A study conducted by researchers from the University of California, Berkeley, has provided a comprehensive analysis of diversity trends in Hollywood films through the use of advanced computer vision techniques. The research, spearheaded by David Bamman, an associate professor at UC Berkeley's School of Information, looked at more than 2,300 films released between 1980 and 2022, amounting to 4,412 hours of footage. This study, recently published in the Proceedings of the National Academy of Sciences, marks a significant leap forward in understanding representation in the film industry.
The researchers employed facial recognition technology to track the appearance time of actors on screen in both "popular" films—defined as the top 50 box office earners each year—and "prestige" films, which include films nominated for "Best Picture" by prestigious award organisations such as the Oscars and the Golden Globes. This automated analysis method, supported by a new federal regulation that allows educational institutions to bypass "digital locks" on DVDs for scholarly research, enabled the researchers to tackle the massive data set effectively.
The study found that Hollywood films have become more diverse since 2010, showcasing greater representation of women and actors from various ethnicities, including Black, Hispanic/Latino, East Asian, and South Asian communities. This trend is evident not only on a large scale across the industry but also at the level of individual films. According to Bamman, this indicates a move towards more inclusive casting practices that go beyond a few widely diverse blockbusters.
A notable finding is that while diversity has increased, there remains a pronounced disparity in the representation of different groups in leading versus non-leading roles, with non-leading roles displaying more diversity. This underscores the need for a deeper examination of casting practices, as leading roles continue to be predominantly occupied by white men.
The study further highlights a significant improvement in the visibility of women on screen, with women's screen time rising from approximately 25% between 1980 and 2010 to about 40% by 2022. Despite these gains, all groups except white men are still underrepresented in leading roles.
Bamman and his team took measures to ensure that their results were aligned with public perception by using Wikidata for actors' gender and conducting surveys to understand how audiences perceive actors' race and ethnicity. This focus on perception aims to approximate the representation that viewers encounter.
The study also sheds light on the underrepresentation of Black actors in award-nominated films when compared to popular ones, especially noted in the award circuit from 1980 to 2010.
In a bid to support open research and transparency, the researchers have made available non-copyrighted elements of their data. These include metadata regarding the frames each actor appears in and the location of faces within each frame.
David Bamman expressed hopes that the study would pave the way for further exploration into how actors are depicted in films and address possible stereotypes and biases. The research has also opened up possibilities for collaboration with movie studios and film researchers.
The study owes its feasibility to the collaboration of multiple departments at UC Berkeley and the exemption granted by the U.S. Copyright Office, influenced by the joint efforts of the UC Berkeley Library and Samuelson Law, Technology and Policy Clinic. Funding support was provided by the Mellon Foundation, ensuring the secure handling of sensitive data through the university's Secure Research Data and Compute platform.
Source: Noah Wire Services