A webring of friends
Suppose you need to evaluate a person. Not just for a job or something like that, but you have to see whether he would make a good friend. One that is reliable and subscribes to the same high values you do: honesty, integrity, friendliness, caring, etc. Now the only problem is that you can’t see nor hear the person. The only clue you have is what other people he visits or what people visit his house. Take a moment and think about this problem.
I think most of you would agree that the only way we could guess whether the person would be a good friend or not is to look at what the people he visits and receives are like, and the quantity of those. If every week a drugs dealer comes by, that would be better than having one to visit him every day. On the other hand, it would still be worse than seeing one of your best friends visit him two times a year. But if your best friend visits him almost every day, that would be far better. Or if all of you friends pay a regular visit, you would think there’s a reasonably good chance the guy/girl would make a nice friend. And if noone would visit this guy, and the guy would visit noone himself, you would be able to tell very little about him.
I think this must be somewhat like the task of a search engine. Part of it is determining whether the page is a reliable candidate to deliver good info for the user. They need to do this without being able to actually read the page. They know what words are on them. I can fill an entire page with the search phrase and this would not be your best friend. Most probably this page would have not many other links to it. It wouldn’t have a lot of friends. Or if it has a lot of links to it, they are most likely to be spam links, the drugs dealers of the not-so-best friend.
If you are that guy who needs to be evaluated, you best avoid visiting drugs dealers or letting them into your house. If you have a website, you best avoid linking to pages with dubious intentions or avoid you become the target of them.
The problem is, a search engine has to see at least some people initially to determine whether they are good people or bad people, and go on from there, inferring about other people who visit them or get a visit by them. What pages could a search engine take as a starting point? What pages are most likely to be informative and not to contain spam? If I were to choose those would be government sites and education sites. Some people think links from these domains get you more points (in Google) than links from other sites. I think that from my perspective this makes sense.