Opened 13 years ago

Last modified 7 years ago

#1203 new enhancement

More configurable rating system

Reported by: robe Owned by: robe
Priority: medium Milestone: PostGIS Fund Me
Component: tiger geocoder Version: master
Keywords: Cc: woodbri

Description

That will allow you to prioritize match weighting of different parts of the address.

E.g you could pass in a rating object metric that allows you to penalize/favor

Some of this was discussed in #1109

e.g.

r.address_number = 10 r.street = 20 r.zip = 20 r.location = 40

etc.

and the rating system would give a penalty of between 0 and valspecified for each that is not an absolute match and then normalize so max bad rating is 100.

Change History (4)

comment:1 by robe, 12 years ago

Milestone: PostGIS FuturePostGIS 2.1.0
Version: 1.5.Xtrunk

now that I have geocoder_settings table, will try to incorporate these in. It will be global but oh well. I think most you probably want it global anyway.

comment:2 by woodbri, 11 years ago

Cc: woodbri added

A better scoring heuristic would be to compute a difference score for each standardized component using something like similarity() or levensthein() then assign a weighting to each component and sum and normalize the result.

You can also do things like check if pre directional and post directional are swapped and not penalize as much for that.

This comparison can be done between the standardized reference record and the standardized request easily because they are both broken into the identical field structures. Making the comparison between the unstandardized reference record and the unstandardized request is much harder and more erroroneous because things like ST. vs SAINT will throw off the scoring.

comment:3 by robe, 11 years ago

Milestone: PostGIS 2.1.0PostGIS Future

comment:4 by robe, 7 years ago

Milestone: PostGIS FuturePostGIS Fund Me

Milestone renamed

Note: See TracTickets for help on using tickets.