- Reviews should be normalised to a reviewer's mean score. If a person hands out 5/5s like there's no tomorrow, those rate as 2.5/5s.
- The whole notion of collaborative filtering is that some information is (usually) better than none. This is often the case, but in a situation where 100 people are tasked with evaluating a skyscraper, jet airliner, pharmaceutical, or software design, where 1 is a domain expert and 99 are not ... I'd slant my weight in favour of the domain expert, generally. (A jury of experts might be more suitable.) Otherwise, the information is simply noise.
- Massive penalties and downratings for both reviewers and retailers who game the system. Zero those motherlovers out. Put some skin in the game.
Amazon (and every other review site) has a massive reputation problem here, and it will (and is) biting them.
- Reviews should be normalised to a reviewer's mean score. If a person hands out 5/5s like there's no tomorrow, those rate as 2.5/5s.
- The whole notion of collaborative filtering is that some information is (usually) better than none. This is often the case, but in a situation where 100 people are tasked with evaluating a skyscraper, jet airliner, pharmaceutical, or software design, where 1 is a domain expert and 99 are not ... I'd slant my weight in favour of the domain expert, generally. (A jury of experts might be more suitable.) Otherwise, the information is simply noise.
- Massive penalties and downratings for both reviewers and retailers who game the system. Zero those motherlovers out. Put some skin in the game.
Amazon (and every other review site) has a massive reputation problem here, and it will (and is) biting them.