Skip to content

fnielsen/wembedder

Wembedder

Embedding with Wikidata.

There is a web application and web service available:

python app.py

When the server is running and the model is loaded the webservice should appear at http://127.0.0.1:5000/

Note you will need to download model(s) from Zenodo for the webservice to start.

Pre-trained models

Wembedder currently works with specially trained Gensim models. Pretrained models are published on Zenodo:

Wembedder wikidata-20170613-truthy-BETA-cbow-size=100-window=1-min_count=20
Wikidata embedding Gensim CBOW trained on truthy dump with a 100-dimensional subspace, a window of 1 and a minimum count of 20. https://zenodo.org/record/823195
Wembedder wikidata-20170613-truthy-BETA-cbow-size=100-window=1-min_count=20-iter=25
Wikidata embedding Gensim CBOW trained on truthy dump with a 100-dimensional subspace, a window of 1 and a minimum count of 20. Trained with 25 iterations. https://zenodo.org/record/827339

The models need to be unpacked under ~/wembedder_data/models/.

Wikimedia Toolforge

The canonical webservice runs from Wikimedia Toolforge at https://wembedder.toolforge.org/

Reference

Finn Årup Nielsen, Wembedder: Wikidata entity embedding web service, arXiv:1710.04099, http://www2.imm.dtu.dk/pubdb/views/edoc_download.php/7011/pdf/imm7011.pdf