Topics
In CBOW (Continuous Bag of Words), we condition the probability of center word based on contedx words. The nomenclature can be broken down as :
- “Continuous” - refers to the dense vector representations (word embeddings) that are learned, as opposed to discrete/sparse one-hot vectors
- “Bag of Words” - because it treats the context words as an unordered set/bag, losing the positional information of words around the target