Computational Stylometric Authorship Attribution of Anonymous Online Extremist Manifestos: A Forensic Linguistic Investigation of Authorial Style, Clustering, and Stylistic Overlap

Abstract

The increasing availability of anonymous online extremist texts has created new challenges for forensic investigators seeking to establish patterns of authorship and identify potential links between documents. Computational stylometry has emerged as a valuable tool for authorship attribution by analyzing recurring linguistic features that characterize individual writing styles. This study applies stylometric authorship attribution methods to a corpus of 45 anonymous online extremist manifestos published between 2015 and 2023 to evaluate the effectiveness of computational stylometry in identifying authorship clusters and potential authorial overlap across documents originating from different extremist movements. The research is grounded in Coulthard and Johnson's (2007) forensic linguistics methodology, Juola's (2015) authorship attribution framework, and the pioneering statistical authorship analysis developed by Mosteller and Wallace (1964). Linguistic analyses were conducted using the JGAAP and Stylo R-package toolkits, employing function word frequency profiles, character n-gram features, and Burrows' Delta distance measures to capture stylistic similarities and differences among texts. Cross-validation experiments demonstrated that stylometric methods classified documents by authorial style with an overall accuracy of 74%, substantially exceeding chance-level performance. Among the linguistic features examined, function word frequency profiles emerged as the most reliable indicator of stable authorial style, outperforming character n-grams and other stylistic measures. Cluster analyses further identified notable stylistic overlap among three manifestos previously considered unrelated, suggesting either potential common authorship or deliberate stylistic imitation. The findings demonstrate that computational stylometry can provide valuable probabilistic evidence for forensic investigations while highlighting the need for cautious interpretation, as stylistic similarity alone cannot establish authorship with certainty. The study emphasizes the importance of integrating stylometric evidence with contextual and investigative information and discusses the methodological limitations, ethical considerations, and legal implications of applying computational authorship attribution to extremist texts. Overall, the research contributes to forensic linguistics by demonstrating the value of robust stylometric techniques for examining anonymous digital writing while advocating responsible and transparent use of computational evidence in investigative and judicial contexts.