Melancholic Ramblings on Twitter: Detecting Depression from Social Media Activity

Camille Marie Ruiz ( 1651218 )

Depression is a serious and prevalent mental illness, affecting 300 million people worldwide. One symptom of depression is rumination, wherein a person thinks about a thought repetitively. Social media sites, such as Twitter, are rich sources of data from which rumination may be detected since these are avenues where people can ramble about their thoughts and feelings. In light of the repetitive nature of rumination, this study uses Twitter data to explore whether or not coherence over a period of time can detect rumination and help in estimating depression. For this task, cosine similarity was used to calculate coherence while time window sizes were applied to account for period of time. Using different SVM kernels to estimate depression, we found that our coherence feature only produced a negligible effect in accuracy across all time window sizes. Nonetheless, the coherence feature was part of the top important features using the SVM linear kernel, indicating its potential. We also found that mental health topics for depressed users and topics on activities for healthy users were used as important features for classifying users regardless of time window size.