A Fine-Tuned BERT-Based Transfer Learning Approach for Text Classification.

Area of Science:

• Natural Language Processing (NLP)
• Data Mining
• Information Retrieval

Background:

• Text classification is crucial for information retrieval and data mining, with applications in healthcare, marketing, and content filtering.
• The proliferation of social media data necessitates automated text classification systems.
• Transfer learning models have shown promise but require evaluation on specific datasets like COVID-19 related content and hate speech.

Purpose of the Study:

• To apply and evaluate transfer learning classification models on COVID-19 fake news and extremist-non-extremist datasets.
• To assess the performance of transfer learning in text classification for sensitive and rapidly evolving data.
• To provide a benchmark for transfer learning model efficacy in these domains.

Main Methods:

• Utilized three datasets: COVID-19 fake news, COVID-19 English tweets, and extremist-non-extremist data.
• Applied transfer learning classification models to the selected datasets.
• Evaluated model performance using accuracy, precision, recall, and F1-score.
• Generated heat maps for detailed model analysis.

Main Results:

• Transfer learning models were successfully applied to datasets not previously used for such experiments.
• Performance metrics (accuracy, precision, recall, F1-score) provided insights into model effectiveness.
• Heat maps visualized classification patterns and model behavior.

Conclusions:

• Transfer learning demonstrates potential for effective text classification on specialized datasets.
• Further research is needed to explore and optimize transfer learning for nuanced text classification tasks.
• The study highlights the importance of evaluating models on diverse and challenging datasets.