munmun de choudhury assistant professor, school of interactive computing, georgia tech march 12,...
TRANSCRIPT
Munmun De ChoudhuryAssistant Professor, School of Interactive Computing, Georgia TechMarch 12, 2014
Online Social Dynamics and Well-being
Photo courtesy NPR
health and wellness
affective disorders are a serious challenge in public health—statistics under-reported in large populations
Three problems
Photo courtesy of NIDA.gov
Examine patterns of activity and emotional correlates for childbirth and postnatal course leveraging activity in social media (Twitter)
(De Choudhury, Counts, Horvitz, CSCW 2013a)Honorable Mention Award
Research Question:
Behavioral Changes of New Mothers
Blue line represents approximate time of childbirth. The red line represents mothers and the green line represents the background cohort
(De Choudhury, Counts, Horvitz, CSCW 2013a)
Individual-level ComparisonCertain mothers show more drastic change in behavioral measures than others
(De Choudhury, Counts, Horvitz, CSCW 2013a)
Language Differences
size of vocabulary (#unigrams)
percentage of unigrams w/ p < .01
mothers w/ large effects 17,117 27.65%mothers w/ small effects 33,785 19.74%background cohort 47,214 3.86%
percent of all unigrams in the language vocabulary used that changed significantly in usage frequency after childbirth, i.e. with p < .01 based on paired sample t-tests
background cohort mothers w/ small effects mothers w/ large effects
now (↓), shit (↑), back (↑), that (↑), day (↓), life (↑), time (↓), them (↑), me (↑), you (↑), fuck (↑), today (↓), sleep (↑), tonight (↓), love (↓), good (↓), here(↓), her (↓), morning (↑), tomorrow (↑), go (↑), know (↑), him (↓), people (↓)
#past (↑), duh (↑), people (↓), photo (↑), post (↑), decision (↓), reunite (↓), women (↑), story (↑), time (↑), asap (↓), do (↑), life (↓), wait (↑), fired (↑), days (↑), happy (↓)
haha (↓), blessed (↑), lol (↓), #lifecangetbetter (↑), awesome (↓), monthly (↑), fantastic (↓), cuddle (↑), home (↑), love (↓), sick (↑), aww (↑), scary (↑)
top unigrams showing the most change (in usage frequency) in the postnatal period, compared to the prenatal phase
(De Choudhury, Counts, Horvitz, CSCW 2013a)
Greedy Differencing Analysis
o determine the number of unigrams whose change in usage frequencies renders the mothers with large effects significantly different
o the deviations for certain new mothers captured by a rather small number of unigrams:• 1.16% compared to mothers
with small effects• 10.73% with respect to the
background cohort
(De Choudhury, Counts, Horvitz, CSCW 2013a)
Facebook & Postpartum Depression• Web survey to recruit mothers who were Facebook
users and who gave birth to a child within the last nine months or less
• To incentivize participation, mothers were entered into the random drawing of four $500 Amazon gift cards
• Survey was active between mid-July and mid-September, 2012
• Advertised through:• Mailing list of new mothers at Microsoft• Neighborhood based mommy blogs in the Seattle metro area• Postings from our organization’s official Twitter and Facebook accounts• Our personal Twitter, Facebook and Google+ accounts• Paid Facebook ads targeting mothers in the age group 20-39 years• Sponsored posts on BabyCenter (babycenter.com)
(De Choudhury, Counts, Horvitz, Hoff, CSCW 2014a)
Patient Health Questionnaire (PHQ-9)• the 9-item questionnaire seeks responses over the past two week period• based directly on the nine diagnostic criteria for major depressive
disorder in the DSM-IV (Diagnostic and Statistical Manual Fourth Edition)• scores on the PHQ-9 range from zero to 27; individuals with scores 15 or
above are considered to be moderately severe to severely depressed (Kroenke et al. 2001) Not at
allSeveral days
More than half the days
Nearly everyday
Little interest or pleasure in doing things
0 1 2 3
Feeling down, depressed, or hopeless 0 1 2 3
Trouble falling or staying asleep, or sleeping too much
0 1 2 3
Feeling tired or having little energy 0 1 2 3
Poor appetite or overeating 0 1 2 3
Feeling bad about yourself — or that you are a failure or have let yourself or your family down
0 1 2 3
Trouble concentrating on things, such as reading the newspaper or watching television
0 1 2 3
Moving or speaking so slowly that other people could have noticed? Or the opposite — being so fidgety or restless that you have been moving around a lot more than usual
0 1 2 3
(De Choudhury, Counts, Horvitz, Hoff, CSCW 2014a)
Prediction Task
Prenatal +10d +20d +30d PostnatalDev. 61.32 57.39 52.58 46.61 58.15LL -30.668 -26.3 -20.34 -17.24 -28.84psd. R2 0.355 0.383 0.439 0.484 0.372sfit 0.72 0.68 0.62 0.57 0.703N 156 156 156 156 156Error df 100 100 100 100 100
• The prenatal period does provide PPD-predictive information together with a brief period of postnatal observations.
• Aligns with findings in the clinical literature where prepartum depression is known to be a good indicator of PPD (Beck 2001).
