Machine learning algorithms usually expect numeric inputs. When a data scientist wants to use text to create a machine learning model, they must first find a way to represent their text as a vector of numbers. These vectors are called word embeddings. The Swivel algorithm is a frequency-based word embedding that uses a co-occurence matrix. The idea here is that words that have similar meanings tend to occur together in a text corpus. As a result, words that have similar meanings will have vector representations that are closer than those of unrelated words.
This model enables you to train the Swivel algorithm on a preprocessed Wikipedia text corpus. For instructions on generating word embeddings on your own text corpus see the instructions in the TensorFlow model repository.
|Domain||Application||Industry||Framework||Training Data||Input Data Format|
|Natural Language||Word Embeddings||General||TensorFlow||Any Text Corpus||Words|
- N. Shazeer, R. Doherty, C. Evans, C. Waterson, “Swivel: Improving Embeddings by Noticing What’s Missing”, arXiv preprint arXiv:1602.02215 (2016)
|Model GitHub Repository||Apache 2.0||LICENSE|
|Model Code (3rd party)||Apache 2.0||TensorFlow Models|
|Data||CC BY-SA 3.0||Wikipedia Text Dump|
Options available for training this model
- Train on IBM Cloud – Watson Machine Learning: follow the instructions in the GitHub README
Resources and Contributions
If you are interested in contributing to the Model Asset Exchange project or have any queries, please follow the instructions here.