Hi Carin,
I've run some experiments with the Cortical.io classify endpoint and have found 
it to be a useful tool. The Python client allows you to query the endpoint [1] 
with positive examples that represent the class, and negative examples that do 
not; it may be useful to use the negative examples as those which may cause 
false positives. In return you get a fingerprint (bitmap) representing the 
class. Now you can classify text examples by first getting their fingerprints 
(via the term or text endpoints [2, 3]), and then calculating the distance to 
the category fingerprints; the API also offers an endpoint with many distance 
calculations [4]. So once you've created multiple category fingerprints you can 
classify text by determining which category is "closest", as defined by the 
distance metrics.

[1] 
https://github.com/cortical-io/python-client-sdk/blob/master/cortical/classifyApi.py
[2] 
https://github.com/cortical-io/python-client-sdk/blob/master/cortical/termsApi.py
[3] 
https://github.com/cortical-io/python-client-sdk/blob/master/cortical/textApi.py
[4] 
https://github.com/cortical-io/python-client-sdk/blob/master/cortical/compareApi.py

Cheers,
Alex

Reply via email to