Google's DeepMind team last year used articles from the Daily Mail website and CNN to help train an algorithm to read and understand a short story. The researchers used the bulleted summaries at the top of the articles to create simple interpretive questions that trained the algorithm to search for key points.
However, Stanford University researchers say they have developed an algorithm that beats DeepMind's results by 10% on the CNN articles and 8% for Daily Mail stories, achieving 70% accuracy overall.
The improvement came through streamlining the DeepMind model.
The advantage of using Daily Mail and CNN articles is the large number of articles the sites produce, according to University of Illinois at Urbana-Champaign researcher Julia Hockenmaier. She notes the more texts the algorithm learns from, the smarter it becomes.
However, it will be difficult to find or create another large set of texts that come with ready-made questions.
There are still other challenges to overcome, such as the difficulty of tracking the articles on the Internet and determining what information is valid and what is not, according to Carnegie Mellon University researcher Robert Frederking.
From New Scientist
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA