Literature Representation using Character Networks based on Sentiment Analysis

Abstract

In this study, we propose a method for expressing literary works using machine learning. The proposed method expresses novels by composing a network of characters appearing in the novel and identifying the flow of sentiment scores between characters who agree or oppose the main character. Characters, represented as nodes on the constructed network, are extracted via the name entity recognizer, whereas edges are constructed based on the emotional words described in the novel. Protagonist and antagonist groups on the character network are classified via signed graph clustering. The novel proceeds through the interaction between the characters, and the groups are segmented to emphasize this through character grouping and network construction. A novel is classified into four acts, and the emotions of each group are emphasized and expressed in each act. In this study, 20 novels are clustered using the proposed method; subsequently, they are compared and tested using other expression methods.