Gregory's ServerKai Labusch from @stabi_berlin is presenting at #swib23 a very sophisticated service for named entity recognition and entity linking in digitized texts, constructed with machine-learning based on #Wikidata entities and their mentions in #Wikipedia. I wonder if one could/should use the #Reconciliation Service API protocol for use cases like this one. I guess the protocol would have to allow to send the context sentence along with the NERed string we want to reconcile. /cc @pintoch
Adrian
7 comments
cneud
@acka47 @stabi_berlin @pintoch That‘s it, we need fairly long context for the disambiguation, which is where the Wikipedia articles are really handy - also the reason why we chose Wikidata over GND as linking target. More on this here https://doi.org/10.1515/9783110691597-012
Adrian
@cneud @stabi_berlin @pintoch I was thinking about e.g. TEI publisher that also supports choosing standard-conformant reconcilation services for – manual – entity linking. E.g. you configre GND as linking target, mark a string and matching candidates from GND will be shown. Might be a big improvement for these kind of use cases to pimp lobid-gnd reconciliation based on what you have developed.
cneud
@acka47 @stabi_berlin @pintoch Would be curious to look into that! Maybe also a potential SWIB24 submission idea?
Adrian
@cneud @stabi_berlin @pintoch Our team will soon discuss possible future projects (and funding options). I will add this to the list and contact you.
Sebastian Majstorovic
@acka47 @stabi_berlin @pintoch Very interesting! Is there a code repo for the project available somewhere? |
@stabi_berlin @pintoch Generally, the protocol already supports it as you can send the context along as additional property. (The service could define the property name to send the context in.)