Automated detection of ChatGPT-generated text vs. human text using gannet-optimized deep learning

Abstract

In the digital era, differentiating text produced by Chat Generative Pre-Trained Transformer (ChatGPT) from human-produced text is critical in a digital setting. As artificial intelligence (AI) increasingly produces content, discriminating between sources becomes significant to prevent spam, improve data accuracy, control content quality, and ensure data reliability. Deep learning (DL), machine learning (ML), and Natural Language Processing (NPL) approaches can distinguish between AI and human-generated text based on superior linguistic context, signals, or patterns frequently used. The ability to proficiently make this alteration has huge achievement effects, from enhancing user contribution to contrasting disinformation and upholding the reliability of online communication platforms. This research paper presents a new Gannet Optimization Algorithm with DL-based detection and classification (GOA-DLDC) technique for ChatGPT and human-generated text. The main objective of the GOA-DLDC technique is to recognize and classify the human and ChatGPT-generated text. The GOA-DLDC technique employs the BERT approach for feature vector generation. The classification method is also implemented using the convolutional gated recurrent unit (CGRU) model. To enhance the classification performance of the CGRU model, the hyperparameter-tuning procedure is executed using the gannet optimization algorithm (GOA). The experimental validation of the GOA-DLDC methodology is performed on a dataset comprising human and ChatGPT-generated text. The investigational outcome of the GOA-DLDC methodology portrayed a superior accuracy value of 94.90 % and 94.40 % under human and ChatGPT datasets.