Wikidata:WikidataCon 2017/Notes/Wembedder: Wikidata entity embedding web service

Abstract

embedding (Q29043227) is a topic that has gained a considerable attention in the machine learning community in recent years. Wikidata as a corpus is often used to train the embedding models that can be used in a variety of applications. I have been experimenting with using Wikidata as the corpus for training the graph embedding (Q32081746) models and implemented a web service running with a JSON-based API on the Wikimedia Toolforge https://tools.wmflabs.org/wembedder/. The current version returns "most similar" Wikidata items based on a query item. Albeit the performance is not optimal, there seems to be great opportunities with embedding models with future Wikidata-for-Wiktionary and the combination of Wikidata and Wikipedia in a combined model.