Term-Based Personalization for Feature Selection in Clinical Handover Form Auto-Filling

Area of Science:

• Information Retrieval
• Machine Learning
• Natural Language Processing

Background:

• Feature selection is crucial for model performance and efficiency.
• Traditional methods apply uniform feature sets, risking performance degradation due to irrelevant features.
• Term-specific feature relevance is often overlooked.

Purpose of the Study:

• To propose a novel term-based personalization approach for feature selection.
• To enhance the accuracy and reduce the complexity of information processing by tailoring features to individual terms.
• To address the limitations of traditional uniform feature set application.

Main Methods:

• Features are provided as input, focusing on selection strategies.
• A term-feature probabilistic relevance model is employed to evaluate feature subset importance for each term.
• A feature searching method generates candidate subsets to manage computational intensity.

Main Results:

• Personalized feature sets are generated for each term.
• Experimental results on the NICTA Synthetic Nursing Handover dataset demonstrate the approach's effectiveness.
• The proposed method shows promising performance improvements.

Conclusions:

• Term-based personalization significantly enhances feature selection.
• The approach effectively mitigates performance issues caused by noisy or irrelevant features.
• This method offers a more nuanced and effective strategy for feature selection in various research areas.