.png){kind=icon}
.png){kind=icon}
Return
Human versus AI generated text classification using deep learning and transformers
.png){kind=icon}
PRE
AI
DOI：10.1007/s00521-026-11872-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Among the significant expansion of large language modules, concerns about their possible abuse occupy the fore. In context to the above statement, the differentiation between human-authored and AI (Artificial Intelligence)-generated texts has surfaced as a challenge, demanding resilient classification mechanisms to trim possible unfavorable repercussions. This document delves into this urgent plight through a thorough quest aimed at distinguishing between human and AI-generated texts, harnessing a wide expanse of machine learning (ML) and deep learning (DL) approaches. By juxtaposing ancient algorithms, amalgamation methods, and state-of-the-art transformer-based frameworks like Bidirectional Encoder Representations from Transformers (BERT), Generative pre-trained transformers (GPT), eXtreme Language understanding NETwork (XLNet), and Text-to-Text Transfer Transformer (T5), the exploration strives to specify adequate detection mechanisms. In totality 22 algorithms have been assessed, encompassing traditional ML, DL methodologies, ensembles, and transformer architectures Through the relative analysis of these models, the scrutiny trials reveal practical detection mechanisms. Through the comprehensive appraisals across three unique datasets, operating metrics like precision, accuracy, precision, F1-score, Area Under Curve (AUC), and training time, the exploration labors to furnish valuable perspectives into AI-generated text classification while examining the moral and juridical significances connected to its abuse. Additionally, the study recognizes the complexities introduced by different languages and cultures, highlighting the importance of adapting and refining processing strategies to ensure that detection mechanisms remain effective across diverse linguistic and cultural settings. Besides, it was sensed that GPT declared unequaled enactment across all three datasets, lauding an accuracy exceeding 99%, underlining its ability to distinguish between human and AI-generated texts.
Keywords:
Large language models (LLMs)
Transformers
Deep learning (DL)
Text classification
GPT
Journal
IF:
4.5
Papers:
808
Citations:
3.2W