Google leverages BERT to match fact checks with news stories

By Miles Deans | 18 Sep 2020

Google announced in its latest update that it will be leveraging BERT, one of its language artificial intelligence (AI) models, to better match news stories with fact checks.

BERT represents an advancement in natural language processing (NLP), a branch of AI that deals with linguistics. It was designed with the intention of understanding more complex keywords and search terms by analysing an entire phrase and looking at the context in which a keyword is used.

What are fact checks and what do they look like?

Fact checks determine whether claims or statements made by publishers are true, false, or partly true.

The way in which they appear varies depending on which Google search feature is used, e.g Google Search, Google Images or, in this case, Google News.

  • For search, results that have been fact-checked will be shown in a box detailing the claim being checked, who made the claim, who is completing the fact check, and a summary of the fact check.
  • For images, Google has introduced fact check labels. The name of the domain doing the fact check may also be present.
  • On news articles, fact checks are labelled ‘Fact Check.’


Why is the BERT update helpful for users?

BERT will help users to understand whether or not stories within Google News Full Coverage are reliable in terms of the facts. The process will involve Google assessing to see if there are connections between articles and the fact check database, in order to better match fact checks with stories. This will help ensure that fact checks are related to main topic stories and will allow users to absorb more reliable content. BERT’s process for dissecting terms will improve results by making better matches between articles within the news section and the fact check database.