Event Details
Stylometric Authorship Verification using Features Merging and Deep Machine Learning Techniques
Presenter: Marcelo Luiz Brocardo
Supervisor: Dr. Issa Traore
Date: Tue, April 14, 2015
Time: 14:00:00 - 00:00:00
Place: EOW 430
ABSTRACT
ABSTRACT:
Writing style is an unconscious habit and the patterns of vocabulary and grammar could be a reliable indicator of the authorship of a document. Research in authorship analysis has so far relied on shallow architectures for machine learning classification. In this paper, we explore for the first time deep learning with many layers of non-linear information processing for authorship verification applied for continuous authentication. Our model is based on a Gaussian-Bernoulli Deep Belief Network, which uses Gaussian units in the visible layer to model real-valued data. Our feature set includes lexical, syntactic, and application specific features. In addition, we propose a method to merge a pair of features into a single feature. Our verification method involves decomposing an online document into consecutive blocks of short texts over which (continuous) authentication decisions happen, discriminating between legitimate and impostor behaviors. We validate our method using different block sizes, including 140, 280, and 500 characters.