Rule-Based Framework IndoNSD Improves Accuracy in Indonesian Sentiment Analysis Models

A newly developed framework named IndoNSD has improved the performance of Indonesian sentiment analysis by integrating rule-based negation scope detection into neural network architectures. In evaluations conducted across three diverse datasets, the method achieved accuracy improvements ranging from 0.72% to 1.71% and F1 score gains between 0.80% and 1.66% over six baseline methods, maintaining stable performance across multiple independent runs.

The study addresses a persistent resource gap in natural language processing for the Indonesian language, where negation fundamentally inverts sentiment polarity. While English sentiment analysis benefits from annotated corpora such as BioScope, SFU Review, and CoNLL datasets to enable supervised negation handling, Indonesian has had zero publicly available resources for negation scope detection. As a result, previous computational approaches relied on simple next-word heuristics that fail to account for complex syntactic patterns involving prepositional extensions, adverbial modifications, and idiomatic constructions.

To overcome these grammatical complexities, the researchers designed IndoNSD as a rule-based negation handling framework that operationalises Sudaryono’s linguistic theory into practical algorithms. The framework introduces four novel linguistically grounded rules intended to systematically capture Indonesian negation structures that were previously overlooked by computational methods.

The system encodes these rules into a four-class negation vector representation distinguishing tokens that are outside scope, cues, within scope, or part of a special construction. These vectors are integrated into BiLSTM-CNN neural architectures through embedding concatenation, representing the first application of linguistically grounded rule-based negation features in Indonesian neural sentiment analysis to establish immediate practical deployment and a foundation for future supervised methods.

Subject of Research: Indonesian sentiment analysis and negation scope detection

Article Title: Novel rule-based negation scope detection with neural integration for Indonesian sentiment analysis

Article References: Ilmawan, L. B., Muladi, M., & Prasetya, D. D. (2026). Novel rule-based negation scope detection with neural integration for Indonesian sentiment analysis. Neural Computing and Applications, 38(21), Article 865. https://doi.org/10.1007/s00521-026-12619-6

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12619-6

Keywords: Indonesian sentiment analysis, negation scope detection, natural language processing, neural networks, BiLSTM-CNN, IndoNSD, computational linguistics, rule-based systems, Novel, rule-based, negation, scope