I am trying to build a classifier that takes as input a tuple of texts:
and learn whether there is a match. For example, for tuples:
[(‘outdoor shoes’, ‘merrel’), (‘outdoor shoes’, ‘salomon’)]
would be positive examples (the target, y, is 1), whereas these:
[(‘outdoor shoes’, ‘spatula’), (‘outdoor shoes’, ‘spoon’)]
would be negative examples.
I am struggling to figure out how to frame this modeling problem. It is not straight text classification, since there is no single text blob as my feature. It is more akin a question answering problem in which each (question, answer) tuple is classified as being correct or not.
What kind of fast ai learner/model do you recommend using in this case?