In the described classifier setup, which data are used as positive samples?

In the described classifier setup, which data are used as positive samples?

• A. (t,c) pairs drawn from negative data.
• B. (t,c) pairs drawn from positive data.
• C. (t,c) pairs drawn from random contexts.
• D. (t,c) pairs drawn from any context.

Answer: (B)

The main idea is that positive samples come from data labeled as positive. In this setup, (t,c) pairs drawn from positive data show what a true or desirable association looks like, so the classifier learns to recognize those patterns and assign high scores to them. If you used data from negative samples, the model would learn to favor the opposite of what positives represent, which undermines its ability to identify true positives. Likewise, drawing (t,c) pairs from random contexts or from any context would blur the distinction between positive and negative instances, adding noise rather than useful guidance. So the data drawn from positive data are used as positive samples.

The main idea is that positive samples come from data labeled as positive. In this setup, (t,c) pairs drawn from positive data show what a true or desirable association looks like, so the classifier learns to recognize those patterns and assign high scores to them. If you used data from negative samples, the model would learn to favor the opposite of what positives represent, which undermines its ability to identify true positives. Likewise, drawing (t,c) pairs from random contexts or from any context would blur the distinction between positive and negative instances, adding noise rather than useful guidance. So the data drawn from positive data are used as positive samples.