kirito at SemEval-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis via Sentence Structure Parsing Preprocessing and Prompt-Enhanced Instruction Tuning

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

Summary

The "kirito" system, developed by Shuangjin Hu for SemEval-2026 Task 3, addresses Dimensional Aspect-Based Sentiment Analysis (DimABSA) by integrating fine-grained aspect extraction with continuous Valence–Arousal (VA) regression. For the DimASR subtask, the system employs a supervised regression approach, utilizing Low-Rank Adaptation (LoRA)-based parameter-efficient fine-tuning alongside a deep nonlinear regression head. For the DimASTE and DimASQP subtasks, it introduces a lightweight sentence structure parsing preprocessing module, which is then combined with prompt-enhanced instruction tuning to achieve unified structured generation of aspect elements and VA scores. Experimental results on official English test sets demonstrate that the "kirito" system surpasses official baselines across most configurations, with syntax-guided prompting enhancing aspect-opinion alignment and the dedicated regression head improving continuous sentiment modeling stability.

Key takeaway

For NLP Engineers developing fine-grained sentiment analysis systems, consider integrating sentence structure parsing and prompt-enhanced instruction tuning. This approach, demonstrated by the "kirito" system, can significantly improve aspect-opinion alignment and the stability of continuous Valence–Arousal (VA) score prediction. You should explore LoRA-based fine-tuning with dedicated regression heads for efficient and robust dimensional sentiment modeling, especially for tasks like SemEval-2026 Task 3.

Key insights

The "kirito" system integrates sentence structure parsing and prompt-enhanced instruction tuning for improved dimensional aspect-based sentiment analysis.

Principles

• Syntax-guided prompting improves aspect-opinion alignment.
• Dedicated regression heads stabilize continuous sentiment modeling.
• LoRA enables parameter-efficient fine-tuning for regression.

Method

For DimASTE/DimASQP, apply sentence structure parsing preprocessing, then prompt-enhanced instruction tuning for unified structured generation of aspect elements and VA scores. For DimASR, use LoRA fine-tuning with a deep nonlinear regression head.

In practice

• Apply LoRA for efficient sentiment model fine-tuning.
• Use syntax parsing to enhance aspect-opinion alignment.
• Implement dedicated regression heads for VA score stability.

Topics

• Dimensional ABSA
• SemEval-2026 Task 3
• Sentence Structure Parsing
• Instruction Tuning
• LoRA Fine-tuning
• Valence-Arousal Regression

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

Related on AIssential

• Aspect-Based Sentiment Analysis in the Accommodation Domain Using the BERTimbau Model — BERTimbau model fine-tuning strategies for Portuguese ABSA show trade-offs between recall and precision.
• Closing the Quality Gap in Low-Resource Text-to-Speech: LoRA Fine-Tuning of VoxCPM2 for Khmer and Korean — LoRA fine-tuning significantly improves low-resource TTS quality, but effectiveness depends on base model weakness.
• 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.
• 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.
• LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification — Iterative imbalance-aware fine-tuning of LLMs, combined with augmentation and post-processing, significantly improves rare-class NLP performance.
• 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.

Open in AIssential →

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