EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Hybrid Graph Attention and Transformer Framework for Deep Sentiment Classification

A. Pandian

Journal of Intelligent Decision Making and Information Science · 23.09.2026 · 3 (14s) · 1653-1669

DOI: 10.17762/jidmis.v3.4246 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Intelligent Decision Making and Information Science, online erschienen am 23.09.2026. Verfasst von A. Pandian. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Aim/Purpose: This study's main goal is to use cutting-edge machine learning models to assess and improve sentiment categorization in Tamil text, a language with limited resources. With transformer-based embeddings for enhanced contextual comprehension, the research examines the efficacy of transfer-based topologies, graph neural networks (GNNs), and extreme gradient boosting (XGBoost) in sentiment classification.

Background

Sentiment analysis (SA), especially for underrepresented languages like Tamil, is essential to comprehending views shared on social media. Linguistic complexity and data sparsity are frequent problems for traditional approaches. Enhancing SA performance with transformer-based embeddings and advanced deep learning models offers better classification accuracy and richer contextual semantics. Methodology: A dataset of Tamil utterances classified into five sentiment classes—positive, negative, mixed, non-Tamil, and unknown—is used in this research. Preprocessing methods such domain-specific feature extraction, stop-word removal, subword tokenization, and text normalization are used. Transformer-based embeddings such as BERT, XLM-RoBERTa, and fastText are used to evaluate three state-of-the-art models: Transfer-based architectures, GNNs, and XGBoost. Metrics including accuracy, recall, F1-score, and area under the curve (AUC) are used to assess performance. Contribution: Three sophisticated models for Tamil sentiment categorization are thoroughly compared in this work. It displays how well GNNs capture structural linkages and illustrates how transformer-based embeddings affect classification results. The research also highlights how effective XGBoost is in terms of interpretability and training speed, which makes it a good substitute for sentiment analysis.

Findings

According to experimental data, the accuracy of sentiment categorization is greatly increased by using transformer-based models. GNNs improve sentiment categorization by efficiently capturing relational connections in textual input. Because of its effectiveness and interpretability, XGBoost is still a viable alternative. The work emphasizes how crucial it is to combine contextualized token representations with deep learning in order to maximize SA performance for low-resource languages. Recommendations for Practitioners: To improve model performance, practitioners working on Tamil SA could think about using transformer-based embeddings. In situations where relationship structures are important for sentiment comprehension, GNNs may be very helpful. Furthermore, where interpretability and computational economy are important considerations, XGBoost provides a trustworthy substitute. Recommendations for Researchers: To investigate hybrid models that include the advantages of boosting strategies, GNNs, and transfer-based architectures. The robustness of sentiment classification models may be further increased by expanding the dataset to include more varied sources and linguistic variances. Examining methods for semi-supervised or unsupervised learning might potentially help low-resource languages. Impact on Society: Better sentiment analysis in Tamil may help better understand public sentiment on social media, which can help with commercial choices, policymaking, and the moderation of digital material. Improving SA for low-resource languages guarantees that underrepresented populations are more adequately represented in online conversation and encourages linguistic inclusion in AI applications. Future Research: Creating multilingual sentiment categorization models that make use of cross-lingual transfer learning may be the main goal of future research. Investigating how to combine textual sentiment analysis with multimodal data—like audio and images—could improve classification accuracy even further. Enhancing real-time applications may also be possible by using reinforcement learning to improve sentiment categorization algorithms.

Zitieren

APA

Pandian, A. (2026). Hybrid Graph Attention and Transformer Framework for Deep Sentiment Classification. Journal of Intelligent Decision Making and Information Science, 3(14s), 1653-1669. https://doi.org/10.17762/jidmis.v3.4246

Vancouver

Pandian A. Hybrid Graph Attention and Transformer Framework for Deep Sentiment Classification. Journal of Intelligent Decision Making and Information Science. 2026;3(14s):1653-1669. doi:10.17762/jidmis.v3.4246

BibTeX
@article{Pandian2026,
  author = {Pandian, A.},
  title = {Hybrid Graph Attention and Transformer Framework for Deep Sentiment Classification},
  journal = {Journal of Intelligent Decision Making and Information Science},
  year = {2026},
  volume = {3},
  number = {14s},
  pages = {1653--1669},
  doi = {10.17762/jidmis.v3.4246}
}
Kostenlos starten →