Recommending Education Materials for Diabetic Questions Using Information Retrieval Approaches

Area of Science:

• Information Retrieval
• Medical Informatics
• Patient Education

Background:

• Effective diabetes self-management relies on accessible, expert-vetted patient education materials.
• Facilitating patient self-management requires robust methods for answering patient queries accurately.

Purpose of the Study:

• To evaluate information retrieval techniques for recommending patient education materials tailored to diabetic patient questions.
• To compare Latent Dirichlet Allocation (LDA) topic modeling and semantic group-based models against the vector space model (VSM) for this task.

Main Methods:

• Compared LDA topic modeling and semantic group-based retrieval models with VSM on diabetic questions from the TuDiabetes forum.
• Utilized a gold standard dataset of 50 questions with expert-assigned relevancy ratings for evaluation.
• Assessed performance using precision of top-ranked documents.

Main Results:

• The topic modeling-based model demonstrated superior performance compared to semantic group-based and VSM models.
• Precision for the top-retrieved document was 67.0% for topic modeling, 62.8% for semantic groups, and 54.3% for VSM.
• Significant differences in word mapping to the Unified Medical Language System (UMLS) were observed (P<.001).

Conclusions:

• Topic modeling effectively addresses vocabulary disparities, enhancing the recommendation of educational materials for patient questions.
• The findings suggest topic modeling as a promising approach for improving patient self-management support in diabetes care.
• Future research should explore generalizability across different diseases, resources, and online platforms.