Social Network Structures in Movies

Michael Schober, Schober@wim.uni-koeln.de Jermain Kaminski, jermaink@mit.edu

Keywords: social network analysis, betweenness centrality, contribution index, movie actors, success factors Although humanities and cultural studies have a long tradition in formalistic interpretations of works of art and literature (e. g. Erlich, 1973; Wellek and Warren, 1956), only few writers have understood these works as networks of characters that (inter)act (with each other). This paper expands this stream of thought by extracting, visualizing and analyzing the evolution of the characters interaction networks in over 900 Movies. The information about actors (nodes) and connections (edges) between them are extracted from movie scripts available on various internet sites. We developed a script parser that automatically identifies the characters and scenes in movie script based on regular expressions. In the following step edges are created between characters appearing in the same scene. The scenes also allow us to use the data for a dynamic analysis, where network metrics of characters change over time. By doing so, we use measures of social network structure such as in- and out-degree and betweenness centrality in combination with an actor contribution index. In the paper, we briefly discuss several decisions that have to be taken during the extractionprocess. What happens to actors who speak in the off? How does the parser handle narrators? With respect to the different layers of the story telling in a movie, the meaningfulness of the extracted social network will be critically addressed. The main part of the paper deals with the analysis of the actor networks. Static visualizations as well as dynamic visualizations from specific movies will be presented. Additionally to the static and dynamic network variables of each movies social network, several other variables are included in the dataset. This allows us to link the network structure to variables like genre, production costs, date of production or ratings from critics. Looking at centrality measures from characters over time, we can identify differences between groups of movies (genre, date of production) and finally come to the final question if successful movies share a common social network structure. Finally we illustrate the potential of this approach in an interdisciplinary research context. Scholars in the humanities can benefit from this just as well as researchers in management science who aim at investigating success factors of movies on the basis of quantitative methods (e.g. Hennig-Thurau et al. 2007). For example, there are not yet many satisfactory quantitative measures reflecting movie script quality that can support managers during the greenlighting process of a production. Considering that big movies have production cost ranging from 150 up to 250 million $, the given set of analysis might also be interesting for the economical perspective.

1: Social Network Structure of "Babel"

2: Social Network Structure of "Schindlers List"

3: Social Network Structure of "Pretty Woman"

2