Abstract
Online health communities have become vital resources for medical information, yet they suffer from a significant mismatch between patient queries and physician responses. This paper presents a novel question-routing framework that adaptively fuses two critical dimensions of a physician's profile: technical expertise and professional attitude. By leveraging hypergraph-based learning to model complex medical knowledge and quantifying behavioral traits like responsiveness and reputation, the system ensures patients are matched with doctors who are both qualified and likely to engage. Experimental results on real-world data from HealthTap demonstrate that this multiview approach significantly outperforms traditional expert discovery methods in both accuracy and subjective quality.
Introduction
As digital health literacy grows, millions of users turn to Community-based Health Services (CHSs) such as HealthTap for medical advice. However, these platforms face a dual challenge: the "vocabulary gap" between patient layman terms and professional medical terminology, and the "silent expert" problem, where highly qualified physicians may be inactive or provide perfunctory responses. Effective question routing—automatically directing a query to the most suitable doctor—is essential to maintain the health and utility of these online ecosystems.
The Proposed Framework
The researchers propose a "Multiview Physician-Specific Attribute Fusion" framework. Unlike previous models that rely solely on text matching, this system evaluates physicians across two distinct "views": Expertise and Attitude.
1. Expertise Modeling
To bridge the vocabulary gap, the model employs Hypergraph-based Learning. While standard graphs represent pairwise relationships, hypergraphs can link multiple physicians who share commonalities, such as co-answering specific questions or peer endorsements. This captures high-order community structures and latent expertise. This is further refined using Kernel Density Estimation (KDE) to map how well a new question aligns with a doctor's historical knowledge base.
2. Attitude Modeling
Technical skill alone does not guarantee a helpful answer. The framework quantifies four behavioral attributes:
- Activity: Response latency and frequency.
- Responsibility: The rate at which a doctor's answers are selected as "best" or endorsed by peers.
- Reputation: General standing within the community based on historical feedback.
- Willingness: The overall likelihood of a physician to accept and answer a routed question.
3. Adaptive Fusion
The core innovation is the Adaptive Fusion mechanism. The system recognizes that the importance of expertise versus attitude varies by question type. A general wellness query might prioritize a highly responsive (high attitude) doctor, whereas a complex diagnostic query requires a deep specialist (high expertise).
The system uses Pace Regression to dynamically calculate a fusion weight ($\lambda$) for each question, balancing these two views to optimize the recommendation.
Experimental Evaluation
The model was tested using a comprehensive dataset from HealthTap, comprising over 3,000 physician profiles and 142,000 questions.
Key Results
- Objective Performance: The proposed model achieved a Hit@20 (H@20) score of 0.347, outperforming standard baselines like Latent Dirichlet Allocation (LDA) and K-means clustering (0.311).
- Subjective Quality: Human medical experts evaluated the top recommendations, awarding the model a Subjective@10 (S@10) score of 0.94, indicating that nearly all top-10 suggestions were clinically appropriate matches.
- Efficiency: The system operates in real-time, processing routing requests in under five seconds.
Conclusion
By treating question routing as a multidimensional matching problem, this research moves beyond simple keyword search. The integration of behavioral "attitude" metrics with advanced hypergraph-based expertise modeling provides a more human-centric approach to online medical consultation, ensuring that patients receive not just accurate information, but timely and empathetic care. Future work will explore the temporal evolution of physician expertise and potential biases in response patterns across different medical specialties.
Summary updated