I’m not sure about the mathematical details, but as described in their FAQ, they presume that it’s inevitable that people will form into local Blue and Green tribes, so they attempt to cluster the population into Blue and Green to not just be a better recommendation engine to both Blues and Greens, but also calculate a nonpartisan score of upvotes by the other side and downvotes by your side.
In general, I thought this was fascinating because it gets to the heart about what voting is for on social websites. If we’re trying to build a recommendation engine, having an extremely diverse set of viewpoints is probably something that we want in the input stream of links and discussion. However, we then don’t want to have everyone’s voting then represent a single score variable, because people are different and have different worldviews. Mixing everyone’s scores together will make a homogenized mess that doesn’t really speak to anyone.
The idea of tracking partisanship not just to Bayes voting to make better recommendations to users, but to get a sense of nonpartisan quality really impressed me as an idea that’s totally obvious...in retrospect. I do wonder how well it scales, as Omnilibrium is fairly small right now.
I’m not sure about the mathematical details, but as described in their FAQ, they presume that it’s inevitable that people will form into local Blue and Green tribes, so they attempt to cluster the population into Blue and Green to not just be a better recommendation engine to both Blues and Greens, but also calculate a nonpartisan score of upvotes by the other side and downvotes by your side.
In general, I thought this was fascinating because it gets to the heart about what voting is for on social websites. If we’re trying to build a recommendation engine, having an extremely diverse set of viewpoints is probably something that we want in the input stream of links and discussion. However, we then don’t want to have everyone’s voting then represent a single score variable, because people are different and have different worldviews. Mixing everyone’s scores together will make a homogenized mess that doesn’t really speak to anyone.
The idea of tracking partisanship not just to Bayes voting to make better recommendations to users, but to get a sense of nonpartisan quality really impressed me as an idea that’s totally obvious...in retrospect. I do wonder how well it scales, as Omnilibrium is fairly small right now.