This is a nice case study on why it's good to predict distributions instead of raw values. Low comment accounts should have a high degree of uncertainty, which

This is a nice case study on why it's good to predict distributions instead of raw values. Low comment accounts should have a high degree of uncertainty, which should translate into weaker similarity scores, of you compute the expected similarity of two accounts.