Sentiment analysis in multilingual context: Comparative analysis of machine learning and hybrid deep learning models

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning (ML)
• Sentiment Analysis

Background:

• Sentiment analysis of e-commerce reviews is crucial for understanding customer feedback.
• Existing text classification models show varying performance across languages, especially for low-resource languages like Bangla.
• Bengali e-commerce platforms generate a mix of Bangla and translated English reviews, requiring multilingual analysis capabilities.

Purpose of the Study:

• To comparatively analyze the efficacy of various machine learning and deep learning models for sentiment analysis.
• To evaluate model performance on both English and Bangla text from a popular Bengali e-commerce site.
• To identify the most effective text classification models for multilingual sentiment analysis in this domain.

Main Methods:

• Implementation of seven machine learning and deep learning models, including Support Vector Machine (SVM), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM).
• Application of preprocessing techniques and the porter stemming algorithm to enhance text data quality and model accuracy.
• Comparative evaluation of model performance using accuracy metrics on distinct English and Bangla review datasets.

Main Results:

• Support Vector Machine (SVM) models demonstrated superior performance, achieving 82.56% accuracy for English and 86.43% for Bangla sentiment analysis.
• The Bidirectional LSTM (Bi-LSTM) model showed the best performance among deep learning approaches, with 78.10% accuracy for English and 83.72% for Bangla.
• Porter stemming algorithm significantly contributed to improved accuracy across tested models for both languages.

Conclusions:

• Support Vector Machine (SVM) is highly effective for sentiment analysis of mixed-language e-commerce reviews.
• Deep learning models like Bi-LSTM offer competitive performance, particularly for Bangla text sentiment analysis.
• This research advances text classification methodologies for Bangla, contributing valuable insights for NLP and sentiment analysis applications.