Publication

Abstract : Recent years witnessed an unprecedented growth of unstructured text data in the healthcare domain. Manual analysis and extraction of useful patterns from these large data collections are extremely difficult. Biomedical named entity recognition (BioNER) identifies entities of clinical importance, such as symptoms, treatment, names of drugs, and organisms, from unstructured text documents such as electronic health records and radiology reports. Even though this is the first step for building artificially intelligent applications such as conversational agents and information retrieval, this is difficult and challenging. Recently pre-trained language models gained high popularity in NLP, considering their capacity to manage contexts satisfactorily. This paper proposes an approach that fine-tunes BioBERT, a state-of-the-art pre-trained biomedical language model for extracting named entities for the Oral Medicine and Radiology domain. This work compares various machine learning algorithms on different feature encoding techniques to establish the superior performance of the proposed approach. We assess the proposed approach by employing metrics like precision, recall, and f-measure and found it to be outperforming our chosen baselines.