AI Can Predict Which Movies Will Succeed Or Fail From Plot Summaries

Interestingly, it is easier to predict what people will dislike than what they will like. Sort of like love and politics. JL

Eva Frederick reports in Science:

Successful movies such as Alice in Wonderland—which scored 80% on the movie-rating website Rotten Tomatoes—have frequent fluctuations in sentiment; unsuccessful ones, fluctuate less. It’s not important whether the films begin or end happily. What’s important is that sentiments change frequently. Sentiment ratings in each summary were compressed into a single score to reflect how often the sentiment changed, a “volatility” measure for movies. Three different methods of arriving at a final score could predict accurately whether a movie would be unpopular.
Artificial intelligence (AI) still can’t see the future, but a new algorithm may come close: Using nothing but written movie summaries, the AI can consistently tell which films will play well—or rottenly—to critics and audiences. If the model can be further refined, it could one day help producers predict whether a movie will be a flop at the box office, before it’s even made.
To test several models, researchers used plot summaries of 42,306 movies from all over the world, many collected from Wikipedia. The models broke up the summaries by sentence and used something called sentiment analysis to analyze each one. Sentences considered “positive,” such as “Thor loves his hammer,” would receive a rating closer to one. And sentences that were considered “negative,” like “Thor gets in a fight,” would be rated closer to negative one.

Generally, successful movies such as 1951’s Alice in Wonderland—which scored 80% on the movie-rating website Rotten Tomatoes—have frequent fluctuations in sentiment; unsuccessful ones, such as 2009’s The Limits of Control, fluctuate less. It’s not important whether the films begin or end happily, the researchers say. What’s important is that the sentiments change frequently.

The sentiment ratings in each summary were then compressed into a single score to reflect how often the sentiment changed—a kind of “volatility” measure for movies. The researchers tested three different methods of arriving at a final score. All three could predict fairly accurately whether a movie would be unpopular, and one method worked especially well for guessing which thrillers and comedies reviewers would hate.The methods were not as efficient at guessing which movies would succeed, but they still predicted the results more accurately than random chance, the researchers reported yesterday at the Storytelling Workshop 2019 in Florence, Italy.
In the future, the researchers say their methods could be refined to predict the amount a movie could earn at the box office and help producers decide which movies to invest in. The system’s impartial judgment might give an advantage to less well-known writers, the researchers add. It could also potentially save the public from having to sit through films like Jaws: The Revenge, which online critics and audiences alike rate as entirely rotten.