Continuous Bag-of-Words (CBOW)

This article provides a comprehensive analysis of the Continuous Bag-of-Words (CBOW) model, a neural network architecture designed for learning dense word embeddings. The model operates by predicting a central target word from its surrounding context words. Its architecture involves representing context words as one-hot vectors, projecting them into a continuous vector space via an input weight matrix, and, crucially, averaging these context vectors to form a single aggregated representation. This averaging process, which gives the model its "bag-of-words" name, creates a smoothed summary of the semantic environment while disregarding word order. The resulting hidden vector is then passed through an output weight matrix and a softmax activation function to produce a probability distribution over the entire vocabulary. The model is trained using a self-supervised approach, where training data is automatically generated from a raw text corpus. Through iterative cycles of forward propagation, loss calculation (typically cross-entropy), and backpropagation, the model adjusts its weight matrices to improve its predictive accuracy. The primary output of this training process is not the predictive model itself but the learned input weight matrix, which serves as a powerful word-embedding lookup table. These embeddings place words used in similar contexts close to each other in a geometric space, thereby capturing semantic relationships. While the model is computationally efficient and effective for frequent words, its averaging mechanism can dilute the influence of rare words, and its static nature produces a single vector per word, failing to account for polysemy. Ultimately, CBOW is presented as a foundational paradigm that marked a critical shift from frequency-based methods to context-driven learning, establishing principles that influence modern natural language processing.