Detection of texts written by human or machine with natural language processing methods

This study aims to develop a method based on natural language processing (NLP) techniques and machine learning models to distinguish between human-written and AI-generated texts. The research utilized datasets consisting of human and AI-generated texts obtained from various sources and implemented comprehensive data processing steps. Initially, the texts underwent preprocessing, where irrelevant words and symbols were removed, and the texts were tokenized and converted into word vectors using the Word2Vec algorithm. The resulting vectors were analyzed using SVM and LSTM models to classify the differences between human-written and machine-generated texts. To enhance model performance, heuristic methods such as genetic algorithms were employed for feature selection, allowing for the reduction of computational costs while optimizing classification accuracy. The developed hybrid model reduced the initial feature set to a more effective subset and was retrained accordingly. As a result, the study achieved high performance in terms of accuracy, ROC curves, and precision-recall analyses, demonstrating the effectiveness of the proposed methods. This research highlights the potential of advanced natural language processing techniques and machine learning models in detecting human and AI-generated texts. The findings provide valuable insights and a solid foundation for future studies in this domain.