Minsu Park*, KAIST (Korea Advanced Institute of Science and Technology), Daejeon, Korea, Republic Of Chiyoung Cha, Ewha Women's University, Seoul, Korea, Republic Of Meeyoung Cha*, KAIST (Korea Advanced Institute of Science and Technology), Daejeon, Korea, Republic Of Track: Research Presentation Topic: Web 2.0 approaches for behaviour change, public health and biosurveillance Presentation Type: Oral presentation Submission Type: Single Presentation Building: Joseph B. Martin Conference Center at Harvard Medical School Room: C-Rotunda Room Date: 2012-09-15 04:00 PM – 04:45 PM Last modified: 2012-09-16

If you are the presenter of this abstract (or if you cite this abstract in a talk or on a poster), please show the QR code in your slide or poster (QR code contains this URL).

Background
Depression is the most commonly diagnosed mental disorder in many developed countries. Despite increasing public knowledge and awareness, many individuals with depression go undetected and untreated. While public programs such as the National Depression Screening Day are an important step towards decreasing the prevalence of undiagnosed depression, their main limitation lies in the selection bias of people they can reach, because the programs are participation-oriented. In terms of reaching vulnerable individuals, one useful addition to existing screening methods is to utilize a large amount of content individuals share on online social networks. In websites like Facebook and Twitter, hundreds of millions of users across a wide spectrum of demographics post their moods and thoughts in real-time. Data embedded in these sites hence could provide a cost-effective way to study health behaviors from non-clinic-based populations.

Objective
We would like to determine the degree to which online sentiments of depressed users in Twitter differ from non-depressed users and to understand the linguistic features in tweets that are reflective of one’s depressive state.

Methods
The analysis was conducted in three steps: (1) classifying 69 participants into three depressive levels based on the Center for Epidemiological Studies-Depression (CES-D) scale, which is a self-judged survey; generally depression symptoms are divided into three groups based on the likelihood of having depression: low (0-15), mild to moderate (16-22), and high range (23-60), (2) collecting the tweets of the participants, and (3) comparing the online sentiments with the offline counterpart. We measured the sentiment scores of online tweets by Linguistic Inquiry and Word Count (LIWC), a text analysis program that counts words in psychologically meaningful categories. We ran multiple regression with CES-D and LIWC scores.

Results
1. Depressed users had a higher chance of using words in certain affect categories such as “anger (e.g., hate, kill, annoyed)”, “causation (e.g., because, effect, hence)”, and “friends (e.g., buddy, neighbor, friend)” . On the other hand, the usage of words related to “communication (e.g., tell, speak, claim)” and “tentative (e.g., maybe, perhaps, guess)” were dropped for the depressed users.
2. The qualitative content analysis suggests that depressed users were more likely to post tweets about themselves than to interact with other users seen by increase usage of first person pronouns compared to non-depressed Twitter users.

Conclusions
Our results imply that it is possible to use online social network data for exploring public behaviors in relation to depression. While preliminary, we expect to gain more insights from using social network data for healthcare purposes in the future with the goal of building a depression screening system in online social networks to decrease the prevalence of undiagnosed depression.