Hi everybody,
I was wondering for a long time, why search engines do not track such things. However, it is not so strait forward as it might seem, to use this kind of data.
I'd argue that in the first place, the algorithm is self feeding. Off course people click on the first entries the most (that is why everybody around this water cooler wants to be in the top 5 / 10 /20). But beyond that the interesting thing is the entries on page 2, 3, 4, ... that still attract attention from the users.
In my opinion the real value lies in timing of events. One click doesn't tell me much about the relevancy of the page. It just says that the short description did not repel the user. But if the same user comes back within a few seconds, then I can conclude he didn't really find what he was searching for and moves on.
I'd say the fastest return (back to the google search page) would stem from any kind of misrepresentation or from an ambiguous search term such as apple (as in apple computer or apple the fruit), where the result page just falls into the wrong category. However it could also stem from someone trying to return to some website she remembers and does remember the general appearance (colors, images, etc.) and theme, rather than the name.
Another pattern that I would be interested in, are people that click on a few entries and then revise their search (add or exclude terms). This would indicate to me, that the clicks they made didn't resemble the results expected. This also leads to the clustering technologies used by some advanced search engines (
dogpile,
Kartoo, etc.), each cluster being a revised sub search.
One remark about scripts (sh, pearl, etc.). They should not influence the results, because scripts usually do not interpret the JavaScript, like a browser does. Has anyone seen a script engine that does interpret JavaScript? As the search engines are blind for JS generated menus etc. so are scripting engines for this code. Anyone proofing me wrong?
I'm saddened, that they obviously do only support this kind of function for IE. This should skew the results towards the standard consumer users (and exclude all Linux, BSD users). I guess the community using alternative browsers is more filled with nerds, researchers or professionals, because they show a more sophisticated taste in browsers and look for certain features. But I might be wrong.
It must be certainly fun to mine all this data. For me to make sense out of it, I'd have lots of focus groups (and research) to get comments on, why people click in the pattern they click. Unfortunately this kind of thing is very expensive. However it would be certainly better than just guessing.
I'm very weary about algorithms, that just measure popularity. Because this is grossly unfair to the new comer site as well as it does severely limit diversity. It leads to the effect that you wake up in your hotel and have breakfast and can't remember what city you are in, because they all look alike. This kind of world is an awfully dull world in my opinion.
Thanks for bringing up this very interesting topic
K<o>