My approach would be to somehow obtain a database of typical word frequencies in English text.

Then the most interesting keywords are those that appear significantly more often than in "normal" English text.

Of course you will need to experiment with cut-offs (for example considering only works that appear at least twice or so), filtering out names etc.