You are viewing a single comment's thread from:
RE: Underrated - Yet another list on Steemit
I like the idea of an "Under-rated" tab. Though there's a bit of a mix about what would qualify for this. If we personalised it; things you might like that are probably underrated, then it could work very well.
However, if the metrics are too crude then the metrics can be gamed. Particularly if the metrics are site-wide and not personalised. For example, I can have a minnowbot army upvote to subvert any upvotes / payout metrics.
I think we have a slight disagreement in quality - but I think the difference is semantic so I'll leave it at that.
Quality is what you vote on only your own votes count. And then we measure the distance between new posts and those that you already voted. This list should be personalized as much as possible in both dimensions quality and topic.
Define distance and how you might measure that. That's the little important detail.
Lexical measures of distance are not a reliable substitute for semantic distance - and we'll need semantic distance for determine topics. And, quality, well, I'd argue that quality is a level of difficulty again above semantic distance.
Don't take this to mean I don't think your idea isn't sound - but we shouldn't over expect it'll deliver perfect results. Probably good enough results though.
I plan to publish some experiments about it. Already started with the extraction of data. For semantic distance, I want to play with gemsim and test doc2vec, word2vec or just tf-idf. For quality maybe similar measures like the utopian.io bot can be useful. A bunch of KPIs that might assemble loosely quality and then the distance between those I voted before and he new ones to identify the personal quality. Personal quality should easier to find then absolute.
You seem to know what you're doing. I think you'll be quite successful actually.