How to Evaluate a Natural Language Processing Model

Our model is a full probabilistic model. It considers all possible jobs in the universe and that size of that universe depends on the taxonomy that's used in code jobs. And it signs to each of those jobs a number. So someone who has worked as a waiter, for example, may have, I don't know, 70% chance of being a waiter that same the next year. But then maybe a 20% chances of being a job similar to a waiter, something like a bartender.