Yes but will require extra processing on the training data and code adjustment of word embedding tool. In our case the word2vec output produce binary model where each word correspond with a numerical vector with no other attributes. Also, most often then not the trained dataset itself is not categorized by article or author.