wtf is dutchembeddings?

Dutchembeddings gives you pre-built "word embeddings" for the Dutch language, essentially a mathematical map of how Dutch words relate to each other. Instead of you having to train a model on millions of Dutch texts yourself, this project has already done that work and packaged the results as downloadable files. Word embeddings represent each word as a list of numbers (a vector) so that similar words end up with similar numbers. This lets software understand that "koning" (king) and "koningin" (queen) are related, or that "hond" (dog) and "kat" (cat) are closer to each other than either is to "computer." The project built these from several Dutch text collections, including Dutch Wikipedia, news articles from the Roularta corpus, and a large web-based corpus called COW. They offer two sizes: 160-dimensional and 320-dimensional vectors, and recommend the larger ones for better quality. This is aimed at anyone building Dutch-language NLP applications, a startup making a Dutch chatbot, a researcher studying Dutch semantics, or a company doing sentiment analysis on Dutch customer reviews. If you're building something that needs to understand Dutch text and you don't have the time or data to train your own word vectors from scratch, this gives you a ready-to-use resource. The files are in a standard word2vec text format, so they load easily into popular Python libraries like gensim. The project also includes a small evaluation tool called relation.py that lets you test how well a given embedding model captures word relationships, similar to the famous "king - man + woman = queen" test. The embeddings are released under a Creative Commons license and the code under GPL, so they're free to use with attribution.