Simulation and annotation of global acronyms.

Area of Science:

• Biomedical Natural Language Processing
• Computational Linguistics
• Data Science

Background:

• Global acronyms in scientific texts often lack definitions, leading to ambiguity and interpretation challenges.
• Supervised machine learning for acronym sense disambiguation requires extensive training data, which is difficult to acquire in clinical settings due to privacy concerns and manual annotation bottlenecks.

Purpose of the Study:

• To develop an automated method for modifying scientific abstracts to simulate global acronym usage.
• To annotate acronym senses without external resources or manual effort, thereby facilitating dataset creation.
• To generate large datasets for training supervised word sense disambiguation models for biomedical acronyms.

Main Methods:

• An approach was proposed to automatically modify scientific abstracts.
• The method simulates the use of global acronyms and annotates their senses.
• A web-based application was implemented to facilitate this process.

Main Results:

• The developed approach enables the automatic generation of annotated datasets for acronyms.
• This method addresses the challenge of data scarcity in biomedical natural language processing.
• The generated datasets can be used to train supervised models for acronym sense disambiguation.

Conclusions:

• The proposed automated method effectively generates large, annotated datasets for biomedical acronyms.
• This approach alleviates the need for manual data annotation and external resources.
• The generated datasets will support the advancement of word sense disambiguation techniques in the biomedical domain.