Previous chapters covered text preprocessing and feature representation methods based on word frequencies, such as Bag-of-Words and TF-IDF. These techniques are effective for certain tasks but struggle to capture the semantic meaning or context of words. For example, they treat 'cat' and 'feline' as entirely unrelated if they don't frequently appear in identical contexts within the training data.

This chapter shifts focus to methods that represent words based on their surrounding context, leading to representations that encode semantic similarity. We will start by discussing the limitations of frequency-based approaches and introduce the concept of distributional semantics, the idea that words appearing in similar contexts tend to have similar meanings.

Following this, we will examine word embeddings: dense vector representations, such as $w \in \mathbb{R}^n$, learned for words. We will look at popular algorithms like Word2Vec (including its CBOW and Skip-gram variations) and GloVe (Global Vectors for Word Representation). Finally, we'll cover techniques for visualizing these embeddings and learn how to utilize readily available pre-trained embedding models for integration into other NLP tasks.