• GloVe = Global Vectors for Word Representation
• Proposed in this paper

How Glove Works

• Co-occurrence Matrix: A co-occurrence matrix is produced where the matrix dimension is $(vocabulary\mathrm{\_}size\ast vocabulary\mathrm{\_}size)$ and each of the element is the number of times both of the words was in a single sentence in the corpus
• Matrix Factorization: Matrix Factorization algorithm is applied on the matrix to reduce the dimension of the word vector and to produce a lower dimensional dense vector.

Pros:

1. Unlike Word2Vec Embedding, GloVe can capture global relationships
2. Like Word2Vec Embedding, it can capture the semantic relationship
3. It produces richer word representation than the Word2Vec Embedding

References

1. https://aman.ai/primers/ai/word-vectors/#count-based-techniques-tf-idf-and-bm25