A novel ai system for detecting academic predatory practices using big data

Academic predatory behavior encompasses unethical actions within scientific writing, wherein individuals or institutions exploit scholars for personal benefit or to further their interests. This work presented a deep learning approach to textual classification of predators and utilizing graph structures derived from attribute-value pairs to encode formalized knowledge about Journals and Publishers. The implemented method explores a loss function that further applied cross-entropy between the true labels and the predicted probabilities. Its formulation allows the model to prioritize identifying the boundary between the most difficult-to-distinguish classes of being suspicious, predatory, and nonpredatory with a notable accuracy of 94%. This is a one-of-a-kind big data analysis wherein we consider the accumulation of 50,000 observations from various origins. The experimental outcomes across mixed databases illustrate enhanced performance compared to classical deep learning algorithms and establish superiority over existing state-of-the-art methods.