Hybrid topic classification approach integrating TF-IDF, BiGRU, and modified builder optimization algorithm (MBOA).

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning
• Artificial Intelligence

Background:

• Automatic topic classification is crucial for organizing large volumes of text data.
• Existing methods often struggle to capture complex contextual information in text.
• Optimization of neural network hyperparameters is key to achieving high performance.

Purpose of the Study:

• To develop a robust framework for enhanced topic classification.
• To introduce a novel metaheuristic algorithm (MBOA) for optimizing neural network hyperparameters.
• To evaluate the performance of the proposed MBOA-optimized BiGRU model on benchmark datasets.

Main Methods:

• Text pre-processing including stop word removal and text normalization.
• Feature extraction using Term Frequency-Inverse Document Frequency (TF-IDF).
• Implementation of a Bidirectional Gated Recurrent Unit (BiGRU) neural network.
• Development and application of the Modified Builder Optimization Algorithm (MBOA) for hyperparameter tuning.

Main Results:

• The MBOA-optimized BiGRU model achieved state-of-the-art performance on both the BBC News and AG News datasets.
• Exceptional accuracy of 98.62% on the BBC News dataset.
• High accuracy of 98.23% on the AG News dataset, demonstrating effectiveness on diverse data.

Conclusions:

• The proposed framework, particularly the MBOA-optimized BiGRU, significantly advances topic classification capabilities.
• The MBOA algorithm effectively optimizes neural network hyperparameters for superior performance.
• This approach offers a powerful solution for accurate and efficient automatic text categorization.