Hi,
I trained two different models that are designed to read websites for information. One of them recognizes names, and the other recognizes job titles. I made them seperate, but I would like to have them in one model so I can compare the positioning of the results and correlate specific names to job titles. Is there any way I could combine these models into one or possibly find a way to run both at the same time so they show me the order in which I found each of the results?
Thanks!
Hi! The most straightforward approach would be to train one model on the combined datasets so you have one entity recognizer that predicts both. Prodigy's train recipe lets you pass in multiple comma-separated dataset names and will merge the annotations before training. An advantage of this approach is that you can potentially achieve better results, because the presence/absence of one label will have an impact on the other label. For example, a token can only be part of one entity, so if you know that something is a job title, the model also knows that it can't be a person.
Thank you!