The One with the Social Network Analysis: the extraction, analysis and modelling of temporal social networks from narratives

Abstract

Narratives tell us about the people, cultures, and time periods in and about which they were written. Therefore, narrative analysis is a powerful tool for understanding culture. One way to analyse narratives is through their social networks, however extracting the network is a complex task. Manually recording characters and their interactions is an accurate, but time consuming method for narrative social network extraction, however efficient automatic extraction methods may introduce errors. In this thesis, we perform a detailed comparative study of narrative social network extraction techniques, and investigate the effect the techniques have on the analysis of the narrative. We use the 1994–2004 television series Friends as a case study to model and compare extraction techniques. By designing a simulated social network and observation processes resembling different network extraction techniques, we find that automated network extraction methods are reliable for computing many network metrics, but can distort the clustering coefficient. Our comparison of extraction techniques allows for many more narratives to be extracted and analysed efficiently. We also analyse and model the social networks of Friends, to gain new insights into the the series, and what made it successful. We show which are the most important characters and relationships, and through modelling social network features we find the most informative features to predict success. Our analysis of Friends provides an example and a building block for deeper understanding about particular narratives and narratives in general.