Using neural networks created to simulate the learning processes of human brains, the AI was taught using sample stories to identify particular traits and patterns that would interest to a broad range of readers.
The crew from Disney Research and the University of Massachusetts Boston says it’s the earliest large-scale study of its set on getting bots to assess the quality of texts, and could ultimately lead to AI systems that are ready to tell a good story as well as point one in a crowd of literary tales.
“Our neural networks had fascinating success in predicting the acclaim of stories,” says one of the team, Boyang Albert Li from Disney Research. “You can’t yet use them to pull out winners for your local book competition, but they can be handled to guide future research.”
To get sufficient data for their system to work with, Disney researchers filled in more than 28,320 of the results from Q&A site Quora, concentrating on those with most upvotes as an symbol of quality many of the results on the site are composed in a story-like fashion, so it looked a good fit for the objectives of this experiment.
Objectively deciding the quality of a novel is of course very tricky, so the upvotes added by Quora users acted like a kind of stenography quality measure.
Three neural networks were then formed: one to look at singular chunks of the stories, one to analyze how those chunks fit collectively, and one to evaluate the stories more frequently. As a combination, the materials were able to learn what departed the better stories from the rest.
The test was to see if the neural networks, once exercised, could take a new story from the database and foretell how many Quora upvotes it got a sign that the AI was getting what makes a story popular and what doesn’t.
Take your time to comment on this article.
