Mapping Hacks

I’m visiting the US for the first time since Google Maps came out, we’re staying in Brooklyn with a friend of Schuyler’s; one afternoon while he’s off visiting the The Open Planning Project, i decide to track down the local highly-recommended organic grocery. Where would i think to try first but Google’s spatial search? The results, i fear to report, were so irrelevant, that i fear the state of spatial search has become worse since i ranted about its missed potential for the book, late last year.

A straight name match for my search didn’t produce any hits, and i should have stopped there, shaking my head over the fact that chi-chi local businesses don’t pay to be listed in slow-moving national business directories. But i widened my search terms, thinking i might have misremembered the name or the zip code.

The image on the right shows the top 10 search results for “organic grocery 11211″

In the writeup in the book, i’d commented on the fact that Yahoo’s local spatial search seemed, from a distance at least, much more relevant - searches for “Doctor in San Francisco” didn’t turn up “Bike Doctor” and “Tofu Doctor” in the first 10 hits; though the results in the Overture advertising displayed on the same search seemed much more relevant - but what tofu doctor would buy an adword in sufficient volume?

We overlapped in San Francisco with hitherto when he came to interview with Yahoo!, and i recall a ranty conversation we’d had about spatial search; he was convinced that this domain would tend towards domination by the big aggregators. The service that can collect the most information in one place would, he insisted, be able to answer most peoples’ questions most of the time; that the utility of spatial search would be in massive collections - a-la Google, Yahoo!, Amazon - and not in detailed but scattered knowledge. I couldn’t really justify my disagreement at the time, but can offer a better justification now.

This seems to be a classic case in which structured metadata and free-text descriptions are mutually complimentary, and produce together much better inference and estimation than either alone. And the problem i see now - particularly with Google’s local search - is that they don’t really have either; they’re clearly not making use of the simple taxonomic potential in business classification codes which the directories use; nor do they have meaningful, user-contributable verbose descriptions about what kind of services businesses actually provide.

A city-local service like the Open Guide has both; long descriptions which, when searched over a collection of a few thousand documents, will usually produce relevant hits; and human-writable categories for places, which can be exposed as structured metadata, without obliging people to conform to a rigid classification scheme. This isn’t carping on principle; for me, the results are just experientially better

This critique isn’t specific to Google Maps / local.google; i find that none of the spatial searches have improved the state of the art in the last year. A common truism is that for mobile search services, getting this right is crucial; through the two-inch-wide pinhole view of the world, if the first ten search results are meaningless, a potential customer has wasted 50 cents and won’t return.

Having been too busy to look at the web for the last three months, a quick survey of the quantity and density of “google map hacks” that have been appearing, has amazed me. So many web developers have been able to pick up on the low-bar access to data, the simple URL-driven interfaces, and do interesting things with javascript and XSL - a definite lesson for the mapserver community and its diaspora. Maps.google has turned out to be a map application provider - (an MAP?) - rather than a map-enhanced search facility. This has been a fantastic thing in terms of bringing a lot of new people to web-facing geospatial tools. But it’s not doing much for the state of spatial search.

Posted in services |

You can follow any responses to this entry through the RSS 2.0 feed. You can skip to the end and leave a response. Pinging is currently not allowed.