I have been playing with lda2vec, Gensim and other great tools but so far have not been able to satisfy this story
As a user wanting to know the subjects in a PDF I found, I want words examined for inter- and intra-document proximity and I want them to be located in a taxonomy I am familiar with (a social science taxonomy). Hopefully there are hyperparameters I can tune so the system performs as expected in my field. As an example, a PDF that talks about initmate partner violence, elder abuse and child welfare should be among terms identified as high salience, then matched into part of the (huge) taxonomy that looks possibly like this
Domestic violence Family violence Initimate partner violence Victims of family violence Elder abuse Violence against women
And the (highest common) hypernym Domestic violence is output as a subject. Note that the PDF does not have the words Domestic violence in it at all. So far in the above tools I get word clusters, but it doesn't have supervised training to know hypernyms.
Thanks so much in advance Craig