Ideally, the probabilistic back-end would return the context-independent
scores
p(uri) * p(sf| uri) or
p(uri| sf ) or whatever variation of this as some sort of priorScore
(poorly named as support in the Lucene / VSM back-end) and the context
dependent component p(uri| tokens) should be returned in the
contextualScore attribute. Finally, we should add to the Lucene / VSM
another attribute called priorScore (with same value as support for now).
The combination of prior and contextual for the probabilistic backend can
be interpreted directly as some kind of confidence, so nothing needs to be
done about that.
If we rename the support, we will break a bunch of clients, so would need
to do it carefully (if at all). First duplicate the value with two
attributes priorScore and support, then send an email to the list, we will
make sure to add a version attribute to our xml/json, and warn the list a
second time, change our reference client implementations, and docs,
before making the switch.
On Jan 15, 2014 6:21 AM, "Joachim Daiber" <[email protected]> wrote:
> Hey ander,
>
> the statistical backend returns a "@similarityScore" value. It's a bit of
> a crude confidence value, as it is very close to 1.0 for most entities, but
> it should indicate entities the system is unsure about. Here is an example
> from the JSON response to the English web demo:
>
> 0: {@URI:http://dbpedia.org/resource/Berlin, @support:42602,…}
> @URI: "http://dbpedia.org/resource/Berlin";
> @offset: "38"
> @percentageOfSecondRank: "0.0013814913515745346"
> *@similarityScore: "0.998229618047692"*
> @support: "42602"
> @surfaceForm: "Berlin"
> @types:
> "Schema:Place,DBpedia:Place,DBpedia:PopulatedPlace,DBpedia:Settlement,Schema:City,DBpedia:City"
> 6: {@URI:http://dbpedia.org/resource/Municipalities_of_Germany,
> @support:7953, @types:,…}
> @URI: "http://dbpedia.org/resource/Municipalities_of_Germany";
> @offset: "234"
> @percentageOfSecondRank: "0.9880708432810636"
> *@similarityScore: "0.3563859722663317"*
> @support: "7953"
> @surfaceForm: "municipality"
> @types: ""
>
> Best,
> Jo
>
>
>
> On Tue, Jan 14, 2014 at 6:32 PM, [email protected] <
> [email protected]> wrote:
>
>> Hello,
>>
>> Is there any way to get some similarity score or confidence value for
>> the entities returned by spotlight statistical using curl sentences?
>>
>> For example Lucene version returns a similarity score for each entity,
>> but statistical does not.
>>
>> thanks
>>
>> ander,
>>
>>
>>
>> ------------------------------------------------------------------------------
>> CenturyLink Cloud: The Leader in Enterprise Cloud Services.
>> Learn Why More Businesses Are Choosing CenturyLink Cloud For
>> Critical Workloads, Development Environments & Everything In Between.
>> Get a Quote or Start a Free Trial Today.
>>
>> http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk
>> _______________________________________________
>> Dbp-spotlight-users mailing list
>> [email protected]
>> https://lists.sourceforge.net/lists/listinfo/dbp-spotlight-users
>>
>
>
>
> ------------------------------------------------------------------------------
> CenturyLink Cloud: The Leader in Enterprise Cloud Services.
> Learn Why More Businesses Are Choosing CenturyLink Cloud For
> Critical Workloads, Development Environments & Everything In Between.
> Get a Quote or Start a Free Trial Today.
>
> http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk
> _______________________________________________
> Dbp-spotlight-users mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/dbp-spotlight-users
>
>
------------------------------------------------------------------------------
CenturyLink Cloud: The Leader in Enterprise Cloud Services.
Learn Why More Businesses Are Choosing CenturyLink Cloud For
Critical Workloads, Development Environments & Everything In Between.
Get a Quote or Start a Free Trial Today.
http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk
_______________________________________________
Dbp-spotlight-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dbp-spotlight-users
