Abstract:
The rapid expansion of e-books and digital libraries has underscored the need for
efficient and accurate genre classification systems, particularly for information
retrieval, content filtering, and book recommendation purposes. This research
"Classification of Bangla Book Genres using Book Cover and Title," aims to develop
a reliable system that leverages both linguistic and visual data for the accurate
classification of Bangla book genres. A comprehensive dataset of over 15,000 Bangla
book covers and titles spanning various genres was collected through web scraping
from the Rokomari e-commerce site and annotated for training and evaluating the
models. Transfer learning architectures employing pre-trained models like Xception
were implemented for feature extraction from images, while the BanglaBERT model
was used to obtain contextualized word embeddings for book titles. The core of the
research involves a comparative analysis of four distinct machine learning and deep
learning models: Logistic Regression, Random Forest, Support Vector Machine, and
Neural Network. The findings reveal that the Neural Network model emerges as the
front-runner, achieving the highest overall accuracy of 78%. This superior performance
underscores the model's ability to generalize well and capture robust features essential
for distinguishing between diverse Bangla book genres. This research demonstrates the
transformative potential of automated content classification systems in enhancing the
accessibility and global reach of Bangla literature, thereby bridging the gap between
traditional literary classification and modern digital behaviors.
