Hyperparameter Tuning for Enhanced Authorship Identification Using Deep Neural Networks

Abstract

Authorship Identification as a task has been long studied and explored. Historically, authorship claims were ratified for copyright issues after the death of the author for unpublished work through style matching. The immense growth in the reach of internet technologies has once again brought to the fore the importance of authorship identification. An application opening up in areas like Intellectual Property Right settlement, Copyrights, Plagiarism, Cyber Crime and Forensics, authorship identification is now an area of active research. The current work presents a Deep Neural Network based approach to authorship identification from a large corpus. The experiments carried out bring out the applicability of Deep Neural Networks for the task and also highlights the importance of hyperparameter tuning for the purpose. Results show that a proper choice and balance in the hyperparameter setting can improve already established outcomes.