Genre Classification using Character Networks

Literature has always had an undeniable impact on society and culture. It facilitates a nuanced understanding of the world around us and serves to pass on valuable knowledge to future generations; hence it is no surprise that the analysis of literature, and especially genre classification, is an essential field of study. However, the attempt to classify literary works into various categories has always been a persistent challenge for scholars. Assigning genres to fictional literature, which have a narrative style, is a particularly challenging task. Most of the existing work in this field have used statistical parameters in their classification algorithms, which neglect a narrative structure. To solve this problem, this research work has used Character Networks, graphs that explore interactions and relationships among various characters of a literary piece, making them singularly relevant to fiction. This paper proposes a novel method for genre classification that utilizes character networks derived from the text of books and compares Eigenvectors of Laplacian matrices of these networks directly against each other. This research work studies about different interpretations of the same character sets to eliminate noise and focus on character interactions. The proposed method can classify genres with 67.4% accuracy and an 80.5% F1 score.