Artificial Intelligence

20th Century Fox has revealed that it will use machine learning to analyze its trailers to determine who will want to see its movies.  An artificial intelligence– or more specifically an “experimental movie attendance prediction and recommendation system” which they call Merlin– will examine trailers frame by frame and label certain “objects and events” and then compare them to other trailers.  The idea here is that trailers with similar labels will attract similar viewers.  By comparing this data, Fox can determine what kinds of movies to make, that will attract theatergoers.  All studios use similar data but gathered the old fashioned way with interviews and questionnaires with actual humans.

The information gleaned from AI analysis is particularly helpful with movies that are harder to classify.  It seems the more narrow a movie can be classified, the better.  If audiences can be broken down into “micro segments” the studio can better understand how to make and market the movie.  The example given is ‘Logan’, which is a superhero movie, but was rated-R and darker and more violent than traditional fare.

Here is what Merlin discovered when it analyzed the ‘Logan’ trailer:

As you can see, the ‘Logan’ trailer contains a large number of trees, facial hair, cars and men.  For whatever reason, although there are many trees, there aren’t as many forests and while there is a lot of facial hair, not all of it was beards.

This information is broken down even further– how long these objects appear in the trailer and at what point they pop up.  As Fox analyst wrote in a report, “For example, a trailer with a long close-up shot of a character is more likely for a drama movie than for an action movie, whereas a trailer with quick but frequent shots is more likely for an action movie.”  The ‘Logan’ trailer indeed included a number of dramatic closeups of Hugh Jackman’s beaten bloody face and is more like a drama than the typical superhero movie.

Below is a chart.  On the left are actual movies that people that watched ‘Logan’ also saw.  On the left are the movies that Merlin predicted that fans of ‘Logan’ would also like.

As you can see, Merlin failed to recommend superhero films like ‘Deadpool’, ‘Ant-Man’, and ‘Captain America: Civil War’, and dramas like ‘Arrival’ and ‘Split’, while it recommended movies that the humans didn’t see, like ‘Jason Bourne’, ‘The Legend of Tarzan’, ‘The Revenant’, ‘Spectre’ and other straight-forward action flicks.

If you’re curious as to why Merlin might suggest, for example, ‘The Legend of Tarzan’, look back at the first graph and notice that the trailer for this movie also contains plenty of trees, atmosphere, forests and light.

So clearly this isn’t a perfected technology yet, but over time, it’s likely that Merlin and similar systems may become more refined and accurate in their predictions and movie studios can certainly use all the information they can get in this highly competitive field.

Source: 20th Century Fox