Camera, motor, Big Data: how film studios are looking for new films using AI
The success of new products with the audience can be predicted using machine learning. This method of forecasting in recent years has been used in the movie business. For information on how to find out with the help of AI, whether to release a new film and how to advertise it, see the Binary District review.
The problem of choosing scenarios and promoting films in recent years has become particularly acute for developed countries. According to the Association of American Cinema (MPAA), the volume of the European film distribution market in 2017 amounted to 10.1 billion dollars, almost returning to the level of 2011 (10.6 billion).
The market of the United States and Canada also reached maturity: in 2015 it grew to 11.1 billion dollars and by the end of 2017 remained at the same level. At the same time, film production continues to grow: in 2017, they were released 8% more than in the past.
One of the biggest failures in 2018 was the film "Han Solo" studio Walt Disney. With a budget of 250 million dollars in the world box office, he earned only 213 million. They say that the reason was in marketing: the teaser came out late, and the advertising campaign started working at full capacity just a month before the premiere of the film.
The question of how to avoid such failures, worries not only producers, but also directors, screenwriters, scientists and entrepreneurs who are trying to figure it out with the help of machine learning. We selected three best examples of such experiments.
One of the first ways to predict the success of a future film with the help of artificial intelligence was the analysis of the texts of screenplays. They were used by ScriptBook , founded in Antwerp in 2015. A year later, the startup attracted $ 1.4 million in investment, and in the summer of 2018 - presented its technology at the Karlovy Vary film festival.
The principle of operation is simple: a PDF file with a script is loaded into the system, and after five minutes it gives a detailed report. The car predicts the age rating, analyzes the characters, indicates the protagonists and antagonists, assesses the emotionality of each character, predicts the audience of the project - including gender and race, and also gives an estimate of the possible box office.
Analysis of the movie "Passengers" using ScriptBook
AI ScriptBook analyzed films that have been released since 2015. According to the developers, he managed to isolate hopeless projects three times better than people did. For this, the AI was given to study the 62 Sony films released during this time, 32 of which failed at the box office.
It turned out that the unprofitableness of the majority of "films-losers" could have been foreseen: ScriptBook predicted a failure for 22 of these 32 films. And to all 30 films that showed themselves well at the box office, the system gave the green light. Before this, the AI managed to practice on 6500 existing scenarios and determined the success of already released projects with an accuracy of 84%.
The algorithm had the best result with the movie “Passengers”, which premiered in the USA in December 2016. Artificial intelligence was almost not mistaken in the financial result: the drama about the journey of the spacecraft in the future gained 110 million dollars instead of the predicted 118.
But the musical "La La Land" AI underestimated almost 8 times: a romantic film with Ryan Gosling in the lead role collected at the box office 446 million dollars instead of the predicted 59 million. True, ScriptBook just gave him the "green light", thanks to a small production budget - only 30 million dollars.
"Quiet place", a frame from the film.
Many fear that such systems will kill the author's cinema, which may not be so profitable and move away from the usual patterns. But, as the ScriptBook blog developers point out, the main thing for analysis is not income, but ROI. In this regard, the film “Quiet Place”, which earned 163 million dollars with a budget of 17 million, looks more profitable than “Fast and the Furious,” attracting a billion, but costing 250 million dollars.
One scenario is still not enough to understand who might like the film and which audience is better to advertise it. Trailers are another matter, say at 20th Century Fox. The development, which the studio presented in November 2018, was named Merlin Video. This is a neural network that categorizes the trailers of different films into categories and analyzes how often the same objects are found in them. As a result, according to the film studio, the AI gives a twice more detailed assessment of the audience than all other methods.
The leaders of the Data Science team at 20th Century Fox talked about how it all works in a scientific article on the Cornell University website and in a blog . For example, they took the film "Logan" - the third and last fantastic Marvel fighter about the harsh fighter named Wolverine, released in 2017 and collected 600 million dollars at the box office (the film’s budget was six times less).
What other films could go viewers who like these pictures? What is so popular with the public this hero? The neural network answered this question after it managed to determine which objects are found in the trailer most often. As Merlin noted, the main thing in Wolverine is a beard. Among the other signs were also "tree", "facial hair", "car" and "man".
In addition to the frequency of repetition of details, the neural network takes into account the time on the screen. Trailers with long close-ups of characters are more likely characteristic of dramatic films. But trailers with fast and frequent scene changes are for action films, the authors of the project noted.
After that, Merlin collected data on the popularity of movies, box office, the number of their views on the Internet and began to compare. According to the algorithm, the film studio combined the twenty films that the fans most likely would have “logged in” with the results of the polls of real viewers. In half the cases, the AI guessed right.
