Enhancing book genre classification with BERT and InceptionV3: a deep learning approach for libraries.

Area of Science:

• Computer Science
• Artificial Intelligence
• Information Science

Background:

• Traditional book genre classification faces challenges with hybrid genres and evolving trends.
• Manual categorization and metadata-based methods have limitations in accuracy and adaptability.
• Automated systems are needed to enhance library organization and information retrieval.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning model for improved book genre classification.
• To integrate visual features from book covers and textual features from titles for genre identification.
• To address the limitations of existing genre classification techniques.

Main Methods:

• A hybrid deep learning model combining InceptionV3 (for visual features) and BERT (for textual features).
• Utilized a scaled dot-product attention mechanism for effective multimodal feature fusion.
• Evaluated the model on the BookCover30 dataset.

Main Results:

• The proposed hybrid model achieved a balanced accuracy of 0.7951 and an F1-score of 0.7920.
• Outperformed baseline models relying solely on image or text features.
• Demonstrated superior performance in classifying book genres.

Conclusions:

• Deep learning offers a powerful approach to enhance automated book genre classification.
• The hybrid multimodal model provides a scalable and adaptable solution for libraries and digital platforms.
• Future work should explore dataset diversity, computational efficiency, and bias mitigation.