Comments on: Building High Precision Classifiers, Taggers, Chunkers, Spelling Correctors, …

By: Text Classification for Sentiment Analysis – Precision and Recall «streamhacker.com

Mon, 17 May 2010 14:46:16 +0000

[...] F-measure to be about as useful as accuracy. Or in other words, compared to precision & recall, F-measure is mostly useless, as you'll see below.Measuring Precision and Recall of a Naive Bayes ClassifierThe NLTK metrics [...]

By: Jacob

Jacob — Tue, 27 Apr 2010 15:21:32 +0000

Here’s a suggestion for naming that non Active Learning process: Manualated Learning. I’m doing something like that right now:
1) Identify uniquely specific keywords for classification
2) Classify documents based on the frequency of those keywords
3) Use those classified documents as training data for supervised classification

We’re currently in the “manualation loop” of 1 & 2, but we’ll soon be feeding 2 into 3 to see how the results are.

By: Breck Baldwin

Breck Baldwin — Mon, 26 Apr 2010 17:49:43 +0000

Customers will almost always ask for 90% or more for whatever metric we hand them, I think it is because in the American school system 90% or above is an A. When they actually see data they tend to be happy with lower performance.