Network Science Predicts Who Dies Next in Game of Thrones

Abstract

Social network analysis and machine learning have found countless applications in recent years. As an example, this short project was carried out in 2017 and was followed by significant media attention, with the following goal: to bring network science and predictive modeling together on the subject of the popular TV and book series, Game of Thrones, and predict which key characters are likely to meet their ends.

A Network Map of The Witcher

“…To extract the social network [5][6][7] of The Witcher, I used the following text-processing procedure on each book. First, I tokenized every book into a list of sentences and then labeled each sentence by the name of the characters appearing in that sentence.…”

“…Second, I defined a window size as the distance of two pairs of sentences and assumed that if two characters are mentioned in two sentences within this window then there is an edge with a weight of one between them. The following two quoted sentences illustrate when the two mentions occur in the same sentence or two sentences apart: Then, I tested several window sizes (1,3,5,6,7), and based on the completeness of the network and the level of noise, I decided to use a window size of five sentences. Following this rule, I finally arrived at a network of 541 nodes (characters) among whom there are 3,306 links total, with a median connection strength of five.…”