Researchers from film studio heavyweight 20th Century Fox have published a paper which reveals it is using machine learning to predict what films people will want to see.
Fox has created Merlin, an “experimental movie attendance prediction and recommendation system” that can extract individual frames from movie trailers and then use machine learning to label items on the screen.
By analysing the content of movie trailers, Merlin was able to identify patterns that would help the studio to better understand audience composition.
Merlin can identify items such as trees, faces, cars and events, which it then compares to data generated for other trailers using information such as movie attendance figures. By doing this it is hoped it will be able to identify patterns in the labels that correlate to box-office hits.
The system can also analyse how long objects are on screen and when they appear in the trailer. Fox engineers explain that this temporal information is particularly rich because it correlates to the movie’s genre.
Researchers summarised their theory, stating: “In other words, if they [movie studios] know who watches what, they will know what movies to make.” It is this detailed sort of data that movie studios seek to augment the information they already have from more traditional means of research such as interviews and questionnaires.
Superhero Movie Logan Used to Test Merlin
The studio tested Merlin using the 2017 superhero movie, Logan, as a trial. Merlin was fed data from the Logan trailer, which it then used to predict what other films might interest those who went to see production.
Merlin was able to successfully identify all of the top five picks, however, it did fail to put them in the same order of importance. It correctly identified John Wick: Chapter 2 as a movie associated with Logan, which is impressive considering John Wick is not a superhero film. However, both films feature handsome, jaded men with excellent fighting abilities and would both be considered darker films.
There were a few mismatches in Merlin’s predictions, which saw Logan paired with Tarzan and The Revenant, but this is likely due to similar tags such as “tree,” “forest,” and “light,” which all feature in their respective trailers.
While Merlin is not yet perfect, the system reveals the movie industry’s willingness to adopt artificial intelligence or emerging technologies for this sort of analysis.
Some companies already claim they can predict the success of a movie using AI to analyse scripts. In the future, this could influence how movies are marketed and even which projects are given the green light.