

Open in AIssential →

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