Are You Robert or RoBERTa? Deceiving Online Authorship Attribution Models Using Neural Text Generators

Abstract

Recently, there has been a rise in the development of powerful pre-trained natural language models, including GPT-2, Grover, and XLM. These models have shown state-of-the-art capabilities towards a variety of different NLP tasks, including question answering, content summarisation, and text generation. Alongside this, there have been many studies focused on online authorship attribution (AA). That is, the use of trained models to identify the authors of online texts. Given the power of natural language models in generating convincing texts, this paper examines the degree to which these language models can generate texts capable of deceiving online AA models. Experimenting with both blog and Twitter data, we utilise GPT-2 language models to generate texts using the existing posts of online users. We then examine whether GPT-2-based text generators are capable of mimicking authorial style to such a degree that they can deceive typical AA models. From this, we find that current AI-based text generators are able to successfully mimic authorship, showing capabilities towards this on both datasets. Our findings, in turn, highlight the current capacity of powerful natural language models to generate original online posts capable of mimicking authorial style sufficiently to deceive popular AA methods. This is a key finding given the proposed role of AA in real-world applications such as spam-detection and the investigation of criminal activity online -- where deceptive texts could be automatically generated to mimic authorship in order to mislead these critical AA systems.

COVIDHealth: A novel labeled dataset and machine learning-based web application for classifying COVID-19 discourses on Twitter

“…Additionally, the focus on Twitter data may limit the generalizability of the findings to other social media platforms, as different platforms may exhibit distinct communication patterns and content structures. To address those limitations, one could consider expanding the training dataset to include longer-form content, allowing the classifier to adapt to diverse text lengths and potentially improving prediction accuracy across various content types [95] . Additionally, to enhance the generalizability of the findings, incorporating data from multiple social media platforms and adjusting the model to account for distinct communication patterns and content structures inherent to each platform would provide a more comprehensive understanding of health-related discussions in the broader digital landscape [96] .…”

ConDA: Contrastive Domain Adaptation for AI-generated Text Detection

“…While older AA methods focused on human authors, more recent efforts (Uchendu et al, 2020;Munir et al, 2021) build models to identify the generator for a particular input text. Recent work also shows how AI-generated text can deceive state-of-the-art AA models (Jones et al, 2022), thus making the task of detecting such text even more important.…”

IDTraffickers: An Authorship Attribution Dataset to link and connect Potential Human-Trafficking Operations on Text Escort Advertisements

“…Authorship Attribution and Cybercrime: Numerous authorship attribution studies have been successfully applied to the fields of forensic (Yang and Chow, 2014;Johansson and Isbister, 2019;Belvisi et al, 2020) and cybercrime investigations (Zheng et al, 2003;Rashid et al, 2013), spam de-tection (Alazab et al, 2013;Jones et al, 2022), and linking vendor accounts on darknet markets (Ekambaranathan, 2018;Tai et al, 2019;Manolache et al, 2022;Saxena et al, 2023). However, to our knowledge, none of the existing studies focus on connecting vendors of HT through escort ads.…”