In this chapter, we will discuss the motivation for understanding semantic relationships between words, and we will discuss approaches for identifying such relationships. In the process, we will obtain a vector representation for words, which will let us build vector representations at a document level.
We will cover the following topics in this chapter:
- Word embeddings, to represent words as vectors, trained by a simple shallow neural network
- Continuous Bag of Words (CBOW) embeddings, to predict a target from a source word, using a similar neural network
- Sentence embeddings, through averaging Word2vec
- Document embeddings, through averaging across the document