To read large volumes of text — pretty exciting, but exhausting. Find in the pile of mediocre tales something worthwhile, too, is not so simple, so developers Disney decided to shift this thankless task on the “shoulders” of artificial intelligence and created an algorithm able to analyze and evaluate texts.
Together with specialists from the University of Massachusetts Boston team Disney Research has created three neural networksthat can work together to read the proposed text in its entirety and find out how good it is. One neural network deals with the analysis of individual parts of the stories, the second evaluates how well all these parts are connected to each other, and the third is already looking at the full text and gives it its own assessment.
“Our neural networks are very good to analyze the text and predict their possible popularity. You won’t be able to use them in a jury of your contest writers, but in our future research, they will definitely come in handy,” says albert Lee of Disney Research.
The latest development Disney Research have shown good results, to cope with the work 18 percent better than their more primitive counterparts. Previously available to the scientists the algorithms could not achieve such impressive results, and the new AI have a great future. According to the developers, the algorithm gradually learns to understand the meaning of reading of the works, assess their quality and, it seems, he will soon learn to write a good fairy-tale books.