sense2vec

Why we included this project

Most word embeddings give every sense of a word the same vector, which muddles meaning. sense2vec fixes this by keying vectors to part-of-speech tags and entity labels, so "duck" as a noun and "duck" as a verb land in separate parts of the space. It builds on the word2vec approach from Trask et al. (2015) and ships as a Python library that loads pretrained models, runs fast nearest-neighbor queries, and trains new vectors from raw text using a spaCy pipeline plus GloVe or Word2Vec. Because it plugs in as a spaCy pipeline component and its vectors are fully serializable, you can bundle sense-aware embeddings directly into existing NLP workflows and model packages.