Cracking Word Embeddings: Why One-Hot Fails and How Word2Vec Actually Works

Automated excerpt

In this model, every word gets two separate vector representations: one used when it acts as a center word, and another used when it acts as a context word. Instead of predicting context from a center word, CBOW assumes that a center word is generated based on its surrounding context words. While training follows a similar gradient optimization process to skip-gram, CBOW typically uses the context word vectors as the final word representations.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.