Prepare your training samples

If some of your categories are too similar, they are not good training samples. For instance, “laptop” and “desktop” are too similar to be distinguished from each other. However, in most customer systems, they are normally treated as two different categories and most likely linked to the same supporting group. Therefore, training the system with this kind of data will lead to low accuracy test results, but has no impacts on the ticket assignment.

For such situations, we recommend you consider combining these similar categories before starting training.