Using Deep Learning to Analyse Movie Posters for Gender Bias
co-author: Aiyaz Miran, Mohammad Shahebaz
It has been historically documented how pop culture and societal norms relevantly play out each other, often mirroring the behavioral trends in the population at large. This becomes more obvious when the creative producers have to market the content, as they are often challenged with creating gripping sneak-peaks of their work to keep the foot traffic high, and in doing so the most interesting characteristics of human psychology intertwined with uncomfortable societal trends come alive for us to ponder.
To find the truth to this thought, the journey started with a compilation of a data-set consisting of 21,000 posters of the movies released since 1970 and its auxiliary data like their IMDB scores and whole movie credits. The idea was to see how the posters were composed of a given movie. Two different AI-models were used to first detect faces(even animated ones!) and to perform facial recognition to identify the gender from the faces featured in the poster. It is given that if a movie has only a male star featuring in its poster, in good faith it is most likely a movie about man/men and vice versa, having said that the findings stand to the point that even if that isn’t the case, then the gender suppression has occurred and analysis still stands. On further holistic analysis of the movie posters with the contextual background data like the movie description, crew/credits, and such, pretty fascinating insights of gender bias were discovered.
Historically critically acclaimed movies were having a predominantly male presence in the posters, but that changed. It is no coincidence that the most highly rated IMDB movie poster has a major male presence, while the lowest-rated movie features a female presence. Trends also point to the fact that before the 2000s, 7 and above rated movies predominantly had mostly male in the posters, but that changed during the post-2000 era where the female actors started getting more limelight in the covers. Feverishly engaging movies like Thor (2011), Spider-Man 2 (2004), got duly equal presence for both the gender roles, changing the trends for good. The diagram also shows that the highest-grossing movies have more men featuring in their posters, and that is almost always. Since the 1990s to 2010s, one trend that hasn’t shied away from the fact of deep gender bias in the film community and society is that the sum of revenues from movies about men or men featuring in the posters has always bagged most money to their banks when compared to movies about women. There is not one single year or instance, that dares to question this trend, although, since the year 2007, the gap is considerably shrinking.
Distribution of genders across different genre
Movie posters of certain genres are the best example of rampant gender stereotypes. The infamous gender stereotypes are unexpectedly amplified when seen the material through genres. Movies featuring topics like “war”, “sci-fi” and “sport” are unfairly biased having men in their stories and posters, while topics like, “animation”, “family”, and “romance” slyly feature female crew in the posters. It is certain that movie posters are heavily stereotyped, that honestly might not be just the posters, as much as it is about the stories the producers and filmmakers tell as well.
Distribution of genders across different Production House
More things to try
Usually, multiple posters are released for the same film in different countries to cater to local needs. One exciting analysis could include how different genders are represented in the same film in different countries.
Another interesting study could be to identify not only the gender but also the race and age of the person in the poster. This information, along with the other metadata like the details of the cast and crews members, could be used to study the diversity and inclusion in the movie industry.