It’s a problem that has plagued the internet for years: When a news story pops up, it’s great to know it’s true, but how can you know which part is fake news and which part isn’t? And it’s not just on- or off-line sites that can turn readers away, the ability to make and place significant emphasis on one specific component without shifting their attention elsewhere is also likely to be a false economy — something startup Wordflow is attempting to correct. Its aims are to use machine learning algorithms to produce multi-sentence news stories that will remain true to the specifics of any given piece of news, even if numerous references diverge — thus maximizing engagement for a reader while minimizing misunderstandings.
Based in San Francisco, Wordflow is the brainchild of Jim Vo, an enterprise software developer, and Alex Wilhelm, a startup consultant. Vo spoke with the New York Times Magazine about his project:
“Our product uses machine learning to generate sentences that represent as accurate a representation of a complex story as we can realistically get, and then exposes individual articles and stories as a whole. Our team of designers and developers takes the complex image our machine is trying to extract from each single story and helps shape what you will be seeing. We like to call it a small, but extremely powerful, proxy for the source.”
Machine-learning algorithms generate a variable for every individual word of a sentence, and by combining those with structured data, such as images, video, audio clips, or other sources of primary information, the computer is able to make inferences. As there are many factors that an article can include to create a well-formed image of a story, it becomes almost impossible to compare Wordflow articles to those produced by hand-written editorials, news summary articles, or keyword-heavy RSS feeds of several hundred keywords.
On the side of personalization, Wordflow has announced that it will be offering pre-written “write what you know” responses to users who ask them to feature their stories on their sites. When a news story becomes a mainstay on a site’s homepage, users are able to select that story by clicking a “pass to submit” button. From the options available on each page, they can see a summary of both the main headline and text, a headline analysis to help them decide if they’re taking the story seriously, and a “written by” message, which provides further context that draws the reader deeper into the story.
Speaking to Digital Trends, Vo said the ultimate goal was to offer a tool to publishers that would allow them to search for story keywords and assign certain topics to certain stories.
In April, Wordflow will be making two versions of its AI available to the public. A free version will allow publishers to create one sentence that produces up to 10 stories, while a $99 version will allow them to “write however you want” and has the capability to generate up to 50 news stories.