Paper Title
A Deep Learning Model for Joint Inference in Authorship Attribution Tasks

Abstract
Authorship attribution (AA) is the task of finding the author of an unknown/disputed document. Modern AA has been thoroughly studied in small datasets via complex feature extraction schemas designed for supervised machine learning methods and multivariate analysis techniques. However, such approaches for large and heterogeneous document collections have limitations in terms of domain adaptation and performance. In this study, we propose a deep learning model for joint learning of stylometric embeddings and author related stylometric evidence for authorship attributions tasks. The proposed deep learning model is easily transferable in between author identification and verification tasks. The macro averaged F-Measure performance on Turkish tweet dataset with 5, 20, and 62 authors indicates that our approach is more accurate than baseline performance and has consistent performance degradation with increased number of authors. Keywords - Authorship Attribution, Deep Learning, Joint Inference.