Deep Learning vs. Conventional Approaches in Stylometry: An Authorship Attribution Study

This research work explores the efficacy of Deep Neural Networks, particularly Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), in the field of Authorship Attribution. We try to check whether these networks provide a significant advantage in this domain. For our investigation, a subset of the Gungor Victorian Author Attribution dataset is used. It comprises 13,184 texts from 10 distinct authors. To ensure a balanced comparison between Deep Learning and traditional methods, we also assessed the performance of a Support Vector Machine (SVM) and a simple Multi-Layer Perceptron (MLP) classifier under identical experimental conditions. Our findings revealed that the conventional classifiers outperformed the Deep Learning classifiers. This suggests that, in its current form, Deep Learning may not be the most effective approach for stylometry.