The purpose of this method is to select K negative target words for a given context word, assuming these pairs are unlikely to co-occur (i.e., negative examples with a target value of 0). In this specific example, during training, K target words are paired with the context word 'orange'. Although there is a small chance that some of these randomly selected pairs might actually be valid (positive) pairs that occur within the context window (e.g. 'orange' and 'of'), this probability is very low using the sampling method (mentioned in video) in real-world corpora. Therefore, it can be safely assumed that nearly all the randomly sampled pairs are negative examples. Hope this explains your question.
Amazing Explanation
I thought it was unsupervised
why orange + of, target is 0?
it can happen, but for this example it in the text we have there is no 'orange of'
The purpose of this method is to select K negative target words for a given context word, assuming these pairs are unlikely to co-occur (i.e., negative examples with a target value of 0).
In this specific example, during training, K target words are paired with the context word 'orange'. Although there is a small chance that some of these randomly selected pairs might actually be valid (positive) pairs that occur within the context window (e.g. 'orange' and 'of'), this probability is very low using the sampling method (mentioned in video) in real-world corpora. Therefore, it can be safely assumed that nearly all the randomly sampled pairs are negative examples.
Hope this explains your question.