Sign In

Communications of the ACM

ACM TechNews

How an Algorithm Learned to Identify Depressed Individuals By Studying Their Instagram Photos

View as: Print Mobile App Share:
The researchers also evaluated the photographs using objective measures such as the average hue, color saturation, contrast, and so on.

Researchers at Harvard University and the University of Vermont have trained a machine to spot depression on Instagram.

Credit: Technology Review

Researchers at Harvard University and the University of Vermont have trained a machine-learning algorithm to spot warning signs for depression on Instagram by analyzing the composition of posted images.

The researchers recruited 170 participants via Amazon's Mechanical Turk service, 70 of whom were diagnosed with clinical depression. For each healthy user, the researchers chose their 100 most recent Instagram posts to be rated, and for depressed individuals, the 100 pre-diagnosis photographs were analyzed. A separate set of Turk participants rated how interesting, likeable, happy, and sad each photo appeared on a scale of zero to five. Photographs also were objectively rated by measuring the average hue, color saturation, contrast, and the number of faces in each image. A machine-learning algorithm then was used to find correlations between depression and image properties, and results show individuals with depression are more likely to post darker images that are bluer or grayer than healthy individuals. Depressed people also were found to post more photos with faces, but they tended to post fewer faces for each photo.

After testing the algorithm on another 100 individuals, the program correctly identified 70% of those with depression; the researchers say their findings support the notion that changes in individual psychology can be identified via social media.

From Technology Review
View Full Article


Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account