Authorship Verification, Neighborhood-based Classification

Daniel Castro, Yaritza Adame, María Pelaez, Rafael Muñoz


The Authorship Analysis task has become a determining tool for the analysis of digital documents in forensic sciences. We propose a neighborhood classification method of Authorship Verification analyzing the similarities of a document of unknown authorship between samples documents of one author, without estimating parameters values from a training data, we implemented two strategies of representation of the documents of an author, an instance based and a profile based one. We will evaluate the methods in different data collections according to the number of samples, the textual genres and the topic addressed. We perform an analysis of the contribution of each function of comparison and each feature used to take as final decision a combination by majority of the votes of each function-feature pair used in the similarity between documents. The tests were carried out using the public data sets of the Authorship Verification PAN 2014 and 2015 competitions. The results obtained are promising and allow us to evaluate our proposal and the identification of future work to be developed.


Authorship detection, author identification, similarity measures, linguistic features.

Full Text: PDF (Spanish)