How was the exampleWordEmbedding example in the text analytics toolbox trained, in detail?
3 views (last 30 days)
Show older comments
William Smith on 19 Nov 2017
Answered: Christopher Creutzig on 9 Mar 2020
The documentation for readWordEmbedding gives a pre-trained embedding, saying only that it was "derived by analyzing text from Wikipedia".
How was it trained?
Should we consider it a 'high quality' word embedding, better than anything a user could generate without extensive work and CPU time? Or is it a quick and dirty starting point, and we are encouraged to train our own for better performance?
Christopher Creutzig on 9 Mar 2020
The embedding is rather low-dimensional (50 dimensions) and has a small vocabulary (with 9999 words). It is unlikely to be “high quality” unless your analysis just happens to need precisely this dataset.
For production use, it is much more likely you'll find fastTextWordEmbedding useful, which downloads data from https://www.mathworks.com/matlabcentral/fileexchange/66229-text-analytics-toolbox-model-for-fasttext-english-16-billion-token-word-embedding for you.
Find more on Sequence and Numeric Feature Data Workflows in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!Start Hunting!