As others have mentioned, trying to automate this task would be close to impossible. But if you want to get some very rough hints of words for yourself, this is something I would try (note, requires some scripting):
Use detex or something to strip the TeX markup and then write a small script that counts the number of time each word has been used in the document. The top words in the list will probably be useless words like a, the, is, etc. But, after those, you might be able to find a few promising words.