Publication

Abstract : The exponential growth of unstructured text data in the healthcare domain has posed significant challenges for the manual analysis and extraction of meaningful patterns and relations. The identification of relevant entities and extracting their meaningful relations are very important for proper medical care and advanced medical research. Biomedical Named Entity Recognition (BioNER) and Biomedical Relation Extraction (BioRE) play pivotal roles in addressing these challenges, as they facilitate accurate and quick recognition of pertinent entities and the extraction of meaningful relations. These tasks are fundamental for developing artificial intelligence applications like conversational agents and information retrieval systems. This study proposes a comprehensive approach to addressing the need for BioNER in the oral medicine and radiology domain. The proposed approach suggested for BioNER involves utilizing a combination of BioBERT embeddings and LSTM to fine-tune the model, facilitating the accurate extraction of relevant entities. The proposed approach was evaluated using various combinations of machine learning models and multiple feature encoding techniques. Among them, BioBERT was found to be outperforming all other baselines assessed in the study.