Young Adults' Instagram Posts and Depressive Moods: A study in Mexico in the Wild

Iván A. Encinas-Monroy, Jessica Beltran, Luis H. Sánchez, Luis-Felipe Rodríguez, Adrián Macías, Cynthia. B. Pérez, Manuel Domitsu, Luis A. Castro

Abstract


Patterns of use of social networking sites like Instagram can be indicators of the mental state of users. Of particular interest to the HCI community are those markers and patterns useful for inferring the mental health of users experiencing depressive episodes or moods. Detecting individuals' depressive mood through their typical Instagram activity remains a challenge due to the diversity of the content posted. Previous research often focus on retrieving content of hashtags related directly to depression for analysis. Thus, although based on real posts, results can be highly biased. Analyzing all user posts in individuals' day-to-day life can yield ecologically valid findings, but it is challenging. We conducted an observational study aimed at detecting depressive moods of users from their Instagram posts. We analyzed text, images, and post behavior using two approaches: inferential statistics, and machine learning. Our results indicate that the time of day and the hue levels of a posted image could lead to the detection of depressive moods. Furthermore, our machine learning approach yielded up to 65\% of accuracy. Although our study yields ecologically valid findings, several challenges remain to be addressed due to the heterogeneity of the dataset, as it typically happens in real world studies.


Keywords


Social Networking Sites; Depressive Mood Detection; Instagram; Machine Learning; Behaviour Analysis; Image Analysis; Text Analysis; Transfer Learning

Full Text: PDF