Abstract

Machine learning is used extensively in authorship attribution of both modern and historical writings. Our research concentrates on 18th century political writing, with a special focus on the work of Thomas Paine. Accurately determining authorship helps achieve a better understanding of the political, ideological, socio-economical and philosophical milieu of the period. This paper presents an enhancement of our prior authorship attribution methodology through the use of artificial neural networks in addition to the previously used support vector machines and centroids. We introduce a method that selects and combines base classifiers. This results in a significantly improved accuracy and confidence in the attribution results. The success of this project is also due to the close collaboration between computer scientists and historians. Automated text analysis brings the best-fitted author into the spotlight, to be cross-examined with historical facts and context. Our methodology has been applied to several 18th century articles. The attribution of two of those articles is also discussed in this paper.