Distance Learning for Author Verification Notebook for PAN at CLEF 2013

This paper presents a distance metric learning method for the 'PAN 2013 Author Identification' challenge. Our approach extracts multiple distance metrics of different document representations from which our system learns to tune each one of these distances and representations to form an combined dis- tance metric. We reach this learning distances by means of linear programming, Support Vector Regression and Neuronal Networks models. As specified by the description of the task our system can be applied to English, Greek and Spanish, and can be configured to be language independent. We achieved moderately suc- cessful results on this PAN competition, with better results on the development test set. We present results with the official test set and our own corpus based on web documents.