Using Artificial Intelligence for Diabetes Distress Detection among the Facebook Community

Diabetes distress is an emotional reaction experienced by patients and caregivers, and to-date; studies have mainly focused on surveys to gauge patients’ distress levels. With the advent of social media and Artificial Intelligence, automatically detecting distress based on textual data is made possible. This paper aims to detect distress among the Facebook diabetes community using emotion analysis, linear regressions and dictionaries. Facebook posts were gathered from official diabetes pages resulting in a total of 2244 posts tagged with user reactions. The emotion analysis performance was found to be satisfactory, with the average scores for all the metrics between 65% and 82%. Three significant Facebook user reactions were observed, namely, anger (angry emotion), love and wow (fear) and wow and sad (sadness). The incorporation of these predictors was found to improve distress detection as the model outperformed the baseline, with an average accuracy of 76%.