October 2017, Washington Post:

“The logic of CrowdTangle’s model is relatively simple (even if the underlying math and software code gets complicated). CrowdTangle tracks clusters of Facebook pages and specific keywords. It gathers historical data on how stories, posts and images tend to perform on these sites, and then highlights the stories, posts and images that are doing best against their own expected baseline performance rate. [The] company then packages this information into a daily email, alerting [its] clients to the content which is likely to perform best on a day-to-day basis.”

Jonathan Albright’s Medium account has some great material.