Self-calibrating Neural-Probabilistic Model for Authorship Verification Under Covariate Shift

We are addressing two fundamental problems in authorship verification (AV): Topic variability and miscalibration. Variations the topic of disputed texts a major cause error for most AV systems. In addition, it is observed that underlying probability estimates produced by deep learning mechanisms oftentimes do not match actual case counts respective training data. As such, poorly calibrated. expanding our framework from PAN 2020 to include Bayes factor scoring (BFS) an uncertainty adaptation layer (UAL) address both problems. Experiments with 2020/21 shared task data show proposed method significantly reduces sensitivities topical variations improves system’s calibration.