Authorship attribution is the problem of identifying the most plausible author of an anonymous text from a set of candidate authors. Researchers have investigated same-topic and cross-topic scenarios of authorship attribution, which differ according to whether new, unseen topics are used in the testing phase. However, neither scenario allows us to explain whether errors are caused by a failure to capture authorship writing style or by a topic shift. Motivated by this, we propose the topic confusion task where we switch the author-topic configuration between the training and testing sets. This setup allows us to distinguish two types of errors: those caused by the topic shift and those caused by the features' inability to capture the writing styles. We show that stylometric features with part-of-speech tags are the least susceptible to topic variations. We further show that combining them with other features leads to significantly lower topic confusion and higher attribution accuracy. Finally, we show that pretrained language models such as BERT and RoBERTa perform poorly on this task and are surpassed by simple features such as word-level n-grams.

Law enforcement faces problems in tracing the true identity of offenders in cybercrime investigations. Most offenders mask their true identity, impersonate people of high authority, or use identity deception and obfuscation tactics to avoid detection and traceability. To address the problem of anonymity, authorship analysis is used to identify individuals by their writing styles without knowing their actual identities. Most authorship studies are dedicated to English due to its widespread use over the Internet, but recent cyber-attacks such as the distribution of Stuxnet indicate that Internet crimes are not limited to a certain community, language, culture, ideology, or ethnicity. To effectively investigate cybercrime and to address the problem of anonymity in online communication, there is a pressing need to study authorship analysis of languages such as Arabic, Chinese, Turkish, and so on. Arabic, the focus of this study, is the fourth most widely used language on the Internet. This study investigates authorship of Arabic discourse/text, especially tiny text, Twitter posts. We benchmark the performance of a profile-based approach that uses n -grams as features and compare it with state-of-the-art instance-based classification techniques. Then we adapt an event-visualization tool that is developed for English to accommodate both Arabic and English languages and visualize the result of the attribution evidence. In addition, we investigate the relative effect of the training set, the length of tweets, and the number of authors on authorship classification accuracy. Finally, we show that diacritics have an insignificant effect on the attribution process and part-of-speech tags are less effective than character-level and word-level n -grams.

No abstract