Classification time proportional to ?

[letronje]

I have observed that as we train graphify more and more, the size of the neo4j database on disk keeps increasing and beyond a point, each classification request takes more than a few minutes and makes it almost unusable.

Is there a way to train graphify for more accuracy but at the same time keep the classification time within usable limits ( like say 30 seconds or under a minute ? )

To understand the slowup, could you tell me which of the following parameters affect the classification time for a text given to it and how ?

• The number of labels/classes already known to graphify from previous training requests
• The total volume of text that has been given to graphify for training.
• The amount of text given to graphify for classification