A classifier has been able to predict who you’ll like to watch on TV by studying a handful of popular series that haven’t aired yet.
The research, led by researchers at the University of Illinois, Chicago, is part of an effort to improve the accuracy of TV classifiers.
The team used deep neural networks (DNNs) to train an algorithm on tens of thousands of TV series, and then analyzed the results to predict how many people would like to see the show they watched in the future.
The algorithm was able to correctly predict that viewers of popular shows like The Big Bang Theory and Orange Is the New Black would watch the show in the near future.
“There’s a lot of shows that are popular right now that have been shown on Netflix, but they’ve never been shown in a large enough number of places to have that degree of predictive power,” said lead author David S. Haddad, a professor of computer science at the UIUC.
The team also created a list of popular TV shows with the most popular episode titles, which they fed into the DNN.
In some cases, they also used data from the show to predict viewers’ likelihood to watch the episodes.
The researchers found that the DLLs trained on the show predicted that the episode title was a factor of 4.5, which is roughly 3.5 times the accuracy a human can achieve on its own.
But that doesn’t necessarily mean that the network is 100 percent accurate.
The results could help inform TV producers, who are in the midst of making TV shows like Homeland and Mad Men that can be easily tailored to the needs of each user.
“I’m hoping that we’re going to see more shows like this come to market in the next few years,” said co-author Andrew D. Smith, an assistant professor of machine learning at the school.
“These DNNs are incredibly good at predicting what you’re going, say, to watch, and there are a lot more things that you can do with them,” he added.
“It’s a pretty powerful tool.”