Saturday, August 1, 2009

Nondeterministic Search Results in Google Maps

While testing the map functionality in Google Maps, one interesting finding has been that the searches returned are not deterministic. That is, if I set the default location to my home address, enter specific search terms (e.g. "coffee+shop"), and click Search, the results vary from search to search.

Go ahead, try it yourself. Go to Google Maps, set your default location, search for "coffee+shop", and then after searching once just hit the search button again. I don't know if Google Map's search algorithm is somehow adjusting the search criteria from inquiry to inquiry, maybe under the assumption that the first search didn't yield good results because you're searching a second time?

I don't know what's going on here, but I find it a bit strange. I was a bit worried about the case in my game where the player searches for points of interests, locates one at a particular address, then closes the app and goes to the address. Upon arriving, they open the app back up, repeat the search, and then find that location is no longer being produced. That would lead to very frustrated players, and would suck. I may prompt the player to repeat the search if this happens, because I don't particularly want to mess with persisting locations if I don't really have to.

Anybody got any ideas on why results change from search to search on Google Maps? Google's regular search engine doesn't work this way.

1 comment:

  1. Google's regular search actually DOES work that way in aggregate, but you never see it. You'd have to use two different computers in your house and honestly get very lucky, you can easily see different results. Most of it is due to experiments with the search algorithm. They'll run a 1% test so that 1% of visitors will see different results, but they'll be marked with an anonymous, but identifiable cookie so that they can give you consistent (experimental) results each time you load (until that experimental cookie expires). Then if the behaviors of those experimental users seem to show better search quality metrics, they'll increase the experiment to larger and larger amounts of traffic and if it's good they'll build it into the standard search.

    As for seeing that kind of behavior in Maps, I don't have a good answer. I think it's unlikely to be search experiments, but it could be that they are trying to find some way of showing you more options for results since the inherent ordering of PageRank is more difficult with physical buildings. :/