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

We are addressing two fundamental problems in authorship verification (AV):\nTopic variability and miscalibration. Variations in the topic of two disputed\ntexts are a major cause of error for most AV systems. In addition, it is\nobserved that the underlying probability estimates produced by deep learning AV\nmechanisms oftentimes do not match the actual case counts in the respective\ntraining data. As such, probability estimates are poorly calibrated. We are\nexpanding our framework from PAN 2020 to include Bayes factor scoring (BFS) and\nan uncertainty adaptation layer (UAL) to address both problems. Experiments\nwith the 2020/21 PAN AV shared task data show that the proposed method\nsignificantly reduces sensitivities to topical variations and significantly\nimproves the system's calibration.\n