Community Clustering and Recommendation Toward Elderly Health Records: Probabilistic Representation Learning and Large Graph Search
Year: 2025
Type: article
Abstract: We develop a probabilistic representation learning method to automatically divide a large-scale elderly population into distinct health management communities by leveraging shared health conditions, treatment preferences, and behavioral tendencies. This approach enables more precise, efficient, and personalized healthcare interventions, wherein the key steps include: • Geometry-based feature selection from high-dimensional health data-This effectively reduces dimensionality while preserving discriminative health patterns. • Probabilistic multi-topic representations of individual health profiles-These latent space embeddings capture complex health-need interactions. • Graph-based community detection via shared clinical/behavioral traits-The algorithm identifies natural clusters with medical relevance. • Personalized intervention ranking by community relevance-This ensures tailored care plan recommendations for each subgroup. Validated on >1M health records, the method achieves a 0.5049 Balanced Error Rate, outperforming six baselines across 42 health categories. Ablation studies show 2.8 – 9.3% accuracy drops when removing key components, with graph mining and ranking mechanisms being most impactful (91.4% baseline precision).
Source: IEEE Access
Author: Huijiong Ding
Institutions: Wuzhou University, Zhejiang University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
OpenAlex ID: W4412836935