Combine NER and doc classification in annotation process


I am going to annotate approximately 10 000 documents, where I should select NER and labels of the document by several options. I understand, that I can split this task into two for NER and txt classification, but in this case, I need to read the document twice (it take some time to understand the content of this document ). So, the main question is it possible to annotate the document for NER and text classification at once (as heartext does it see example bellow ) ?

Best regards,

Hi! That's definitely possible using a combined interface with a ner_manual and a choice block. See here for an example: This one actually shows exactly what you're looking for (with an added free-form input in this case).

If your goal is to just label all individual examples as they come in, doing NER and text classification annotations jointly definitely make sense. If you want to use a more semi-automated approach and are looking to pre-select the best examples to annotate, it would probably make sense to separate the two tasks because the optimal examples you may want to select for text classification vs. NER are likely different. So you may want to train both components on different subsets of the data selected based on what works best for you model.