slinkp
commented
11 years ago Also happens with stuff that's in "Boston" but nothing more specific. We're getting those on the demo. The "center of boston" has moved to a spot in the water near william J day boulevard; this caused issue #215 ("news piling up on william j day blvd") and contributed to #192 (lots of stuff in "unknown" locations), since that spot on the water is in no zip code or neighborhood.
Example map demonstrating this: http://demo.openblockproject.org/maps/?c=-71.01741_42.33416&z=16&l=t3t9t1t2t7t4&s=09%2F01%2F2011&e=09%2F08%2F2011
When we geocode eg. a neighborhood name, we resolve that to a single point (the centroid), when actually all we really know about the news item is that it happened within the boundaries of the neighborhood.
This means that geocoded news can pile up on that (arbitrary) point and make it look like a lot of stuff is happening at a very specific location that just isn't true. At best this is misleading, at worst harmful (imagine your block being wrongly portrayed as a violent crime hotspot).
This is a variant of (but different from) #89: We don't want to exclude news for eg. "Roxbury" in Boston, we just don't want to pretend we have a more specific location than we do.
Desirable behavior: Those news items should be findable when browsing or searching by the relevant neighborhood, but not when browsing by block - and they should not be displayed as points on maps.
The browsing part could maybe be supported by having some sort of flag on NewsItems that marks these as "local but not really hyperlocal", and assigning NewsItem.location_object to eg. a neighborhood, and then fixing #93 (or, better, creating some NewsItemLocations since there could maybe be relevant locations of different types ... but there are scary triggers that will wipe those out if/when the Locations change.)
It's worth noting that I'm 99% sure this is exactly the use case for which NewsItem.location_object was created. Everyblock.com used them for news types eg. NY police reports where the available data only told you the relevant precinct, not a specific coordinate.
It's not obvious to me how the REST API should represent such items though. We're not going to want to duplicate a huge polygon for each newsitem's coordinates. Might have to grow a field in the API's output for that.