THE MAIN WAYS TO IDENTIFY AI-GENERATED DATA IN ACADEMIC AND ONLINE PUBLICATIONS USING MACHINE LEARNING AND NLP MODELS

Abstrak

This article comprehensively analyzes contemporary methodologies for identifying text generated by Artificial Intelligence (AI) across academic and online publications. The rapid maturation of generative language architectures has fundamentally altered the processes of information synthesis, significantly complicating the crucial task of determining true authorship origin. The primary objective of this research is to investigate the practical efficacy of identifying synthetic texts by employing Natural Language Processing (NLP) techniques alongside advanced Machine Learning (ML) models. The study conducts a comparative evaluation of stylometric analysis, "zero-shot" probabilistic models, and transformer-based deep learning architectures, such as RoBERTa. Findings demonstrate the systemic ineffectiveness of traditional plagiarism frameworks and validate the high accuracy of hybrid detection models. Specifically, the reliability of ensemble approaches is proven under conditions involving complex algorithmic attacks and automated text paraphrasing. The derived conclusions provide vital practical recommendations for higher education institutions and academic journal editorial boards.