A novel approach for Arabic business email classification based on deep learning machines

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Linguistics

Background:

• The increasing volume of business emails necessitates efficient content classification.
• Arabic text classification remains an under-researched area despite its growing importance in official correspondence.
• Existing email classification algorithms often lack specialized support for Arabic content.

Purpose of the Study:

• To develop and evaluate machine learning models for classifying Arabic business emails.
• To address the specific challenges of Arabic text classification in a business context.
• To categorize emails based on urgency, sentiment, and topic.

Main Methods:

• Utilized a large dataset of 63,257 Arabic business emails.
• Employed natural language processing (NLP) techniques combined with machine learning.
• Developed and tested multiple convolutional neural network (CNN) models, each tailored for a specific classification task (urgency, sentiment, topic).
• Incorporated a lexicon of words to enhance email identification and classification.

Main Results:

• Achieved high accuracy, exceeding 92%, across all classification tasks.
• Maintained a low loss rate, below 8%, indicating robust model performance.
• Demonstrated the effectiveness of CNN models for Arabic email classification.
• Validated the correctness and robustness of the proposed classification approach.

Conclusions:

• The developed NLP and CNN models provide a highly accurate solution for classifying Arabic business emails.
• This research significantly contributes to the field of Arabic text classification.
• The findings support the practical application of these models in managing large volumes of business correspondence.