Figure 1.

Illustration of the natural language processing pipeline. The full-colour boxes indicate main parts of the data or model development: (1) definition of the variables, (2) development of annotation schema, (3) annotation of the documents, (4) development of initial model, (5) fine-tuning of the model using active learning procedures and (6) deployment of the model on all clinical notes. The outlined boxed indicate processing of information and calculations between main steps: (1) calculation of the quality of notes and using ones with maximum quality, (2) changing schema through exercise with clinicians, (3) splitting the gold corpus on development and validation part and (4) validation of performance for developed and fine-tuned model.