Author attribution using network motifs

Abstract

The problem of recognizing the author of unknown text has concerned linguistics and scientists for a long period of time. The authorship of the famous Federalist Papers remained unknown until Mosteller and Wallace solved the mystery in 1964 using the frequency of functional words. After that, many statistical and computational studies were published in the fields of authorship attribution and stylistic analysis. Complex networks, gaining much popularity in recent years, may have a role to play in this field. Furthermore, several studies show that network motifs, defined as statistically significant subgraphs within a network, have the ability to distinguish networks from distinctive disciplines. In this paper, we succeed in the utilization of network motifs to distinguish the writing style of 10 famous authors. Using statistical learning algorithms, we achieved an accuracy of 77% in classifying 100 books written by ten different authors, which outperformed the results from other works. We believe that our method proved the importance of network motifs in author attribution.