Toward an Advice Agent for Diet and Exercise Based on Diary Texts

摘要

Many people are interested in the prevention of disease and health-enhancing activities. In Japan, most of these people are anxious about some health-related issue such as decreased strength, stress, and chronic diseases. To improve one's health, it is very important to manage diet and exercise. Various systems give advice on information easily captured by sensors, including body weight and exercise time. Considering that such health activities are likely to be recorded as text, we are developing a system to generate humanlike advice for diet and exercise based on health related blog articles. When a user posts a blog article reporting his/her diet and/or exercise activities, the system posts advice for the article. The system extracts health information (diet and exercise) from blog articles using machine learning-based models that are trained on pairs of blog and advice articles. Herein, we report the current status and future work of our research. In the first phase, we created a corpus consisting of blog articles on health and advice on some of these, as the knowledge resource for the health advice system. To date, 1402 blog articles and 386 advice articles have been created. We are currently annotating the articles to extract health-related information.