(De Choudhury, Counts, Horvitz, Hoff, CSCW 2014a)
Low-cost, privacy-preserving mechanisms to identify new mothers’ behavior can improve social support and encourage postpartum wellness
(De Choudhury, Counts, Horvitz, CSCW 2013a)Honorable Mention Award
Study Methodologyo Ground truth data on clinical depression
condition of 476 individuals was collected through a behavioral studyo crowdsourcing (Mechanical Turk) driven methodologyo use of the CES-D depression screening survey (Center for
Epidemiologic Studies Depression Scale); an auxiliary screening test—Beck’s Depression Inventory was used to reduce noisy responses
data collection methodology (twitter)
(De Choudhury, Gamon, Counts, Horvitz, ICWSM 2013)
Social media characteristics of MDD
(De Choudhury, Gamon, Counts, Horvitz, ICWSM 2013)
Depressive language useTheme UnigramsSymptoms anxiety, withdrawal, severe, delusions, adhd, weight,
insomnia, drowsiness, suicidal, appetite, dizziness, nausea, episodes, attacks, sleep, seizures, addictive, weaned, swings, dysfunction, blurred, irritability, headache, fatigue, imbalance, nervousness, psychosis, drowsy
Disclosure fun, play, helped, god, answer, wants, leave, beautiful, suffer, sorry, tolerance, agree, hate, helpful, haha, enjoy, social, talk, save, win, care, love, like, hold, cope, amazing, discuss
Treatment medication, side-effects, doctor, doses, effective, prescribed, therapy, inhibitor, stimulant, antidepressant, patients, neurotransmitters, prescriptions, psychotherapy, diagnosis, clinical, pills, chemical, counteract, toxicity, hospitalization, sedative, 150mg, 40mg, drugs
Relationships, life
home, woman, she, him, girl, game, men, friends, sexual, boy, someone, movie, favorite, jesus, house, music, religion, her, songs, party, bible, relationship, hell, young, style, church, lord, father, season, heaven, dating
(De Choudhury, Gamon, Counts, Horvitz, ICWSM 2013)
Predicting MDD
mean frequency—the average measure of the time series of a feature at any given day: µi=(1/N)∑tXi(t).
variance—the variation in the time series: (1/N)∑t(Xi(t) −µi)2.
mean momentum—relative trend of a time series, compared to a period before: (1/N)∑t(Xi(t)-(1/(t-M))∑(M≤k≤t-1)Xi(k)).
entropy—the measure of uncertainty in a time series: −∑tXi(t)log(Xi(t)).
(De Choudhury, Gamon, Counts, Horvitz, ICWSM 2013)
Social media depression index
least squares regression fit yields correlation of 0.52
actual (NIMH data) predicted (SMDI)
( ) ( )( ) d d s s
d s
n t n tSMDI t
standardized difference between frequencies of depression-indicative and standard posts, compared to a period before between k and t-1 (1≤k≤t-1)
(De Choudhury, Counts, Horvitz, WebSci 2013)
(De Choudhury, Counts, Horvitz, WebSci 2013)
The Atlantic
(De Choudhury, Monroy-Hernandez, Mark, CHI 2014)Best Paper Award
A Case Study
• The Mexican Drug War is an example of protracted trauma that has exposed people to persistent acts of violence.
• Many Mexican cities affected• rapid increase of shootings and homicides, loss
of life of innocent civilians. • increase of criminal activities such as extortions,
and kidnappings• as of 2011, the Drug War had claimed 60,000
lives and had displaced between 230,000 and 1.6 million people. However unconfirmed reports set the homicide statistics over 100,000 victims (Booth, 2012)
(De Choudhury, Monroy-Hernandez, Mark, CHI 2014)
Challenges
• As early as 2010, local health officials had reported a significant increase in the number of people seeking mental health help with post-traumatic stress disorder (PTSD) induced by drug-related violence (O’Connor, 2013).
• The international news media have reported how Mexicans are “numb to carnage” (Archibold & Cave, 2012) and even kids are “exposed to such violence that they’re desensitized” (Hopewell, 2013).
(De Choudhury, Monroy-Hernandez, Mark, CHI 2014)
Goal: Study affective responses in social media and how they might indicate desensitization to violence experienced in communities embroiled in an armed conflict
(De Choudhury, Monroy-Hernandez, Mark, CHI 2014)
Negative Affect
Towards the beginning of the time period of analysis, i.e., early on in 2010 or early 2011, the peaks in number of homicides are actually correlated with those in NA. However over time, especially in 2012, that ceases to be the case
(De Choudhury, Monroy-Hernandez, Mark, CHI 2014)
Activation, Dominance
Over time, from the trends of activation, we observe a general increase, implying that Twitter users mentioning the four cities in their postings, were increasingly using higher intensity emotions.Dominance shows a rise with persistent violence (ref. the slope of the linear fit), indicating that users are increasingly using dominating and aggressive emotions
(De Choudhury, Monroy-Hernandez, Mark, CHI 2014)
borderlandbeat.com
Thanks!Questions?
[email protected]@munmun10