A problem with link-based reputation systems
Generally, the more links that a site (or blog) gets can be used a measure of the importance of the site in question. Links are one type of vote for the site's reputation (see my previous post about
Technorati as a reputation system). One problem with this is that
we sometimes link to things that we don't like. Sometimes we link to things to express our disagreement, anger, or outrage. These types of links really should detract from the reputation of the site in question -- but how can a link-based reputation system tell the difference between a good link and a bad link? Humans can do this, by reading the adjacent text and making a subjective determination as to the authors opinion of the link destination. Is it possible to design an automated reputation system, that can understand the semantics of a link and its surrounding text? I suppose such a system could attempt to achieve this by spidering the text and scanning for known positive and negative words, and, considering the proximity of those words to the link itself, calculate a probability that the link has a postive (thus increasing reputation) or negative (thus decreasing reputation) intent. I guess you could take this concept a step further and assign a -10 to +10 reputation weighting to every link. I wonder how close such a system would come to the human interpretation of link-reputation intent.
