Muhammad Shihab Fathurrahman Jondien, Taqwa Hariguna, Dhanar Intan Surya Saputra

Explainable Aspect-Based Sentiment Analysis with Contrast-Aware IndoBERT for Indonesian Public Service Reviews

Introduction

Explainable aspect-based sentiment analysis with contrast-aware indobert for indonesian public service reviews. This study presents Explainable ABSA using Contrast-Aware IndoBERT for Indonesian public service reviews. Gain 83.4% accuracy & transparent insights for policy evaluation.

Abstract

This study presents an Explainable IndoBERT with Contrast-Aware Attention framework for Aspect-Based Sentiment Analysis (ABSA) on Indonesian public service reviews. The proposed model integrates automated aspect labeling using KeyBERT with a contrast-aware mechanism to handle mixed or opposing sentiments within a single sentence. By leveraging IndoBERT as the base transformer, the system captures context-sensitive sentiment cues while maintaining interpretability through attention-based rationale extraction. Experimental results on the SMSA dataset demonstrate an accuracy of 83.4%, with strong precision in positive and negative sentiment detection. The contrast-aware module improves clause-level understanding, while the attention-based explainability module provides transparent, token-level rationales that align with human judgments at an average rate of 87.7%. Although a modest performance decline occurs compared to non-explainable baselines, the proposed model offers significant gains in semantic transparency, making it suitable for evidence-based policy evaluation and citizen feedback monitoring. This research contributes a practical, interpretable, and linguistically grounded solution for explainable sentiment analysis in low-resource languages, advancing the application of responsible AI in public service analytics.

Review

This study introduces an Explainable IndoBERT with Contrast-Aware Attention framework for Aspect-Based Sentiment Analysis (ABSA) tailored for Indonesian public service reviews, a significant contribution to responsible AI in low-resource language contexts. The authors effectively integrate automated aspect labeling via KeyBERT and a novel contrast-aware mechanism to address the complexities of mixed sentiments within a single sentence. Leveraging IndoBERT as its backbone, the model demonstrates a strong capability to capture context-sensitive cues while prioritizing interpretability, making it highly relevant for evidence-based policy evaluation and citizen feedback monitoring. The methodology is well-articulated, highlighting a robust system that achieves an 83.4% accuracy on the SMSA dataset, accompanied by strong precision for both positive and negative sentiment detection. The inclusion of a contrast-aware module is particularly commendable, addressing a common challenge in fine-grained sentiment analysis and enhancing clause-level understanding. Furthermore, the attention-based explainability module provides transparent, token-level rationales that impressively align with human judgments at an average rate of 87.7%. While the abstract notes a modest performance decline compared to non-explainable baselines, this trade-off is consciously managed, underscoring the research's commitment to semantic transparency and practical utility over raw accuracy alone. Overall, this research presents a valuable and timely solution for advancing interpretable natural language processing in practical applications. The explicit focus on explainability and the handling of linguistic nuances like contrastive sentiments are key strengths. Future work could potentially explore strategies to further mitigate the performance disparity without compromising interpretability, perhaps through more advanced fine-tuning techniques or alternative explanation methods. Nevertheless, the proposed model stands as a linguistically grounded and highly practical tool, offering significant strides in making AI applications more transparent and trustworthy, especially crucial for public service domains and low-resource languages.

Full Text

Comments

You need to be logged in to post a comment.