Machine learning is the key to finding patterns in data. That's why companies love it. Patterns help predict the future, and forecasting the future is a great way to make money. It's sometimes unclear how these things fit, but there's a perfect example of a 20th-century Fox movie studio that anticipates whether AI wants to see a movie.
A researcher from the company last month published a paper explaining how to use machine learning to analyze the content of a movie trailer. The machine vision system inspects the trailer footage frame by frame, labels the objects and events, and compares them to the data generated for other trailers. A movie that uses a similar set of labels is an idea that attracts a similar set of people.
As the researchers described in this paper, this is exactly the kind of data that movie studios like. (They already produce many similar data using traditional methods such as interviews and questionnaires.) "Filmmakers investing in stories about uncertain commerciality are important to understand the detailed audience composition,
It would be better if you could divide this audience structure into smaller, more precise "micro segments", a good example being 2017 Logan For example, a superhero movie There is a conspiracy to draw a dark subject and a slightly different audience, so can you use AI to capture these differences? The answer is:
Make "Experimental Movie Attendance Prediction and Recommendation System" (Merlin) For the 20th century, Fox used Google's server and open source AI framework, TensorFlow, In the attached blog post, the search giant explains an analysis of Merlin's Logan .
At first Merlin searches for trailers and searches for "face hair", "cars" "
This graph only records the frequency of these labels, but the actual data generated is more complex: you need to consider how long those objects appear on the screen and how long the trailer will be displayed.
As the engineer at 20th Century Fox explains, this ephemeral information is particularly rich because it relates to the genre of the movie. "For example, a trailer with a close-up of a character is more likely to be a drama movie than an action movie, A quick and often taken trailer is more likely to be an action movie. Logan, If Hugh Jackman's slow shot has a bloody and beaten trailer with many trailers.
Compare this information with the analysis of other trailers to see what movies Merlin can predict Logan but things are a bit tricky here
The graph below is the top 20 movies that people have gone to see Logan . (19659012) Merlin X-Men: Apocalypse
The column of Merlin shows the prediction and the column on the left represents the actual data 19659013] Many of the films are correct, including other superhero movies, such as Dr. Strange and Batman v Superman: Dawn of Justice . John Wick The impressive intuition after John Wick is certainly not a superhero movie, but it is not In the end Merlin identifies all the top five items even if they can not be sorted in order of importance
What does the inconsistency get bigger? I predict Tarzan's legend will make a big hit with fans of Logan Google and 20th Century Fox do not provide a description of this, but found in Logan 39; "Forest", "Tree", & "Light". Tarzan Trailer.
Likewise, Revenant is a drama heavy Oscar bait, not a smart superhero movie with many flora and face hairs. Merlin misses as a bait for audiences like Deadpool 2 Ant – Man and . was a superhero film with a quick-cut trailer, but like Logan's Wolverine cure, he took a bold approach to the heroes.
20th Century Fox already knew all this, but could not see what AI could find. The movie industry is eager to adopt AI for this kind of analysis, and some companies already claim that machine learning can be used to exploit scripts to predict the success of a movie. However, this analysis shows that the computer is not yet a movie lover. They have to stay longer in the movie before they truly predict our taste.