LAUSR

Author verification is a fundamental problem in authorship attribution, and it suits most relevant applications where it is not possible to predefine a closed set of suspects. So far, the most successful approaches attempt to sample the non-target class (all documents by all other authors) and transform author verification to a binary classification task. Moreover, they follow the instance-based paradigm (all documents of known authorship are treated separately). In this paper, we propose two algorithms, one instance-based and one profile-based (all known documents are treated cumulatively) that are able to outperform state-of-the-art methods in several benchmark datasets. We demonstrate that the proposed methods are capable of taking advantage of the availability of multiple documents of known authorship and that they are robust when text length is reduced.