Among the coincidences, there were not only superhero films - such as "X-Men: Apocalypse", "Doctor Strange" and "Batman vs. Superman: Dawn of Justice". Merlin did not ignore the crime thriller "John Wick 2" - it is the same as in Logan, a brutal hero of an exhausted look.
The discrepancies, however, were more revealing: for example, the algorithm considered that the audience of “Logan” should like “Tarzan” - apparently because of the abundance of trees in the cinema. But he missed such obvious recommendations as “Ant-Man” and “Deadpool” - their trailers were less dramatic.
The algorithm was constantly used in the 20th Century Fox after the release of the musical "The Greatest Showman" in November 2017. The model built by the developers, in addition to analyzing trailers using Merlin Video, also includes studying the text of the screenplay using Merlin Text (for more details, see the article on the Cornell University website).
The forecast for the audience of the film can be made both for already released tapes, and for those who are still being prepared - 6-8 months before their official release. To find out how this prediction coincided with reality, data on ticket sales in cinemas is used. Now the analysis is supplemented with data on which films users buy and rent online.
Commercials and scripts are not the only thing that can determine the financial success of films. Specialists from the University of Iowa in 2015 found that to reduce the budget of motion pictures - and improve its profitability - can be at the stage of its production. To analyze the data, scientists created a database of 4,000 films released in 2000–2010 and taught the program to search for features common to commercially beneficial tapes.
The feature set consisted of four groups:
Then, using artificial intelligence, researchers tried to identify signs that most clearly correlate with the return on investment. To estimate the last parameter, we used the value (revenue - budget) / budget.
It turned out that the main sign of success is the director’s fame and the earnings of his previous films. The celebrity of the actors guaranteed revenue growth of almost half (46%), but the return on investment was no more than 17%. Of course, the audience will be more eager to go to the cinema for famous actors, but in order to invite them, much more money must be spent, the researchers note.
The study did not take into account that film companies may have other sources of income besides ticket sales in cinemas. So, the Disney company sells toys and other goods based on the most successful tapes. But such revenues also depend on the popularity of the movie with the audience.
You can learn how to use face recognition, neural network and machine learning in business at the AI for Business intensive course. The course speakers from Microsoft, Nanosemantics and Home Credit Bank will tell you how to use different types of AI and what tools are available for this. The nearest intensive will take place on March 30-31.
Why AI studios
The problem of choosing scenarios and promoting films in recent years has become particularly acute for developed countries. According to the Association of American Cinema (MPAA), the volume of the European film distribution market in 2017 amounted to 10.1 billion dollars, almost returning to the level of 2011 (10.6 billion).
The market of the United States and Canada also reached maturity: in 2015 it grew to 11.1 billion dollars and by the end of 2017 remained at the same level. At the same time, film production continues to grow: in 2017, they were released 8% more than in the past.
One of the biggest failures in 2018 was the film "Han Solo" studio Walt Disney. With a budget of 250 million dollars in the world box office, he earned only 213 million. They say that the reason was in marketing: the teaser came out late, and the advertising campaign started working at full capacity just a month before the premiere of the film.
The question of how to avoid such failures, worries not only producers, but also directors, screenwriters, scientists and entrepreneurs who are trying to figure it out with the help of machine learning. We selected three best examples of such experiments.
How to make the AI read between the lines
One of the first ways to predict the success of a future film with the help of artificial intelligence was the analysis of the texts of screenplays. They were used by ScriptBook , founded in Antwerp in 2015. A year later, the startup attracted $ 1.4 million in investment, and in the summer of 2018 - presented its technology at the Karlovy Vary film festival.
The principle of operation is simple: a PDF file with a script is loaded into the system, and after five minutes it gives a detailed report. The car predicts the age rating, analyzes the characters, indicates the protagonists and antagonists, assesses the emotionality of each character, predicts the audience of the project - including gender and race, and also gives an estimate of the possible box office.
Analysis of the movie "Passengers" using ScriptBook
AI ScriptBook analyzed films that have been released since 2015. According to the developers, he managed to isolate hopeless projects three times better than people did. For this, the AI was given to study the 62 Sony films released during this time, 32 of which failed at the box office.
It turned out that the unprofitableness of the majority of "films-losers" could have been foreseen: ScriptBook predicted a failure for 22 of these 32 films. And to all 30 films that showed themselves well at the box office, the system gave the green light. Before this, the AI managed to practice on 6500 existing scenarios and determined the success of already released projects with an accuracy of 84%.
The algorithm had the best result with the movie “Passengers”, which premiered in the USA in December 2016. Artificial intelligence was almost not mistaken in the financial result: the drama about the journey of the spacecraft in the future gained 110 million dollars instead of the predicted 118.
