20th Century Fox and Google have developed a technology that predicts whether those who watched the trailer will go to the cinema
IT specialists from 20th Century Fox and Google Cloud have developed a machine-based technology that analyzes trailer views and predicts the likelihood that people will go to these films and "related" pictures to the cinema.
The technology itself was named Merlin. The system recognizes objects and trailer plot patterns to “understand” the picture itself. Merlin scans trailers and identifies objects like “a man with a beard”, “gun”, “machine”, deciding what kind of picture it is and in what context the object is shown.
For example, if the protagonist is shown long enough / often close-ups, most likely the film's genre is drama. When the characters are shown in fragments and the frames often change, the film is most likely an action. Merlin is able to predict all sorts of catastrophes - for example, if a car moves too fast, there will likely be an accident with an explosion. Well, chases with accidents - this is clearly an action.
Next, the system “comes up with” tags for each of the pictures and groups the tags into categories. At the moment, Merlin analyzed the stony films and millions of groups of data on the attendance of each film by viewers in cinemas. According to the developers of the system, it is completely anonymous, the data is depersonalized. But it is unclear exactly what data on visits are used for analysis. But it is known that this information is analyzed together with “basic demographic” data at the individual level.
All this is necessary in order to be able to predict the popularity of a particular picture based on its genre and the most frequently encountered objects (a man with a beard, a car and a pistol. The film company uses this to forecast ticket sales in cinemas.
In addition, the film company is going to use Merlin in order to study the best marketing methods for a particular picture.
According to some experts, such a system will not work normally, simply because, in addition to various kinds of objects found in films, pictures are popular or unpopular due to a number of other factors. It can be jokes, hints, play of actors, special effects and much more. Yes, the computer can recognize the "man with a beard", but this is not the main factor in the success of the film.
Tags for the movie "Logan" with a maximum repetition rate
Merlin is able to predict whether those who have watched a particular trailer will go to other films with similar objects and plot. But sometimes there are inconsistencies. For example, for the film “Logan”, Merlin predicted that its viewers would go to movie theaters for films like The Magnificent Seven, Jason Bourne, John Wick 2 and The Legend of Tarzan. Actually, here it is clear that in the first three pictures there are “men”, “beards” and weapons. But in the case of Tarzan, the forecast was made only on the basis of the "trees" tag, which are abundantly enough in Logan.
According to the developers of Merlin, the platform is able to predict correctly about 50% of the films that can go to those who watched the trailer for a particular picture. Based on these data, the company can plan the costs of marketing campaigns and the configuration of campaigns to promote films.
So far, the platform has a number of drawbacks. For example, it is likely that the viewer who watched "Logan" will watch other films related to superheroes with interest. But in the list of probabilities of Merlin, no such picture is indicated, which indicates the inability of the platform to analyze the plot of the picture and recognize its main characters.
On the other hand, all this is only the first experiments. Companies will improve their platform, teaching it new tricks of analysis of pictures. And there is no doubt that success will be achieved with the capabilities of the 20th Century Fox and Google Cloud.