Bio-Tag Recognition Method in Indonesian-Speaking Conversations Between Doctors and Patients

Abstract

This invention relates to a method for automatic medical information extraction from doctor-patient conversations on Indonesian-language telemedicine services. Previous Biomedical Named Entity Recognition (BioNER) methods are generally limited to identifying specific entities, such as anatomy (ANAT) and disease (DISO), so they are not able to represent clinical information comprehensively. This invention proposes a BioNER method with a coverage of eight types of medical entities, namely anatomy (ANAT), disease (DISO), chemical or drug (CHEM), duration (DRTN), frequency (FRKW), direction (DRCN), medical procedures (PROC), and patient age (AGE). The method starts from the stage of dataset collection, preprocessing, data annotation, to model formation using the XLM-RoBERTa + BiLSTM + CRF architecture. The use of BiLSTM and CRF has been proven to improve entity extraction performance compared to the baseline model. In the best scenario, the model produces an F1-score of 0.79 for ANAT, 0.74 for AGE, 0.70 for DISO, 0.65 for DRTN, and 0.64 for CHEM. Overall, the proposed method is able to improve the accuracy of medical information extraction from unstructured conversational text and has the potential to support faster and more accurate clinical decision-making in telemedicine systems.