But the musical "La La Land" AI underestimated almost 8 times: a romantic film with Ryan Gosling in the lead role collected at the box office 446 million dollars instead of the predicted 59 million. True, ScriptBook just gave him the "green light", thanks to a small production budget - only 30 million dollars.
"Quiet place", a frame from the film.
Many fear that such systems will kill the author's cinema, which may not be so profitable and move away from the usual patterns. But, as the ScriptBook blog developers point out, the main thing for analysis is not income, but ROI. In this regard, the film “Quiet Place”, which earned 163 million dollars with a budget of 17 million, looks more profitable than “Fast and the Furious,” attracting a billion, but costing 250 million dollars.
Why watch trailers with AI?
One scenario is still not enough to understand who might like the film and which audience is better to advertise it. Trailers are another matter, say at 20th Century Fox. The development, which the studio presented in November 2018, was named Merlin Video. This is a neural network that categorizes the trailers of different films into categories and analyzes how often the same objects are found in them. As a result, according to the film studio, the AI gives a twice more detailed assessment of the audience than all other methods.
The leaders of the Data Science team at 20th Century Fox talked about how it all works in a scientific article on the Cornell University website and in a blog . For example, they took the film "Logan" - the third and last fantastic Marvel fighter about the harsh fighter named Wolverine, released in 2017 and collected 600 million dollars at the box office (the film’s budget was six times less).
What other films could go viewers who like these pictures? What is so popular with the public this hero? The neural network answered this question after it managed to determine which objects are found in the trailer most often. As Merlin noted, the main thing in Wolverine is a beard. Among the other signs were also "tree", "facial hair", "car" and "man".
In addition to the frequency of repetition of details, the neural network takes into account the time on the screen. Trailers with long close-ups of characters are more likely characteristic of dramatic films. But trailers with fast and frequent scene changes are for action films, the authors of the project noted.
After that, Merlin collected data on the popularity of movies, box office, the number of their views on the Internet and began to compare. According to the algorithm, the film studio combined the twenty films that the fans most likely would have “logged in” with the results of the polls of real viewers. In half the cases, the AI guessed right.
Among the coincidences, there were not only superhero films - such as "X-Men: Apocalypse", "Doctor Strange" and "Batman vs. Superman: Dawn of Justice". Merlin did not ignore the crime thriller "John Wick 2" - it is the same as in Logan, a brutal hero of an exhausted look.
The discrepancies, however, were more revealing: for example, the algorithm considered that the audience of “Logan” should like “Tarzan” - apparently because of the abundance of trees in the cinema. But he missed such obvious recommendations as “Ant-Man” and “Deadpool” - their trailers were less dramatic.
The algorithm was constantly used in the 20th Century Fox after the release of the musical "The Greatest Showman" in November 2017. The model built by the developers, in addition to analyzing trailers using Merlin Video, also includes studying the text of the screenplay using Merlin Text (for more details, see the article on the Cornell University website).
The forecast for the audience of the film can be made both for already released tapes, and for those who are still being prepared - 6-8 months before their official release. To find out how this prediction coincided with reality, data on ticket sales in cinemas is used. Now the analysis is supplemented with data on which films users buy and rent online.
How to find out who will be the best advertisement for a movie
Commercials and scripts are not the only thing that can determine the financial success of films. Specialists from the University of Iowa in 2015 found that to reduce the budget of motion pictures - and improve its profitability - can be at the stage of its production. To analyze the data, scientists created a database of 4,000 films released in 2000–2010 and taught the program to search for features common to commercially beneficial tapes.
The feature set consisted of four groups:
- “Who” - this group included the parameters of success of the director and actors;
- “What” - the genre and age rating of the film;
- “When” - what time of year the film came out and how successful this year was for the film industry;
- a group with “hybrid characters”: for example, were guest actors cast before or did they appear in this genre before?
Then, using artificial intelligence, researchers tried to identify signs that most clearly correlate with the return on investment. To estimate the last parameter, we used the value (revenue - budget) / budget.
It turned out that the main sign of success is the director’s fame and the earnings of his previous films. The celebrity of the actors guaranteed revenue growth of almost half (46%), but the return on investment was no more than 17%. Of course, the audience will be more eager to go to the cinema for famous actors, but in order to invite them, much more money must be spent, the researchers note.
The study did not take into account that film companies may have other sources of income besides ticket sales in cinemas. So, the Disney company sells toys and other goods based on the most successful tapes. But such revenues also depend on the popularity of the movie with the audience.
You can learn how to use face recognition, neural network and machine learning in business at the AI for Business intensive course. The course speakers from Microsoft, Nanosemantics and Home Credit Bank will tell you how to use different types of AI and what tools are available for this. The nearest intensive will take place on March 30-31.
Source: https://habr.com/ru/post/436988/