JMIR Publications

ABSTRACT

Background:

Health education emerged as an important intervention for improving the awareness and self-management abilities of chronic disease patients. The development of information technologies has changed the form of patient education materials from traditional paper materials to electronic materials. To date, the amount of patient education materials on the Internet is tremendous with variable qualities, which makes it hard to distinguish the most valuable materials by individuals lacking medical backgrounds.

Objective:

The aim of this study is to develop a health recommender system to provide appropriate educational materials for chronic disease patients in China and to evaluate the effect of this system.

Methods:

A knowledge-based recommender system was implemented using ontology and several natural language processing (NLP) techniques. The development process was divided into 3 stages. In stage 1, an ontology was constructed to describe patient characteristics contained in the data. In stage 2, an algorithm was designed and implemented to generate recommendations based on the ontology. Patient data and educational materials were mapped to the ontology and converted into vectors with the same length, then the recommendations were generated according to the similarity between these vectors. In stage 3, the ontology and the algorithm were incorporated into an mHealth system for practical use. Keyword extraction algorithms and pre-trained word embeddings were used to preprocess the educational materials. Three strategies were proposed to improve the performance of keyword extraction. The system evaluation was based on a manually assembled test collection for 50 patients and 100 educational materials. The recommendation performance was assessed using the macro precision of top-ranked documents and the overall mean average precision (MAP).

Results:

The constructed ontology contained 40 classes, 31 object properties, 67 data properties and 32 individuals. A total of 80 SWRL rules were defined to implement the semantic logic of mapping patient original data to the ontology vector space. The recommender system was implemented as a separate web service connected with patients' smartphones. According to the evaluation results, our system can achieve a macro precision up to 0.97 for the top 1 recommendation and an overall MAP score up to 0.628.

Conclusions:

This study demonstrated that a knowledge-based health recommender system has the potential to accurately recommend educational materials to chronic disease patients. Traditional NLP techniques combined with improvement strategies for specific language and domain proved to be effective for improving the system performance. One direction for future work is to explore the effect of such systems from the perspective of patients in a practical setting.