Contextual deep learning for accurate news article categorisation with pre-trained embeddings.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Digital Journalism

Background:

• The proliferation of online news necessitates automated classification systems.
• Manual document classification is inefficient and prone to errors.
• Automated systems offer advantages in managing large volumes of digital content.

Purpose of the Study:

• To develop and evaluate a deep learning approach for automated news content classification.
• To compare the effectiveness of deep learning against classical machine learning methods.
• To investigate the impact of dataset characteristics on classification performance.

Main Methods:

• Utilized pre-trained word embeddings to capture semantic meaning.
• Employed a hybrid neural network architecture incorporating local and distant text dependencies.
• Compared deep learning models with classical machine learning algorithms.
• Evaluated performance on the AG News corpus and the News Category Dataset V3.

Main Results:

• Deep learning achieved over 91% accuracy on the balanced AG News corpus.
• Performance was considerably lower on the more complex and unbalanced News Category Dataset V3.
• Demonstrated significant improvement over classical machine learning methods.

Conclusions:

• Contextual and semantic deep learning models are effective for news categorization.
• Dataset complexity and class imbalance significantly impact the performance of classification models.
• Automated news classification systems offer a viable solution for digital journalism.