Omin should find frequently used phrases in the prompt based on the model, and use them to find related words (frequently used together). Whether Omin makes use of those phrases shouldn't be based on chance but only by how good the match is.

Omin should also build a list of how much he likes certain phrases that were matched and save those preferences for the future — so if he speaks positively of one, he'd be unlikely to speak negatively of them later.