tedster - 11:54 pm on Feb 21, 2011 (gmt 0)
As I see it Google's guiding vision is far different from what we have conventionally understood as IR (Information rRtrieval). It's NOT "the user entered this text, let's find the best documents with that text."
Instead, it's "the user entered this text. Our data shows a certainty of NN% that they actually want this such & such. So let's serve them a kind of mix, using our intention engine assessment as a guide."
So yes, it is quite a mix of factors, and the recipe for any particular SERP may not be the same at all as another SERP. In fact, Google is aiming to drive their user to the best content for their query, and not the content that best lines up technically in any traditional way.
I think I understand where they are coming from. After 18 years, it's pretty clear that many web publishers are not going to get on the technically-accurate bandwagon. But the average user could care less. They just want to get to whatever it is that they hoped to find.
I agree that it's very much a work in progress. Sometimes Google search results can be painfully poor. More often, they just leave something to be desired.
But it was much worse with the pure IR model, even before spamming became what it is today. If you were active in search in the early or mid nineties, you can remember doing a search on Alta Vista and finding 8 out of 10 results that were completely out past left field and over the wall (and in foul territory, too, not a home run.)
I predict that 2011 will continue to see SERPs that we feel are crazy. This is a machine learning model, guided by humans but more and more, I think it's machine learning. They're training their little toddler. And he's not properly toilet trained yet.