[Evaluation of the Automatic Full Form Retrieval Method from Abbreviation Using Word2vec for Terminology Expansion]

Area of Science:

• Medical Informatics
• Natural Language Processing
• Radiology

Background:

• Medical literature contains numerous abbreviations, complicating information retrieval and terminology management.
• Automated methods are needed to expand these abbreviations for improved consistency and accessibility.

Purpose of the Study:

• To automatically extract full forms of abbreviations using Word2vec.
• To optimize Word2vec parameters for maximum accuracy in abbreviation expansion.
• To enhance terminology consistency and efficiency in medical imaging research.

Main Methods:

• Collected ~300,000 PubMed abstracts on "image diagnosis" (1994-2018).
• Preprocessed text: lowercasing, symbol removal, compound word recognition (RadLex, JTR).
• Generated distributed representations using CBOW and Skip-gram models with varying iterations (3-85) and dimensions (50-1000).

Main Results:

• The Skip-gram model achieved the highest accuracy of 74.3% with 200 dimensions and 10 iterations.
• Skip-gram consistently outperformed the CBOW model across all tested parameter combinations.
• Identified full forms by cosine similarity between abbreviations and generated word vectors.

Conclusions:

• Word2vec achieved 74.3% accuracy in extracting full forms from abbreviations.
• This method enhances the consistency of medical terminology.
• The approach improves the efficiency of terminology expansion in the field of image diagnosis.