Efficient Sentiment Classification of Bangla E-Commerce Reviews IEEE Transactions on Artificial Intelligence · 2026
BanglaLiteFormer is a lightweight, purpose-built transformer architecture for binary sentiment classification of Bangla e-commerce product reviews. Designed for low-resource NLP settings, it achieves 97.00% test accuracy with near-perfect prediction confidence (99.48%) — outperforming Bangla-BERT, GRU, LSTM, and transformer ensemble baselines — while using just ~150,000 parameters (vs. ~110M for Bangla-BERT).
Authors:
- Tanjin Adnan Abir — M.Sc. Applied Physics & Electronics, Jahangirnagar University
- Md. Tanvir Mahmud Himel — B.Sc. Computer Science & Engineering, Comilla University
- Shukdev Datta, Student Member, IEEE — M.Sc. Computer Science & Engineering, University of Dhaka
Try it now on Hugging Face Spaces:
👉 https://shukdev3-banglaliteformer-new-design.hf.space
Try it now on Github Page:
👉 https://shukdevtroy.github.io/BanglaLiteFormer/
Built with TensorFlow & Gradio · Bengali NLP · Low-Resource Machine Learning
Cloned from @shukdevtroy