Clustering Nursing Sentences - Comparing Three Sentence Embedding Methods.

Area of Science:

• Health Informatics
• Natural Language Processing
• Clinical Nursing

Background:

• Qualitative analysis of large text datasets in health sciences can be augmented by text embeddings.
• Effective searching and clustering of health information are crucial for data analysis.

Purpose of the Study:

• To evaluate three sentence-level embedding methods for clustering nursing narratives.
• To identify the most effective embedding technique for analyzing patient care episodes.

Main Methods:

• Utilized two BERT-based language models and one Sent2Vec method for sentence embeddings.
• Clustered sentences from 20 individual patient care episodes.
• Manually evaluated the quality of the generated clusters.

Main Results:

• Embeddings from a BERT model fine-tuned for predicting nursing subject headings produced the best clustering outcomes.
• This approach demonstrated superior performance in organizing and analyzing nursing narrative data.

Conclusions:

• Fine-tuning BERT models for specific proxy tasks can enhance their effectiveness in health information analysis.
• Sentence-level embeddings, particularly from fine-tuned BERT models, show promise for augmenting qualitative data analysis in nursing.