We have updated Wikipedia2Vec, our open-source tool for obtaining high-quality embeddings (vector representations) of words and entities from Wikipedia. In addition to the source code, we provide pretrained embeddings for 12 languages and offer benchmark comparisons with other competitive models.

In this release, major improvements have been made to the Wikitext parser, especially for non-English Wikipedia editions. Also, wheel packages are now available, aimed at significantly reducing installation time.

Wikipedia2Vec website