On learning your search preferences

New Scientist reports on experiments that give search engines metadata about the person searching, with the hope of increasing the relevance of search results. It seems that some of it’s promising, though some relies a lot on, say, all young white males being roughly alike, at least in terms of what they’re likely to search for:

Demographic data can help, say Ingmar Weber and Carlos Castillo at Yahoo Research Barcelona, Spain. For example, they say that when US women type in the search term wagner, they are most likely to be thinking of the 19th-century German composer. US men, on the other hand, may well be thinking about the makers of spray painters.

By giving a search engine some basic demographic information, such as age, gender and educational background, it is possible to boost the engine’s chances of identifying user intent correctly, say Weber and Castillo. That personal information can be gleaned when people sign up to the other services, such as email, that search engines provide.

The wagner example is perfectly anti-Barry: I can absolutely guarantee that I would never be looking for spray painters, and would almost certainly, in a search for wagner, be interested in the composer. I know I’m not a typical guy in these regards, but, still... I know lots us U.S. men who would also never imagine searching for anything related to spray paint.

Generalizations are tricky things, and if one generalizes and the result is sometimes improving results, but never (or rarely) making things worse, we can cope. This sort of thing, though, strikes me as using too broad a brush (or, well, too broad a spray, I suppose) in painting a picture of a search. It seems likely to make things much worse, too often.

Happily, search engines such as Google have a lot more information, and, in particular, a lot more information specific to the individual. They know what you’ve searched for over time, they know what you’ve clicked on, they know, for those who also use their other services, what’s in your email, where you took your photos, and so on. When I search Google for wagner, I get, in order, the Wikipedia entry for the composer, the home page for Wagner College on Staten Island, and, in third place, the home page of the leader in consumer paint sprayers. (That’s followed by more stuff about Richard Wagner, and then the web site of Wagner Vineyards in New York’s Finger Lakes region.)

Personalization is more likely to help my search results than just trying to generalize on arbitrary demographic data.

The other experimental technique that I have serious questions about is mouse tracking:

However, the mouse serves analysts’ needs almost as well, say Qi Guo and Eugene Agichtein at Emory University in Atlanta, Georgia. Recent studies suggest a surprisingly high number of us — 70 per cent — use the mouse cursor to help our eyes follow text on screen. Tracking the cursor offers a cheap way to check how people read a search results page — vital information for companies keen to target their adverts at people who are likely to respond.

70% of us point the mouse at what we’re looking at? Really?

I’m skeptical.

I’m certainly different in that regard, as well. I tend to go the opposite way, moving the cursor out of my way so it doesn’t threaten to obscure what I’m reading. And with scroll wheels on mouse devices, I don’t even need to use the mouse cursor to grab the scroll bar. I’d have to do some data recording to be sure, but my guess is that for me, the mouse position has a negative correlation with where my interest is.

Yet, in their study — which I haven’t read; I’m only considering the news article — they found that [t]he volunteers’ mouse movements revealed their intent with 96 per cent accuracy. Wow! I should go check out their methodology.

Of course, nothing is going to be perfect. Even looking at what people actually click isn’t going to tell you for sure whether you found what they were looking for; accidental clicks and browses of interesting-looking search failures are common also. We’re just aiming to make an improvement, not to get it right all the time.

That’s where I think personalized learning will win every time.