Yes, you have to. :/ I tried to explore ways to grab the data without needing to click it, but I needed a well defined action (going to the next page is valid to me) that would tell the classifier to train on the data. I'll see if I can find multiple ways to direct certain actions to train the classifier.

Without seeing the source it's hard to ask relevant technical questions, but why not train every time a link is opened?

Online training? I need to read more about it, but I believe there is a partial implementation of it in it weirdly enough.

Dev Source: https://github.com/normano/ycChrome

Naive Bayes is an online classifier anyway, training on click would be the equivalent of clicking a single link and then clicking next. I don't see why it would be hard to do...