ALMs: Authorial Language Models for Authorship Attribution

• DOI: 10.48550/arxiv.2401.12005
• Publication Date: 2024-01-01
• Authors: Weihang Huang; Akira Murakami; Jack Grieve
• Keywords: Perplexity; Authorship Attribution; Macro; Stylometry

In this paper, we introduce an authorship attribution method called Authorial Language Models (ALMs) that involves identifying the most likely author of a questioned document based on perplexity calculated for set causal language models fine-tuned writings candidate author. We benchmarked ALMs against state-of-art-systems using CCAT50 dataset and Blogs50 datasets. find achieves macro-average accuracy score 83.6% Blogs50, outperforming all other methods, 74.9% CCAT50, matching performance best method. To assess shorter texts, also conducted text ablation testing. found to reach 70%, needs 40 tokens 400 while 60% requires 20 70 CCAT50.