Biggest Size of your vocabulary by Nodes

j0ej0e

Hi,

Whats the biggest size vocabulary in your project in terms of the number of nodes?
Mine is 427, anyone has similar or bigger?

What are your experiences with a Classifier of big number of nodes?
Based on the documentation, the expected accuracy and completeness is on an average of 0.3 to 0.5 for Classifiers, anyone manage to get higher accuracy and completeness?

rgds,
joe

Migrateduser

Hey Joe,

Accuracy can go up substantially if you switch to a single-label classifier or use classification in combination with recognition and a rules engine to add obvious codes and prune out obvious errors. You will also want to use larger k values (~ 30) with a fairly large number of well-vetted training docs to get the accuracy up.

As far as number of nodes, managing a training set for more than 1000 categories will be difficult due to the "crowding-out" effect, where two categories can ligitimately apply, but one category's examples crowd-out all the examples of the second category.

Rgs
Clark