A pragmatic introduction to building NLP models

Description

Many NLP tasks—such as sentiment analysis, named entity recognition, or machine translation—require each word in the input text to be mapped to a vector of real numbers. Words that are semantically similar correspond to vectors that are close together; that way, word embeddings capture the semantic relationships between words. Just like for other machine learning applications, you can either learn these embeddings from scratch or use transfer learning and fine-tune pretrained embeddings. The latter is usually preferable, especially if you use pretrained embeddings trained on huge datasets that generalize well.

Julien Simon dives into a quick introduction to word vectors, why they’re so important, and which are the most popular algorithms to compute them (Word2Vec, GloVe, BERT). He runs several demos showing you how to solve typical NLP problems by either computing embeddings or reusing pretrained ones using Jupyter notebooks based on Apache MXNet and Gluon NLP toolkit, as well as built-in NLP algorithms implemented in Amazon SageMaker (BlazingText, Object2Vec).