AI Now Watches Movies Trailers And Predicts What Movies You Will Like

The key of machine learning is analyzing huge databases, searching for patterns, and implementing that in selecting future examples to fit those patterns. And 20th Century Fox doesn’t miss this chance to earn more profits from movies.

Sometimes, studios take risks to earn a good deal of money. However, they usually invest in the types of films that have already succeeded. Previously, studios would collect data on ticket sales, demographic figures, and surveys in which they directly ask people about their movie preferences. But now that we have AI, 20th Century Fox thinks the process can be upgraded.

With this idea in mind, they have tied up with Google to use its servers and TensorFlow that is Google’s open-source AI framework. They are building Merlin with this framework.

Merlin, which was named after a wizard similar to the legendary King Arthur, watches through every frame in movie trailers, then picks out events and objects. Next, Merlin compares them with other trailers, searching for common features which may attract people watching the first trailer to watch the second too.

If AI can make predictions on people’s favorite kinds of films, studios will know which films to produce.

Take “Logan” (2017) for example, it’s basically a superhero movie. However, it’s also a movie with darker tones, a beaten-down main character, and a sad story behind. The audience that favor this movie would be different from the one that like “Superman”.

As illustrated in the graph below, Merlin scanned and picked out some objects such as “facial hair”, “forest” and “car”.

The graph only indicates their appearance frequency, but in fact, there are more variables being measured, such as how long they remain on screen and when they appear in the trailer. For example, the facial hair showed up at a close look and remained for a long time. Although the trailer is filled with quick cuts that hint an action movie, it is more likely a drama one.

Unluckily, movies are not always exactly that way.

The next graph indicates the top 20 movies that Logan’s watchers also watched in the cinema. In the left column are the movies predicted based on surveys and the right ones illustrates Merlin’s projection. It is shown that the AI does some good guesses with similar superhero movies namely “Doctor Strange”, “Suicide Squad”, “X-Man: Apocalypse” and “Batman v Superman”. Notably, it is also right about the movie “John Wick: Chapter 2”. Though this movie is not about superheroes like “Logan, it features a beaten and bloody character along with a series of actions and dramas.

Yet, the framework lost some movies. It predicts the targeted audience would also watch “Tarzan” which includes forest and trees features. The same with “The Revenant”, despite a similar setting, the film is actually a drama-heavy Oscar movie, which differs from “Logan”. And Merlin also misses some other films such as “Deadpool” and “Ant-Man”, perhaps because their trailers have much faster scenes that look more like action movies.

Though it’s obviously still in work, Merlin will soon develop to make accurate predictions on most people’s movie choices. This technology may help studio to decide on more reliable products, as well as make them more willing to take risks and try new franchise that we may like.