Hi Pajolma,
> You're right, I am using the statistical version but the paper you're
> pointing ("Improving efficiency and accuracy in multilingual entity
> extraction") isn't giving me enough helpful information.
I understand. The other paper I linked to goes in a bit more detail
about what happens during the spotting, candidate selection, and
disambiguation phases. Neither paper is clear about to which endpoints
this corresponds. But this kind of background information could help to
formulate sensible hypotheses about time performance.
> My concern is that I'm not sure the 'candidates' endpoint and the
> 'disambiguate' one are totally separated between each other in the
> sense that I've the impression that a part of the disambiguation logic
> might be already performed during the candidates generation
> (contextual score calculation) which makes me doubt about the
> significance of just comparing the different endpoints time
> performance... Let me know if you see it differently.
Right, as far as I know there is indeed overlap between /candidates and
/disambiguate. My suggestion is that /spot + /disambiguate should be
equivalent to /annotate. If you are able to verify this, measuring /spot
will indicate how long the spotting phase takes, and /disambiguate how
long candidate selection, ranking, and selecting the top candidate
takes. To make any further distinction between candidate selection and
disambiguation, you will unfortunately need to dig deeper into the Java
code.
Cheers,
Alex
------------------------------------------------------------------------------
_______________________________________________
Dbp-spotlight-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dbp-spotlight-users
