2

We all know precision and recall. What if two algorithms have the same precision and recall but one algorithm makes more predictions with the data available.

Example: "I love samsung. Apple is terrible."

Algorithm 1: "I love samsung. Apple is terrible." Makes prediction Neutral.

Algorithm 2: "I love samsung. Apple is terrible." Makes prediction Positive for Samsung and Negative for Apple. Two predictions.

So far both have 100% Precision and 100% Recall. However, one is clearly making more predictions, segmenting more, and giving more information than another.

iuppiter
  • 408
  • 3
  • 13
  • Although one algorithm is "giving more information" than another, it has been trained with different data and with different target. Whereas one seems to estimate the sentiment of a saying such love+terrible = neutral, the other estimates two (or more) target variables. You are thereby comparing apples and pears. Or I did not understand the question - in that case: please clarify. – Nikolas Rieble Feb 21 '17 at 20:19
  • You did understand it correctly. It is apples to oranges in a way, but one model is clearly saying more with the same data with the same level of precision/recall. I just wanted to know if there is a technical way to vocalize that. – iuppiter Feb 21 '17 at 20:21

0 Answers0