Fusing Stylometric and Embedding Systems to Estimate Authorship Likelihood Ratios in Japanese

This study presents the first application of the likelihood ratio framework for authorship attribution in Japanese texts, integrating stylometric features with embedding-based systems. Using approximately 1,000-character excerpts from blogs, the fused system achieved improved calibration and discriminability, with a log-likelihood-ratio cost of 0.32484. This advancement is significant for practitioners as it demonstrates the potential of combining traditional stylometric methods with modern embedding techniques to enhance authorship analysis in non-English languages.