DBPedia Spotlight pre-compiled models: where to find updated versions?

Services and Applications DBpedia Spotlight
1

Hello everyone!
I am using DBpedia Spotlight pre-compiled models for entity recognition (italian and english models). However, the latest pre-compiled models available, according to https://sourceforge.net/projects/dbpedia-spotlight/files/ are 3 years old (october 2016). Is it correct? I wonder whether there exist newer version of these models, since a lot of new entities can not be spotted because of this issue.

In case there are no recent pre-compiled models, is there a way to create my own models based on the latest version of DBpedia? I saw that the most recent DBpedia dataset was released on August 2019 (https://wiki.dbpedia.org/develop/datasets/dbpedia-dataset-2019-08-30-pre-release).

Any help is appreciated.

2

Hi,
in the past, it was possible to follow these steps https://github.com/dbpedia-spotlight/model-quickstarter but it is not up to date and does not use the DBpedia Databus for the moment.

According to the steps inside https://github.com/dbpedia-spotlight/model-quickstarter/blob/master/index_db.sh it will do a DBpedia RDF extraction on the newest Wikipedia dumps and these extracted DBpedia dumps are then used as input to create a new spotlight model.

The extraction is a quite resource expensive process, so it will consume a lot of memory and hard disk space and will take some time depending on your core count. ( >64GB memory, 300GB hard disk). As you mentioned there is an already new extraction provided. (official 2019-08-30, inofficial 2019.10.01 e.g https://databus.dbpedia.org/marvin/generic/)

It should be possible to adapt the index_db.sh script to use the latest already extracted DBpedia datasets from the Databus (e.g. https://databus.dbpedia.org/dbpedia/collections/pre-release-2019-08-30).

I created an issue, but you can have a look as well if you want.