Figure (A) Excerpt from the multiplex film citation network containing different types of citations. (B) Exemplary illustrations of adapted centrality measures to identify milestone films. (C) Inference of distinguishing markers of milestone films based on various film attributes.
Each year the film industry invests billions based on estimations of audience numbers and box-office revenue. Accordingly, much of today's research about the success in cinema is based on such simple measures. However, due to its limited perspective, the purely economically-driven approach to the assessment of success is denounced by critics and film fans alike. Their suggestions on the other hand usually promote films with common traits, revealing more about personal taste than about the highlights of our film heritage. Efforts to quantify the long-term success of films thus pose numerous challenges and are still missing.
We propose a novel approach to the analysis of cinematic influences based on the multiplex film citation network maintained by the Internet Movie Database. The network is continuously expanded by millions of users and includes records of six types of connections (references, features, remake of, follows, edited from, and spoofs) between over 40,000 international feature films going all the way back to the beginning of cinema. Taking into consideration the time ordering of the films given by their release years, as well as the distinct nature of the citation types, we adapt and extend existing centrality indices to a framework for the quantification of success in art citation networks. In the case of film, these centralities account for the number of received citations, for the time frame in which individual films were influential, for the propagation of cinematic influence through the network, for the out-degree of citing films, and for the importance of films that start influential series. A discounted cumulative score combining individual centralities is thus deduced directly from the self-organizing network of film citations, which is shaped almost exclusively by the preferences of the filmmakers themselves. This measure captures inherent patterns of creative influence rather than external factors, thereby overcoming the most severe limitations of existing measures of success for films.
Our measure of success enables us to identify the influential milestones of film history. Check out the Top 13 or download the Top 100.
Furthermore, based on a wide set of film attributes (such as the genre, technical specifications, cast, and crew) we use the state-of-the-art machine learning technique of random forests to find the characteristic attributes that turn a film into a milestone. Our computational approach to address the question that has preoccupied filmmakers for decades now, What makes a film to be remembered?, results in inferred markers of the coveted milestone films. It also complements established qualitative methods from film aesthetics and contributes a quantitative understanding of success in cinema.
Supported by the Karl-Steinbuch Scholarship for innovative IT- and media-related projects; MFG Foundation (Baden-Württemberg, Germany): innovation.mfg.de