BanglaTextDistinguish: A Dataset and Hybrid BiLSTM-SVM Model for Detecting Human and ChatGPT-Generated Bangla Texts

Abstract

ChatGPT, a conversational language model, has largely influenced many aspects of life, bringing numerous advantages and some reasonable concerns. While extensive research has been conducted to differentiate text generated by humans and ChatGPT in various languages, a gap persists specifically in the context of the Bangla language. Due to its extensive vocabulary and complex linguistic structure, differentiating between human and ChatGPT-generated Bangla text poses a considerable challenge. To address this limitation, we introduce a dataset named BanglaTextDistinguish, comprising 6,644 texts written by humans and ChatGPT. Additionally, we propose a hybrid model incorporating Bidirectional Long Short-Term Memory (BiLSTM) networks and Support Vector Machines (SVM) for accurately classifying human-written and ChatGPT-generated texts. This model achieves an impressive accuracy of 82.83%, demonstrating its efficacy as a reliable solution for distinguishing between human and ChatGPT-generated Bangla texts.

Ähnliche Arbeiten