How can I programmatically extract topics from text?

[Kristy Atkins]

I've been working with LDA for about 6 weeks full time and have read a lot of of tutorials and seen youtube videos on it.

I have a working Python app that uses Gensim. I'm not getting anywhere near the results I hoped for.

I suspect that I'm not using: gensim.models.ldamodel.LdaModel correctly.

How can I programmatically determine what settings would produce the best results (most accurate word groupings)?

How can I get somewhat reliable groupings out of the text?

How can I programmatically determine each actual 'topic'?

I'm aware of Wordnet Synset hypernyms and thought if I can get the word groupings per topic more accurate, maybe I can get common hypernyms from them and use those.

I'm stumped. I don't know how to extract valid topics or even word groupings from LDA.

Here is my data, results and code: