My latest interview is with Google's Sep Kamvar. We talked about the personalization algorithms currently in use by Google in detail. Sep spells out in detail for us:
The two signals that we use right now are the search history and the location. We constantly experiment with other signals, but the two signals that have worked best for us are location and search history.
We do talk quite a bit more about the types of signals they experiment with. Net-net most of the other signals they have experimented with are "noisy" in nature. What he means by that is that their test has shown that the input from such signals does not really help them improve the quality of search results for their users.
It underscores the fact that it's not as simple as we are all inclined to think. There are lots of things that we can guess make for good signals for a search engine to use. But many times, these signals really don't match up with a user's search intent.
For example, just because a user indicates a personal preference for something, that doesn't mean that this really related to what they are searching for at a particular time. In fact, when Google tried to get users to specify their interests, they found that it was not a good signal for them to use.
The key element really is what they user's intent at the time they perform the search. Turns out, that is not that simple to determine.