Open Access

Die Gliederung der Arbeit beruft sich auf die Fragen, die Eder als relevant für jede Figurenanalyse bezeichnet: „Welche Gegenstände – welche Welten und Themen – werden auf welche Weise dargestellt, aus welchen Gründen und mit welchen Wirkungen?“ (Eder 2014: 125). Dementsprechend wird zunächst im Kapitel 2 die Synopsis des Films Get Out und seine Konstruktion einer post-rassistischen Ideologie als Gegenstand genauer vorgestellt. Kapitel 3 der Arbeit widmet sich der Gestaltung von Chris Perspektive und den Faktoren die diese beeinflussen; also der Art und Weise, wie die Ideologie dargestellt wird. Im 4. Kapitel wird aufgezeigt welche Funktionen diese Gestaltung erfüllen kann, um in Kapitel 5 das Potential von Chris Perspektive analysieren zu können. Zum Abschluss wird eine Zusammenfassung der Beobachtungen vorgestellt und ihre Eignung zur Beantwortung der Forschungsfrage reflektiert.

With the rise of streaming platforms and direct-to-consumer business models, big data and audience analytics have emerged as powerful new ways to define target audiences and predict movie revenues in the film industry’s public discourse. Yet while major studios have since launched proprietary internal data science efforts, the role such technologies could play for the practice of independent producers remains unclear. In an effort to close that gap, this dissertation examines how the way feature film producers define their target audience and forecast box office revenues might be improved upon by the use of large sets of individual user data and digital methods of modeling movie preferences. Drawing from the productions studies approach, we establish in an exploratory qualitative study that German independent feature film producers mostly base their target audience definitions on vague, anecdotal evidence, and use intuitive box office forecasts mainly to win over financiers. We then take a look at scientific studies on target audience definition and revenue prediction for feature films and identify reasons for why their results have not been more widely adopted among practitioners. Subsequently, we propose an approach to evidence-based target audience definitions building on recommender algorithms and movie preference modeling at the individual level. Using several datasets on German moviegoers (n = 2,374, n = 6,564, and n = 700, respectively), we confirm that past movie choices indeed provide reliable information on future behavior that can be exploited for efficient targeting. Targeting by as few as three past movie choices also is on average more efficient than targeting by demographic proxies (gender and age). The increase in targeting efficiency as compared to targeting by gender is statistically significant. Outlining a possible practical application, we go on to show that it is possible to develop clusters of similar viewers based on such data, and that such clusters can be used efficiently for target audience definition. Concerning revenue prediction, we utilize a large open dataset on individual users’ movie preferences to compute movie similarities for a sample of 1,093 movies released in the United States between 2004 and 2015 and show that regression models trained on sub-samples of similar movies only lead to better revenue prediction results than models built on the entire dataset. Finally, we address some critical concerns and lay out an agenda for further research on the road to practical implementation of our insights.