Researchers at Disney Research and the University of Massachusetts Boston (UMass Boston) say they have developed neural networks that assess short narratives to predict which stories will appeal to large audiences.
The researchers note a major challenge in developing automated evaluation of story quality is the lack of large databases of stories that have been assessed by humans and can be used to train artificial intelligence systems.
UMass Boston researcher Tong Wang says they used the question and answer site Quora as a data source because many of its answers are in the form of stories.
In addition, reader upvotes measure a story's popularity and serve as a proxy for narrative quality.
The researchers collected about 55,000 answers and developed an algorithm to classify them as either stories or non-stories. The process produced more than 28,000 stories with an average of 369 words.
During testing, the algorithm exhibited an 18% improvement over a baseline text evaluation system.
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA