What is unsupervised learning?
The applications that we worked with in earlier chapters were based on data that was manually categorized by human annotators. For example, each review in the movie review corpus that we have used several times was read by a human annotator and assigned a category, positive or negative, based on the human’s opinion. The review-category pairs were then used to train models, using the machine learning algorithms that we previously learned about to categorize new reviews. This whole process is called supervised learning because the training process is, in effect, supervised by the training data. The training data labeled by humans is referred to as the gold standard or ground truth.
Supervised approaches have some disadvantages, however. The most obvious disadvantage is the cost of developing the ground-truth data because of the cost of human annotators. Another consideration is the possibility that the manual annotations from different annotators...