Data-Driven Research for Film History: Exploring the Jean Desmet Collection

In the past decades, film archives have invested vast amounts of energy and resources into digitizing significant parts of their holdings, including collections of early cinema.¹ This steadily growing digital archive presents a treasure trove for researchers attempting to understand the early years of the medium and its role in modern societies. It also provides new opportunities for audiences to engage with cinema history in attractive ways—opportunities we have only just begun to mine.

At the same time, the exploration and analysis of these digital archives require new means, as does the presentation of results to audiences. Digital tools are helpful here because they can provide comprehensive overviews of collections and reveal macrolevel patterns and structures within them. They are useful starting points for the formulation of new questions and for the development of innovative research paths.² However, the use of these tools also requires reflection on their underlying assumptions and limitations. Digitization transforms physical objects into data, sometimes at the cost of the information available in the original sources (e.g., in the case of film, edge mark information or material features that do not easily get captured in the course of digitization).³ Online collection overviews tend to emphasize factual information such as dates and locations, privileging the known at the expense of the unknown and not always doing justice to a collection’s variety.⁴ Cultural heritage metadata vary enormously, both in extent and quality, and this may skew the interpretation of research results.

In the recent project “Data-Driven Film History: A Demonstrator of EYE’s Jean Desmet Collection,” researchers explored the potential and pitfalls of digital methods for research into early cinema history.⁵ It was a small-scale project lasting eight months part-time that involved academic researchers at the Universities of Amsterdam and Utrecht, along with staff from EYE Filmmuseum and two technical partners.⁶ It centered around the recently digitized collection of early film distributor and cinema owner Jean Desmet (1875–1956), which is part of EYE’s renowned corpus of silent films and related materials. The collection includes films, photographs, posters, and Desmet’s business archive and was digitized as part of the Images for the Future project (a large-scale digitization effort funded by the Dutch government between 2007 and 2014).⁷

From an academic perspective, the objective of the “Data-Driven Film History” project was to try out, and in one case to develop, tools for studying the distribution, screening, and stylistic features of the films in the Desmet Collection. In particular, we sought to test the usefulness of such software in establishing relations between the distribution and screening of the films, and between their screening or programming and some of their aesthetic qualities, specifically their use of color. For EYE, the objective was to gain a better understanding of the quality of its metadata on the Desmet Collection and to experiment with new ways of providing access to information about the films. The software partners were interested in contributing to technologies that could be used for research and access purposes within the heritage sector more broadly.

As part of the project, we developed a mapping tool and experimented with software for the visualization of color patterns in films. The mapping tool was designed to help us visually represent contextual information and map it onto the titles of films in the collection. The visualization software in turn was chosen to help us discern textual patterns in the films themselves, specifically chromatic ones. Combining these tools, we hoped, would allow us to suggest some new pathways for the study of colors in silent cinema and enable researchers to capitalize on the opportunities afforded by recent digitization efforts. Along the way, the project also presented some intriguing methodological issues, which provoked much of the reflection we present here. The process of building a mapping tool, and, in particular, the ancillary work of combining metadata originating from different sources—three different...