3 ms·
The problem with user supplied metadata is users lie. For example, the "keywords" meta field was mainly used for SEO spamming.
by narrator 9y ago
The problem with user supplied metadata is users lie. For example, the "keywords" meta field was mainly used for SEO spamming.
- williamscales 9y agoI dunno, I think as long as you do statistical analysis that's vulnerable to adverse inputs you're going to see the same problem. Meaning that you can infer the keywords and prevent obvious lies only until the users figure out how to trick your keyword machine into inferring the wrong keywords.
- hobofan 9y agoAs someone who only briefly dipped their toes into the semantic web topic, I found Googles' Knowledge Vault[0] to be a very interesting approach to the problem of "everybody lies". [0]: https://en.wikipedia.org/wiki/Knowledge_Vault https://en.wikipedia.org/wiki/Knowledge_Vault
- GistNoesis 9y agoIf you do the counting properly, this can be taken into account. Just like in poker, where you can bluff sometimes but if you do it always, people will call you upon. It's called Reification Done Right (in quadstore terminology), where you store metadata about the statement. Then it's just about building a reputation score for each statement. You can use simple bayesian updates like it's done in spam filters. Pretty quickly there is very little incentive to lying in an open book world. Even more interestingly you can compute a score between author of the statement and viewer of the statement, basically telling you whether or not you can trust the information given your current belief system. For example if you are a flat-earther, you can get spaceX news filtered quite easily. From the computational point of view it's faster not to compute the whole matrix of score but use some lower dimensional embeddings so you do a matrix factorization, and that's how you get all the user recommendation systems ala Netflix.