Supervised Classifier
Is a classifier that is trained through supervised learning. Eg. bayes-classifier
We generally supply the classifier with [see page 10, 3 sets] of data:
- Training -> estimate probability distribution of discrete classifications.
- Development -> Used to fine tune classifier parameters to select best model.
- Test -> Evalaute the classifier (make sure it doesn't reproduce training result)
Training data should be:
Aspect | Meaning |
---|---|
[see page 28, representative] | Even if the result is wrong, it should be derivable from the training data. |
[see page 33, large] | To avoid missing words in our vocabulary leading to inaccurate classifications. |