It’s not pure chance that the above calculus shakes out, but it doesn’t have to be that way. If you are embedding on a word by word level then it can happen, if

It’s not pure chance that the above calculus shakes out, but it doesn’t have to be that way. If you are embedding on a word by word level then it can happen, if it’s a little smaller or larger than word by word it’s not immediately clear what the calculation is doing. But the main difference here is you get 1 embedding for the document in question, not an embedding per word like word2vec. So it’s something more like “document about OS/2 warp” - “wiki page for ibm” + “wiki page for Microsoft” = “document on windows 3.1”