TY - GEN
T1 - Modelling the Relative Contributions of Stylistic Features in Forensic Authorship Attribution
AU - Çagatay Sat, G.
AU - Blake, John
AU - Pyshkin, Evgeny
PY - 2025/9
Y1 - 2025/9
N2 - This paper explores the extent to which stylistic features contribute to the task of authorship attribution in forensic contexts. Drawing on a filtered subset of the Enron email corpus, the study operationalizes stylistic indicators across four groups: lexical, syntactic, orthographic, and discoursal. Using R Programming Language for feature engineering and logistic regression modelling, we systematically assessed both the individual and interactive effects of these features on attribution accuracy. Results show that n-gram similarity consistently outperformed all other features, with the combined model of n-gram similarity and its interaction with other features achieving accuracy, precision and F1 scores of 91.6%, 93.3% and 91.7% respectively. The model was subsequently evaluated on a subset of the TEL corpus to assess its applicability in a forensic setting. The findings highlight the dominant role of lexical similarity and suggest that integrating interaction effects can yield further performance gains in forensic authorship analysis.
AB - This paper explores the extent to which stylistic features contribute to the task of authorship attribution in forensic contexts. Drawing on a filtered subset of the Enron email corpus, the study operationalizes stylistic indicators across four groups: lexical, syntactic, orthographic, and discoursal. Using R Programming Language for feature engineering and logistic regression modelling, we systematically assessed both the individual and interactive effects of these features on attribution accuracy. Results show that n-gram similarity consistently outperformed all other features, with the combined model of n-gram similarity and its interaction with other features achieving accuracy, precision and F1 scores of 91.6%, 93.3% and 91.7% respectively. The model was subsequently evaluated on a subset of the TEL corpus to assess its applicability in a forensic setting. The findings highlight the dominant role of lexical similarity and suggest that integrating interaction effects can yield further performance gains in forensic authorship analysis.
UR - https://aclanthology.org/2025.ranlp-1.123/
UR - https://www.scopus.com/pages/publications/105034110572
U2 - 10.26615/978-954-452-098-4-123
DO - 10.26615/978-954-452-098-4-123
M3 - Conference publication
AN - SCOPUS:105034110572
T3 - International Conference Recent Advances in Natural Language Processing, RANLP
SP - 1066
EP - 1073
BT - Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing, RANLP 2025
A2 - Angelova, Galia
A2 - Kunilovskaya, Maria
A2 - Escribe, Marie
A2 - Mitkov, Ruslan
T2 - 15th International Conference on Recent Advances in Natural Language Processing - Natural Language Processing in the Generative AI Era, RANLP 2025
Y2 - 8 September 2025 through 10 September 2025
ER -