How to derive suggestions for categories from data?

With freely available pretrained models like GloVe, classification has become quite easy.

But where to start if I do not know which categories are there?

In the example linked above we "suggest" categories to the classification algorithm:

# Category -> words
data = {
'Names': ['john','jay','dan','nathan','bob'],
'Colors': ['yellow', 'red','green'],
'Places': ['tokyo','bejing','washington','mumbai'],
}


Please give me a hint whether and how it could be possible to derive suggestions for possible categories from a larger data set. Output could be for example:

# Category suggestions
categories_suggestions = {
'cat1': ['fish','bird', 'snake'],
'cat2': ['wood', 'metal','plastic'],
..
}


So that a human could look upon it and say, ah ok, let's name cat1 "animals" (even if "pets" could appear as more appropriate afterwards) and cat2 "materials".

• Please provide more detail: so will be all documenrs' words mapped to clusters? – J. Doe May 12 '19 at 11:09