NLP Implementation For AI Generated Text Detection (ChatGPT) Using Naive Bayes Method

The development of artificial intelligence (AI) technology, especially large language models like ChatGPT, presents challenges related to the authenticity and validity of digital content. AI's ability to produce human-like text opens up opportunities for misuse, such as plagiarism and information manipulation. This study aims to develop an AI text detection system using the Multinomial Naive Bayes algorithm, due to its ease of use and high effectiveness algorithm has become a popular choice for text classification.. The dataset used is the Human ChatGPT Comparison Corpus (H3C), sourced from the ELI5 subreddit on Reddit, consisting of 800 entries of questions and answers from both humans and AI. The labeling process involves combining answers into a single column and assigning labels based on the source. Preprocessing steps include case folding, removal of digits and punctuation, tokenization, stopword removal, normalization, and text finalization. Text features are extracted using the TF-IDF method, limited to the top 1000 features. The model is trained on 80% of the data and tested on the remaining 20%. The evaluation shows an accuracy of 93%. These findings suggest that the Naive Bayes method is effective in distinguishing AI-generated from human-generated text and has potential as an automatic AI content detection tool.