Automatic Labeled Dialogue Generation for Nursing Record Systems.

Area of Science:

• Health Informatics
• Natural Language Processing
• Artificial Intelligence

Background:

• Digital voice assistants are crucial for improving nursing productivity in residential care.
• Training natural language understanding (NLU) modules requires extensive labeled data, which is costly and time-consuming to obtain, especially in specialized domains like nursing.

Purpose of the Study:

• To propose an automatic dialogue labeling framework for NLU tasks in nursing record systems.
• To overcome the limitations of manual data labeling by developing efficient, automated methods.
• To enhance the development of NLU modules for digital voice assistants in healthcare settings.

Main Methods:

• Utilized data augmentation techniques to generate diverse sample utterances.
• Investigated deep generative models, including a character-based Long Short-Term Memory (LSTM) model, for data generation.
• Implemented intent and entity labeling using feature embeddings and semantic similarity-based clustering, evaluating various embedding methods like fastText.

Main Results:

• The LSTM model achieved 90% accuracy and generated reasonable texts with a BLEU score of 0.76.
• fastText embeddings demonstrated strong performance in intent labeling (0.79 accuracy) and entity labeling (0.78 F1-score).
• Clustering tasks achieved silhouette scores of 0.67 for intent and 0.61 for entity labeling, indicating effective data representation.

Conclusions:

• The proposed automatic framework effectively addresses the challenge of limited labeled data for NLU in nursing.
• Automated labeling using data augmentation and embedding-based clustering significantly enhances NLU model development for nursing documentation.
• This approach offers a scalable and cost-effective solution for deploying advanced voice assistant technologies in healthcare.