What are the best practices for classifying uncoded data?


Is there is a certain percentage amount that you recommend that I manually code by hand in my sample before using the uClassify system? Should I do 10%, or would a larger percentage like 40% give better results?

We advise iteration until you do not see an improvement that justifies further coding efforts. 

Once you have rebuilt the classifier 2 or 3 times, use the classifier on a set of uncoded data. Then filter by a single code at the 85%-95% level. Sample that list and code it.

If you get 90% accuracy, then you can decide if more accuracy is required.


Have more questions? Submit a request


Powered by Zendesk