Authorship verification is the task of comparing two or more documents with regard to their stylistic properties in order to determine if they were written by the same author. One of the most effective authorship verification methods to date is "authorship unmasking", which was developed by Koppel and Schler (2004) and extended by us for application on short texts (Bevendorff et al., 2019).
The idea behind unmasking is to build representations of two texts and let a classification system continuously discriminate between those while gradually degrading the representations. The resulting accuracy curves tend to drop faster for two texts written by the same author, largely due to more similar function word distributions. [demo]