There were some papers using Twitter to predict depression.
Park et al. (2012) applied sentiment analysis on tweets and showed
that Twitter provided meaningful data for clinical studies on
depression. De Choudhury et al. (2013) from Microsoft Research
compared tweet text of depressed Twitter users to those of the
normal users and highlighted the potential of Twitter as a tool for
predicting MDD. Harman, Coppersmith, and Dredze (2014) pointed
out that although Twitter users are not a representative sample of
the entire population suffering from MDD, individual level and
population level analysis can still be made because of the diverse
set of quantifiable signals related to MDD in Twitter. A recent survey
revealed that 26 percent of the online U.S. adults discussed
their health information online, and 42 percent of them use social
media to post or seek information about health conditions (GE
Healthcare, 2012). Rudd et al. (2006) showed that signs identified
by the American Association of Suicidology were frequently
included on websites. Among all the signs, the largest individual
category for psychological symptoms is about depression and
anxiety states. Additionally, they found that young adults like to
text to share their feelings.