Researchers at Salesforce have been busy developing an algorithm that proves just how well a computer can summarize documents. It’s a fantastic tool that will save people a lot of time and could soon become the norm to rely on a machine to sum something up for you. The algorithm uses various machine-learning tricks to do this and although it’s not quite as good as a person yet, it is something that could very well become automated soon.
The following summary demonstrates what the algorithm is able to produce and was taken from an article in the New York Times about Facebook combating fake news:
Social network published a series of advertisements in newspapers in Britain on Monday.
It has removed tens of thousands of fake accounts in Britain.
It also said it would hire 3,000 more moderators, almost doubling the number of people worldwide who scan for inappropriate or offensive content.
This new algorithm that’s been developed is far better than anything else out there does a similar job. Richard Socher is chief scientist at Salesforce and he says, “I don’t think I’ve ever seen such a large improvement in any [natural-language-processing] task.” Although the team recognizes the algorithm isn’t perfect they’re still confident of its abilities and could even be used in the future on Salesforce’s own platform. In order to generate its own summaries, the system experiments through a process of reinforcement learning, inspired by the way in which animals learn.
Kristian Hammond is a professor at Northwestern University and founder of Narrative Science and feels that while the Salesforce research is a good step in the right direction it also shows how limited machine learning still is. “At some point, we have to admit that we need a little bit of semantics and a little bit of syntactic knowledge in these systems in order for them to be fluid and fluent,” says Hammond. Startup company Maluuba was acquired by Microsoft earlier this year and has recently produced its own system that can generate a series of questions from the text by using both reinforcement learning and supervised learning.
More News to Read