January 2018
Beginner to intermediate
284 pages
8h 35m
English
On top of the novelty of the new algorithms, such as Skip-Gram models (with negative sampling), CBOW (with hierarchical softmax), NCE, and GloVe, compared to traditional count-based approaches, there are also many new hyperparameters or preprocessing steps that can be tuned to improve performance. For example, subsampling, removing rare words, using dynamic context windows, using context distribution smoothing, adding context vectors, and more. Each one of them, if used properly, would greatly help boost performance, especially in practical settings.
Read now
Unlock full access