I was reading several Skipgram Negative Sampling codes such as this one in Github and became curious how to exclude positive samples among sampled negative ones as the code seems not properly doing it. As I understood, the original paper says that the negative sampling is sampling a few unrelated words and minimizing their likelihood. Don’t we have to strictly exclude positive samples being picked from negative sampling? I guess we have to further check whether samples from negative sampling includes one from positive samples.
Please correct me if I’m wrong.