A Visual Interpretation-Based Self-Improved Classification System Using Virtual Adversarial Training
The research proposes a self-improving spam classification model combining BERT with Virtual Adversarial Training (VAT). It uses a two-stage process: sentiment analysis followed by spam detection. Visual interpretation techniques, like Integrated Gradients, are used to identify important features—specifically URL parts—which are then extracted and added as new labels to further enhance the model's accuracy and robustness.