First, we generate benchmarks to identify the expected values. To generate the benchmarks, we take the last 100 posts from a given account and of a given post type (link post, image post, etc.). We drop the top and bottom 25% of those 100 posts, and calculate the mean number of interactions that the middle 50% of the posts have at each age (15 minutes old, 60 minutes old, 5 hours old, etc.).
Then, when a new post comes in from that account, we compare it to that average, and multiply the difference by the weights in each dashboard (if the dashboard has weights set).
For details on how we do that comparison and generate the scores, learn more here.