Publication

Abstract : Acronyms are commonly used in technical fields, such as science and medicine, but their potential expansions can be difficult to understand without proper context. This is where acronym disambiguation (AD) comes in - the process of determining the appropriate extension of an acronym in text. AD is essential for comprehending technical jargon in plain language, and requires effective representations of "words", "phrases", "acronyms", and "abbreviations" depending on their context. However, understanding the appropriate expansion for an acronym is not always straightforward. Words in natural language often have multiple meanings, and the context in which they are used can further increase the complexity of the text. For example, the acronym "AC" can potentially expand to "Adriamycin Cyclophosphamide", "ante cibum" i.e., before meals, or "Anticoagulation", among other possibilities. To address this challenge, we have developed a transformers-based model that can learn from contextual data and select the correct full form from a set of possible options for a given abbreviation. This study was conducted on the Clinical Acronym Sense Inventory (CASI) dataset. This study involved comparing different clinical versions of the BERT transformer architecture, including SciBert, ClinicalBert, and BioBert and observed that ClinicalBert performed the best with F score of 91.49%.