Learn from the Sentiment140 classifier

[bwbaugh]

We could use the Sentiment140 API for classifying tweets during the training stage. We could learn from those as if they were annotated data in order to update our model.

This is different than self-learning because our current model isn't used during the training process. Instead, a different model (Sentiment140) is used for classifying additional unlabeled training instances. Our model would still solely be used for classifying test data.

I'm not sure if this is related (or if so, how much) to classifier compression using active learning.

If nothing else, it at least be as a comparison against other methods (especially over the Sentiment140 baseline itself).

Idea was originally recorded 3/12/13 at 23:01.