Investigation of fine-tuned BERT models for sentiment analysis in COVID-19 tweets using a fuzzy logic-based ensemble approach
Journal of Computer & Electrical and Electronics Engineering Sciences · 25.04.2026 · 4 (1) · 17-26
DOI: 10.51271/jceees-0033 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Computer & Electrical and Electronics Engineering Sciences, online erschienen am 25.04.2026. Verfasst von Mustafa Sefa Evgin und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Aims
With the beginning of the COVID-19 pandemic, social media applications such as twitter used more than usual because people started to work at their homes rather than offices. Thus, data on this application has become more important to manage crisis of COVID-19. While conventional deep learning methods have shown success in sentiment analysis, they often encounter challenges in capturing the inherent semantic ambiguity and informal linguistic structures prevalent on social media platforms. To find ambiguity on these texts propose ensemble model enhanced by fuzzy logic designed to improve sensitivity and capability.
Methods
Architecture uses BERT model, fine-tuned for specific data to supply dynamic attributes for MLP, LSTM and BiLSTM elements. Their shared executive is regulated via Mamdani Fuzzy Interference System. Then dynamic weights results from defuzzification after mapping prediction of confidence and validation accuracy value on 7 level fine-grained rule set.
Results
Experiment performed on Kaggle Corona NLP dataset resulted in 91.28% accuracy, 91.23% F1 score and 91.38% precision. System’s robust performance is demonstrated by Mean Square Error of 0.2301.
Conclusion
Relative analysis demonstrates dominance of this approach against traditional models. Fuzzy Ensemble model proposes more trustworthy solution for obstruse tweets with successfully straining noise and dealing semantic uncertainties which is naturally present in social media data.
Zitieren
APA
Evgin, M. S., & Toklu, S. (2026). Investigation of fine-tuned BERT models for sentiment analysis in COVID-19 tweets using a fuzzy logic-based ensemble approach. Journal of Computer & Electrical and Electronics Engineering Sciences, 4(1), 17-26. https://doi.org/10.51271/jceees-0033
Vancouver
Evgin MS, Toklu S. Investigation of fine-tuned BERT models for sentiment analysis in COVID-19 tweets using a fuzzy logic-based ensemble approach. Journal of Computer & Electrical and Electronics Engineering Sciences. 2026;4(1):17-26. doi:10.51271/jceees-0033
BibTeX
@article{Evgin2026,
author = {Evgin, Mustafa Sefa and Toklu, Sinan},
title = {Investigation of fine-tuned BERT models for sentiment analysis in COVID-19 tweets using a fuzzy logic-based ensemble approach},
journal = {Journal of Computer & Electrical and Electronics Engineering Sciences},
year = {2026},
volume = {4},
number = {1},
pages = {17--26},
doi = {10.51271/jceees-0033}
}
EUVIMED