Instagram photos reveal predictive markers of depression.
In a study conducted by a Harvard researcher Andrew G. Reece, it was concluded that there were key indicators of depression that lie within one's Instagram profile. The computer trained model was able to outperform the average general practitioner whose success rate for diagnonsis was around 42%.
Picture Perfect is a piece of software which
- fetches a user's instagram profile
- downloads all the images from posts
- uses several key indicators of depression from social media to assign a health score
- displays the likelihood of that instagram user having depression
A 34-page research paper was written on how Instagram photos reveal predictive markers of depression. The researchers include Andrew G. Reece who is part of the Psychology Department at Harvard University.
Abstract Using Instagram data from 166 individuals, we applied machine learning tools to successfully identify markers of depression. Statistical features were computationally extracted from 43,950 participant Instagram photos, using color analysis, metadata components, and algorithmic face detection. Resulting models outperformed general practitioners’ average diagnostic success rate for depression. These results held even when the analysis was restricted to posts made before depressed individuals were first diagnosed. Photos posted by depressed individuals were more likely to be bluer, grayer, and darker. Human ratings of photo attributes (happy, sad, etc.) were weaker predictors of depression, and were uncorrelated with computationallygenerated features. These findings suggest new avenues for early screening and detection of mental illness.
There were several criteria we considered when constructing the health score, mainly derived from the research paper.
- Hue, Saturation, Brightness
Depressed individuals were found to have a more blueish tint in their photos which is quantified through a higher value in hue. Depressed individuals were also more likely to posts photos that did not have a high saturation value nor a high brightness, meaning their photos would be on the darker side with a more blue tone.
- Comments to Likes ratio
It was found that depressed individuals were more likely to received more comments on their posts, while simultaneously receiving less likes. To implement this in our program we parse the meta data of the user's posts and found the number of likes and number of comments and created a ratio of # comments / # likes
. This ratio would indicate a higher likelihood of an individual being depressed if it were closer to one.
- Facial Detection
There lies a relationship between the number of people in a photo and a user's mental health. Having an image where there are more people present indicates mental wellbeing. To implement this in our program, we ran each image through a facial recognition software that outputs the number of faces there are in a given image which allowed us to assign a score to the user based on this factor.
- Number of posts
A mentally unhealthy individual was found to post more on average and per day than a mentally healthy individual. To implement this, we found the metadata concerning number of posts user had. The higher this number the higher score they were assigned.
- Caption Analysis
Future Steps Lastly, we plan to use sentiment analysis on the caption of each image which can contribute to the health score of the user.
To get up and running with this project you need google application credentials and python3. First clone this project.
git clone [email protected]:maaslalani/PicturePerfect.git
Google Application Credentials
After you get you secret credentials type the following command, replace the filepath with the path to your secret credentials.
export GOOGLE_APPLICATION_CREDENTIALS=~/Documents/Projects/Hack-Harvard-2018/PicturePerfect.json
Install python3
with homebrew
brew install python3
Install dependencies
pip3 install -r requirements.txt
To run the main program use the following command.
python3 main.py
--help
, --verbose
, and --debug
are valid flags.