Abstract:
This study examines the effectiveness of deep learning models for sentiment analysis of
Bangla customer evaluations on the Daraz platform. The researchers developed and
evaluated various models, including CNN, LSTM, GRU, and a hybrid CNN+BiLSTM,
focusing on their ability to accurately classify sentiments. The experimental setup involved
exhaustive preprocessing of Bangla text and using TensorFlow and PyTorch frameworks
for model training. The CNN+BiLSTM model achieved the highest accuracy and
precision, indicating its superior performance in identifying positive sentiments. The CNN
model showed balanced performance with high accuracy and F1-Score, making it reliable
for general sentiment classification tasks. The CNN+BiLSTM model was the most
effective for precision sentiment predictions, while the CNN model proved a reliable
choice for balanced sentiment analysis. The research aims to construct sentiment analysis
algorithms for multilingual e-commerce platforms, as online stores like Daraz have a
significant amount of customer feedback, making these critiques more credible than other
forms of advertising material.
