Although there are a number of evidence-based clinical guidelines available for diabetes screening, diagnosis, and management, these are usually general in nature and do not take into account the special property of AI patients. If we directly apply these recommendations to AI patients without considering their special socio-economic, cultural, ethnic, and geographical status, the usefulness of the recommendation will be jeopardized. For example, a proper diet recommendation is crucial for diabetes patients. We can find such general recommendation from existing guild lines. However, AI patients may have their unique diet preferences. Also, as many tribes are located in “food desert”, in which too many foods (such as seafood) are either too expensive or unavailable for the AI patients. If recommended with unavailable or unaffordable food, AI patients cannot get the benefits at all. Therefore, making personalized recommendation is especially important for them.
To address this issue, we propose an ontology-enhanced recommendation system to provide real-time personalized recommendation for AI diabetes patients. Shared ontologies (defined in Task 1) are used to achieve a common understanding of the domains in which the system operates. We will employ rule-based knowledgebase together with the predefined ontology, which is description logic in nature, to make personalized recommendation. In particular, we will model existing recommendations/guidelines regarding diabetes with First Order Logic (FOL). Then we make use of ontological logic to infer appropriate recommendations for AI patients, taking into account their physical condition and socio-economic, cultural, ethnic, and geographical status.
Details of this task are still under development. Currently, we have implemented a smart meal recommender and physical exercise recommender as a mobile application. The following screenshots illustrate the use of the system. More functions such as grocery shopping recommendation, nutrition recommendation, and social activity recommendations will be added into the system, and the prototype will be evaluated with AI users.