[Developers] Results from the reconciliation service?

Shawn Simister narphorium at gmail.com
Tue Feb 17 23:26:47 UTC 2009


Tom Morris wrote:
> Somehow I've managed to completely overlook the Reconciliation service
> until just now.
>
> http://www.freebase.com/view/guid/9202a8c04000641f8000000007beed56
> http://blog.freebase.com/2008/05/13/new-api-service-reconciliation/
>
> Is anyone using it?  Any results to report?  Tips to make it work better?
>   
I've experimented with that service in the past and ,like you, I found 
it lacking in several ways. Luckily, there's a new and much improved 
reconciliation service <http://mqlx.com/reconciliation/> in early 
development right now. It lets you add a lot more constraints which 
gives much better results than the old one.

For example, entering the following query into the new service should 
give you what you're looking for:

{
  "/type/object/name":"GM",
  "/type/object/type":"/business/company"
}

> Right now I'm using my own query against name (and aliases) and then
> fetching the associated Wikipedia article for some additional analysis
> to help disambiguate and punting if I'm not extremely confident of a
> match.  Anyone have any other successful strategies or tips that
> they're willing to share?
>   
I've generated some good results by filtering my reconciliation results 
against a hand-crafted black-list of incompatible types. So for example, 
if you're reconciling a list of companies, you know you can throw out 
any results that are typed as People, Locations, Buildings, etc. because 
they can't possibly be what you're looking for. Of course, you're never 
going to be able to come up with an exhaustive list of all the 
incompatible types but as long as you get the major ones you should see 
a noticeable improvement in the quality of your results.
> On a related note, has there been any discussion of generalizing and
> opening up the end user work queues (á là the voting queue or the
> Typewriter queues) to allow tasks to be contributed or is the thought
> that each domain/application will have to replicate this?
>
>   
I know that a couple of the Metaweb folks have mentioned this idea 
before. I haven't seen a formal feature request in JIRA but I know that 
the developers who built Typewriter have created another project called  
the Task Queue Library 
<http://acre.freebase.com/#app=/user/dfhuynh/task-queue-lib> which seems 
to be what you're describing. There are also two derivative projects 
called Genderizer <http://genderizer.dfhuynh.user.dev.freebaseapps.com/> 
and Geographer 
<http://geographer.stefanomazzocchi.user.dev.freebaseapps.com/> which 
seem to be built on the same framework as Typewriter.

Shawn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freebase.com/pipermail/developers/attachments/20090217/3b19c3d3/attachment.htm 


More information about the Developers mailing list