Celeste Lecompte writes for Nieman Reports about how the Associated Press and assistant business editor Philana Patterson are using automated software to write earnings stories.
Lecompte writes, “The Associated Press’s data is provided by Zacks Investment Research; that company uses human analysts who review Securities and Exchange Commission data, stock pricing, and press releases to build a custom feed of the numbers the AP has requested. That data is sent to Automated Insights, and Wordsmith assembles the stories following the rules Patterson and her colleagues helped set.
“Translating even the simplest data means converting the loose guidelines a human reporter might follow into concrete rules a computer can follow. For example, a human reporter might have a general idea of when a company’s performance was very different from analyst expectations, based on their knowledge of the industry. But for the algorithm, the AP had to specify exact ranges for which the spread between actual earnings and expectations is considered large or small. Wordsmith uses such metrics to decide both which words are used to describe the data and how the story is structured— for example, whether the fact that a company missed analyst estimates should be mentioned in the headline. The story-assembling algorithm uses a predetermined set of vocabulary and phrases (known as a corpus) that follows the AP’s strict stylebook rules.
“‘It’s a lot!’ Patterson says. ‘To come up with a system to trigger the right type of story, we as reporters and editors and programmers have to figure out this stuff ahead of time.'”
Read more here.
