Differentiating Chat Generative Pretrained Transformer from Humans: Detecting ChatGPT-Generated Text and Human Text Using Machine Learning

Iyad Katib, Fatmah Y. Assiri, Hesham A. Abdushkour, Diaa Hamed, Mahmoud Ragab

Recently, the identification of human text and ChatGPT-generated text has become a hot research topic. The current study presents a Tunicate Swarm Algorithm with Long Short-Term Memory Recurrent Neural Network (TSA-LSTMRNN) model to detect both human as well as ChatGPT-generated text. The purpose of the proposed TSA-LSTMRNN method is to investigate the model’s decision and detect the presence of any particular pattern. In addition to this, the TSA-LSTMRNN technique focuses on designing Term Frequency–Inverse Document Frequency (TF-IDF), word embedding, and count vectorizers for the feature extraction process. For the detection and classification processes, the LSTMRNN model is used. Finally, the TSA is employed for selecting the parameters for the LSTMRNN approach, which enables improved detection performance. The simulation performance of the proposed TSA-LSTMRNN technique was investigated on benchmark databases, and the outcome demonstrated the advantage of the TSA-LSTMRNN system over other recent methods with a maximum accuracy of 93.17% and 93.83% on human- and ChatGPT-generated datasets, respectively.