DeepMood: Forecasting Depressed Mood Based on Self-Reported - PDF document

DeepMood: Forecasting Depressed Mood Based on Self-Reported Histories via Recurrent Neural Networks ∗ † Yoshihiko Suhara Yinzhan Xu * Alex ‘Sandy’ Pentland Recruit Institute of Technology Massachusetts Institute of MIT Media Lab 444 Castro Street Suite 900 Technology 20 Ames Street Mountain View, CA 94041 77 Massachusetts Avenue Cambridge, MA 02139 suharay@recruit.ai Cambridge, MA 02139 pentland@mit.edu xyzhan@mit.edu ABSTRACT Depression is a prevailing issue and is an increasing problem in many people’s lives. Without observable diagnostic cri- teria, the signs of depression may go unnoticed, resulting in high demand for detecting depression in advance automati- cally. This paper tackles the challenging problem of forecast- ing severely depressed moods based on self-reported histo- ries. Despite the large amount of research on understanding individual moods including depression, anxiety, and stress based on behavioral logs collected by pervasive computing devices such as smartphones, forecasting depressed moods is still an open question. This paper develops a recurrent neu- ral network algorithm that incorporates categorical embed- Figure 1: Depressed mood forecasting task. ding layers for forecasting depression. We collected large- scale records from 2,382 self-declared depressed people to conduct the experiment. Experimental results show that cause of disease burden globally. In the United States, men- our method forecast the severely depressed mood of a user tal disorder problems are among the top five conditions for based on self-reported histories, with higher accuracy than direct medical expenditure, with associated annual health SVM. The results also showed that the long-term historical care costs exceeding $30 billion [29]. information of a user improves the accuracy of forecasting Due to the lack of physical symptoms, diagnosing depres- depressed mood. sion is a challenge. People cough or may have a fever when they are physically ill; these symptoms can lead them to go Keywords to hospitals for appropriate treatment. Without these phys- Depression; Neural Networks; Mobile Applications ical symptoms, the signs of depression may go unnoticed. Existing studies [24][35] show effective vital signs for depres- sion detection. A standard method is to measure biomarkers 1. INTRODUCTION such as serotonin to provide obvious evidence for depres- Depression is a prevailing mental health care problem and sion. However, obtaining such biomarkers requires special is a popular keyword due to the increase in mentally disor- apparatus and often invasive sensing. Not many people can dered patients including potential numbers. The WHO esti- adopt the approach to assess their mental health in daily life. mates that 676 million people in the world (nearly one in ten Because we lack a system to reveal mental illness through people) suffer from depression 1 . Current predictions by the physical signs, we need an external system that helps us de- WHO indicate that by 2030 depression will be the leading tect depression in a noninvasive manner. This need has led ∗ These authors contributed equally to this work. to a large body of work on depressed mood detection by † The author is also affiliated with MIT Media Lab. pervasive computing devices [3][5][10][30][36][48][50]. Early detection of depressed mood is essential to pro- 1 http://www.who.int/healthinfo/ vide appropriate interventions for preventing critical situa- tions. Despite a large amount of research on existing depres- sive mood prediction [3][5][10][30][36][48][50], forecasting de- � 2017 International World Wide Web Conference Committee c pressed moods has not been well studied. Therefore, we de- (IW3C2), published under Creative Commons CC BY 4.0 License. fine a novel task of forecasting depressed mood and develop WWW 2017, April 3–7, 2017, Perth, Australia. a predictive model for this task. In this paper, we distin- ACM 978-1-4503-4913-0/17/04. guish forecasting from prediction, emphasizing the meaning http://dx.doi.org/10.1145/3038912.3052676 of predicting future mood instead of existing mood. Par- ticularly, we focus on forecasting severe depression among several types of depressive moods in this paper. Severely . 715