Ensemble and Transformer Models for Emotion Recognition in Bengali Consumer-Goods E-Commerce Comments

DOI

10.2991/978-94-6239-664-7_39How to use a DOI?

Keywords

Bengali Emotion Classification; E-commerce Feedback; Bengali Sentiment Analysis; Ensemble Methods; Transformer Models; BanglaBERT

Abstract

The rapid growth of the internet and e-commerce in Bangladesh has generated a vast repository of customer reviews in Bengali across domains such as foods, fashions and electronics. It is important to understand what users say in these multilingual user opinions, which leads to a better understanding of customer satisfaction and improved service quality. But most of the sentiment analysis language models are for English only, which misses out on a major portion of fine-level emotions, indicating the need to be developed in the case of Bengali text. This paper introduces a complete Bengali emotion recognition framework for online business that leverages contextual embeddings from BanglaBERT and integrates a soft-voting ensemble of classical machine learning classifiers (Logistic Regression, Random Forest and Multilayer Perceptron). The dataset contains three emotion classes (Negative, Neutral, Positive) and a balanced 3972-comment corpus constructed through preprocessing, human annotation, and inter-annotator agreement validation. Four experimental pipelines were evaluated: (1) traditional machine-learning classifiers (Logistic Regression, Random Forest and Multilayer Perceptron); (2) BanglaBERT embeddings combined with these classifiers; (3) IndicBERT embeddings combined with the same set of classifiers; (4) the proposed softvoting ensemble integrating Logistic Regression, Random Forest and Multilayer Perceptron. BanglaBERT achieved the highest performance of 93.33% when combined with Random Forest, and the proposed Voting Ensemble model obtained the highest macrorobustness F1-score of 0.90, offering the most consistent performance across all classes. Macro F1 is selected as the primary metric due to its balanced evaluation in multiclass settings. The results validate that the stacking diverse classifiers (Logistic Regression, Random Forest, MLP) on top of contextual embeddings improves robustness and interpretability for multi-domain sentiment analysis in low-resource languages. The proposed framework provides a scalable platform for developing automated sentiment and emotion analysis in Bengali e-commerce that may have implications for recommender systems, consumer analytics, and cross-language applications.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Md. Abdul Amin
AU  - Habiba Begum
AU  - Kazi Md. Jahid Hasan
AU  - Md. Jalal Uddin Chowdhury
PY  - 2026
DA  - 2026/06/08
TI  - Ensemble and Transformer Models for Emotion Recognition in Bengali Consumer-Goods E-Commerce Comments
BT  - Proceedings of the International Conference on Intelligent Data Analysis and Applications (IDAA 2025)
PB  - Atlantis Press
SP  - 565
EP  - 578
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-664-7_39
DO  - 10.2991/978-94-6239-664-7_39
ID  - Amin2026
ER  -