An Initial Study of Abbreviation Disambiguation in Vietnamese Clinical Texts

Abbreviations are ubiquitously used in clinical texts of electronic medical records (EMRs) in the world due to their record simplification when created under high pressure. However, they made EMRs less understandable and sharable for both human users and computer programs. Such a case hinders automatically processing EMRs in many various activities of both medical and other related fields. Therefore, abbreviation resolution and its subtask, abbreviation disambiguation, in clinical texts have been well studied worldwide. Nonetheless, most of the existing works were proposed for English clinical texts. No work has been defined for Vietnamese clinical texts. This must be a shortage when EMRs get more and more popular in Vietnam. On the other hand, abbreviation disambiguation faces the data imbalance problem that has not yet been discussed much in the existing works. In such a context, we propose an initial study of abbreviation disambiguation in Vietnamese clinical texts. The study is the first one that explores three different approaches and figures out the best effective solution to the task. It also defines new features in addition to the traditional one to make the solution better than the existing ones. The experimental results on real Vietnamese EMRs confirm the contributions of our work.