This package contains code for the Entity Recognition and Disambiguation Challenge. It took me some time to configure all the stuff and I think it could be useful for other teams participating in the challenge.
The package contains two utilities:
- code to index the freebase <-> wikipedia file provided by the organizers, in order to convert the wikipedia labels to freebase id (and to filter out invalid entities);
- code to start the REST service (for now, just the short track).
Once you have downloaded the entity.tsv file provided in the Datasets page, you can index it running
./script/index.sh entity.tsv mapdb
This command will create a folder mapdb containing an index with all the mappings. At runtime you can access
the index creating a WikipediaToFreebase object:
WikipediaToFreebase wikiToFreebase = new WikipediaToFreebase("mapdb"); // the folder name
And then use:
wikiToFreebase.getLabel("Diego_Maradona");
in order to retrieve the freebase-id for the entity Diego_Maradona. Please observe that wikipedia labels are case sensitive and that I split
out the common prefix /wikipedia/en_title/ to make things more compact.
I set up a REST service for the short track as required in the challenge. I put both a POST and a GET service, the GET service is useful to test if everything works. In order to use it you only have to patch the Annotator object and make sure that you return your list of Annotations. When you are ready, you only have to run the command
mvn jetty:run
and the rest service will answer at the address:
http://$(your-ip-address):8080/erd-challenge/rest/shortTrack
I hope this will help ;)
Diego