I can understand that, but I have regularly found that GM posts, and controversial content tends to acquire more replies than high-quality content.
Taking any interaction at face value in a recommendation system is going to cause issues, over and over. Context matters.
Alternative approach?
Analyzing content based on context.
I had worked on a Nostr content recommendation system about 2 years ago, and it was pretty accurate.
I’m considering eventually resuming this after I finish more important stuff.
Eventually we’ll need to use a system like this. In the meantime, on our Web app we provide alternative trending feeds including a trending w/o “GM” messages
They still use the same metric value system but we‘ve found them to be less spammy
