Deep Learning-Based NER for Digitizing Unani Medicines Text

Abstract

Traditional medical systems such as Unani Medicine (UM) encompass centuries of therapeutic wisdom derived from Greco-Arabic traditions and enhanced by South Asian influences. Nevertheless, much of the Unani literature is still in non-digitized forms and various classical languages, making it difficult to integrate with contemporary healthcare technologies. This research introduces a named entity recognition (ner) machine learning preprocessing tool method to digitize and extract structured data from Unani medicinal texts systematically. Utilizing the reference work Classification among Unani Drugs with English and Scientific Names”  by Ahmed and Nizami, a specialized dataset was created through OCR-based text extraction, followed by thorough preprocessing and manual annotation. We developed and assessed a BiLSTM-CRF model employing furtherness is pre-trained GloVe embeddings to discern and group four main entity types: Unani drug names, English names, scientific (Latin) names, and temperament attributes. An accuracy of  94.0% and an F1-score of 92.7%, indicating strong performance despite challenges related to OCR noise and multilingual diversity. In a comparative analysis, it can be observed that our approach competes favorably with other NER models published in the recent past in the non-biomedical and biomedical domain. The results point to the possibility of deep learning-assisted NER as an effective tool of digitalizing classical Unani texts enabling the development of structured databases to be used in expert systems, knowledge graphs, and AI-driven drug discovery. The research contributes to the maintenance of cultural heritage and helps to combine the traditional medicine with the data-driven healthcare system. Future studies will  test what the combination of transformer-based models can do to nest an NER strategy, and extend the multilingual corpora with a view to further increasing recognition accuracy and domain-adaptability.