wikipedia2vec

Why we included this project

Wikipedia2Vec learns word and entity embeddings in the same vector space, so a name like "Apple" gets its own representation tied to the Wikipedia article rather than being treated as an ordinary token. That co-embedding makes entity linking, related-entity retrieval, and knowledge-aware classification noticeably easier to build. You can train it from a plain Wikipedia dump with a single command, no annotations or schema setup needed, and pretrained embeddings for twelve languages are available if you'd rather skip that step. The bundled NABoE example, an entity-aware text classifier built on a neural attention model, shows a practical starting point whether you're exploring the technique or wiring it into a pipeline.