I am trying to use patent texts to make a binary classification of a large number of patents as either 1. related to automation or 2. not related to automation.
I have manually classified a small non-random subset of patents and would like to use them to train a classifier. I formed this subset by looking at all firms that had patented an automation technology in a small window of time. Non-automation patents in the subset are thus going to be quite different than non-automation patents in the whole data because they are filed by firms that do some work in automation.
My question has two parts:
Is there a classification method that can be used with "biased" training data? (My hunch is no but never hurts to ask.)
Alternatively, is there a classification method that will simply identify which patent texts are most like the ones I classified as automation patents?