The word2vec model, developed by Mikolov et al., provides state-of-the-art word embeddings. This paper explains the negative-sampling method used in the model. The model is based on the skip-gram model, which aims to maximize the probability of word-context pairs in a corpus. The conditional probability p(c|w) is modeled using a softmax function, but this is computationally expensive due to the large number of contexts. To address this, Mikolov et al. propose negative sampling, which replaces the softmax with a binary logistic function. The objective function is then modified to include both positive (training data) and negative (randomly sampled) examples. The negative sampling approach optimizes the log-likelihood of the positive examples and the log-probability of the negative examples. The negative sampling method is more efficient than the softmax approach, as it reduces the computational complexity. The negative sampling method is based on the idea that words with similar meanings should have similar vector representations. The model uses a dynamic window size and subsampling to improve the quality of the embeddings. The effectiveness of negative sampling is attributed to the increased effective window size, which includes more context words. The paper concludes that the negative sampling method produces good word representations, although the exact reason is not fully understood.The word2vec model, developed by Mikolov et al., provides state-of-the-art word embeddings. This paper explains the negative-sampling method used in the model. The model is based on the skip-gram model, which aims to maximize the probability of word-context pairs in a corpus. The conditional probability p(c|w) is modeled using a softmax function, but this is computationally expensive due to the large number of contexts. To address this, Mikolov et al. propose negative sampling, which replaces the softmax with a binary logistic function. The objective function is then modified to include both positive (training data) and negative (randomly sampled) examples. The negative sampling approach optimizes the log-likelihood of the positive examples and the log-probability of the negative examples. The negative sampling method is more efficient than the softmax approach, as it reduces the computational complexity. The negative sampling method is based on the idea that words with similar meanings should have similar vector representations. The model uses a dynamic window size and subsampling to improve the quality of the embeddings. The effectiveness of negative sampling is attributed to the increased effective window size, which includes more context words. The paper concludes that the negative sampling method produces good word representations, although the exact reason is not fully understood.