A lot of online engagement data looks like a power law distribution on large platforms. The mean (obviously) has huge variance. Say you have multiple metrics pe

A lot of online engagement data looks like a power law distribution on large platforms. The mean (obviously) has huge variance. Say you have multiple metrics per user like this. You can still apply k-means but the huge variance makes it rather ineffective. That's what I've found, anyway.