It's neither, I think. It's basically a way to get training data. The novel part is splitting sentences into phrases and asking about the sentiment of each part

It's neither, I think. It's basically a way to get training data. The novel part is splitting sentences into phrases and asking about the sentiment of each part, which provides the algorithm richer data to train off.