Introduced by Youhee Choi, Jeff Gonda, Kexin Huang, Jackie Kim and Yuntao Wang
When so much of our digital presence is dictated by our social media accounts, we find ourselves subconsciously having a “first impression” of people based on their profiles.
So last semester, a team of UC Berkeley students, Jackie Kim, Jeff Gonda, Kevin Feng, Kexin Huang, Youhee Choi and Yuntao Wang, set out to learn more about the connection between social media and personal traits by looking into whether Facebook profile photos can help you predict something as subjective and complex as personality.
The students used 32,000 photos from members of self-identified Myers-Briggs Facebook groups for the project. They applied Amazon Rekognition image analysis API to the photos, and used machine learning algorithms such as KNN, Logistic Regression, SVM, Random Forest and CNN to build predictive models.
Ultimately, they were able to predict each Myers-Briggs Type Indicator component with about 55 percent accuracy, which is not decisive. Nonetheless, the students say they were able to glean some interesting insights about the tendencies associated with each component of the Myer-Briggs Type Indicators.