Abstract

The advances in mobile technology and natural language processing have made chatbots suitable for personal health care management. When the world’s population is aging, diabetes becomes one of the most common chronic diseases in the world. After being discharged from a hospital, diabetes patients must conduct personal health care management such as monitoring blood glucose, professional diet advice, and regular exercise reminders to control the disease. Unfortunately, it has been found that there are few chatbots designated for diabetes patients, especially in the Chinese language. To fulfill the need, this research proposes a diabetes self-management chatbot to assist patients in recording their blood glucose, exercise, and diet through conversation. The proposed chatbot system consists of four main components: a dialog controller, a neural network, a personal database, and a diabetic management rule base. Iterated Dilated Convolutional Neural Network with Conditional Random Field (ID-CNN-CRF) is applied for Named Entity Recognition (NER) to achieve high chatting quality. The experiments show that the ID-CNN-CRF outperforms the other three popular CNN, LSTM, and Bi- LSTM-CRF models regarding intention prediction accuracy and response time. In addition, the chatbot can advise people with diabetes on proper diet and exercise. The feedback from the diabetes caregivers and patients shows that the proposed chatbot is recognized as a convenient self-management tool.