Best Paper Award winning paper presented at ASONAM 2015.
Derek Doran, Samir Yelne, Luisa Massari, Maria-Carla Calzarossa, LaTrelle Jackson, Glen MoriartyDept. of CSE, Professional Psych, Wright State University, USADept. of Electrical, Computer, and Biomedical Eng., University of Pavia, Italy
7 Cups of Tea, Inc.
http://knoesis.wright.edu/doran
Bicycle Safety in Focus: Preventing Fatalities and Seeking Justice
Stay Awhile and Listen: Emotional Support Processes on the Social Web
1. Stay Awhile and Listen:
User Interactions in a Crowdsourced Platform
Offering Emotional Support
Derek Doran, Samir Yelne, Luisa Massari, Maria-Carla Calzarossa, LaTrelle Jackson, Glen Moriarty
Dept. of CSE, Professional Psych, Wright State University, USA
Dept. of Electrical, Computer, and Biomedical Eng., University of Pavia, Italy
7 Cups of Tea, Inc.
http://knoesis.wright.edu/doran
3. The Web as an emotional support tool
They are effective
Giving and receiving emotional support in CMSS groups has positive
effects on emotional well-being for breast cancer patients with higher
emotional communication (Yu et al., CHB 2015)
[Participation is] correlated with activity level: the higher the number of
posts and replies, the lower the level of distress in following months
(Barak et al. 2006)
They are popular
(Wang et al., CSCW 2012): Over 1.5 million online forum posts on cancer
support exist
In 2005: 36M people in the US alone claimed they were members of
online emotional support groups
In 2006: 94k online support groups on Yahoo!
And now we are 10 years past these numbers!
4. The Web as an online emotional support tool
But they are not well studied…
Meta-analysis of medical, social science, other online pub databases
revealed a handful (514) from 2004 – 2013 (Chou et al. 2013)
267 (52%) are reviews and commentary
213 (41%) are descriptive studies
34 (6.6%) involve intervention studies
And now, the social web is changing everything…
6. Emotional Support in the Social Computing Age
Traditional OSNs and social media are not suitable avenues to give/receive
emotional support
“Our study highlighted a set of goals that people pursue in order to
enhance their health—emotional support, motivation, accountability, and
advice—and identified the tension between the pursuit of these goals with
impression management” (Newman et al., CSCW 2011)
Enter 7 cups of tea: A new online social service designed to anonymously
connected those needing help with a crowd of active listeners
7cot follows a pioneering model for professional psychology:
Light-weight, 24/7 access to emotional support through
paraprofessionals
(Contrast: heavy-weight, scheduled, expensive support
through a licensed clinician)
7. Research Questions
7cot is an amazing community to study how people utilize, connect, share,
and engage with others in a “Social Web” emotional support system
An exemplar example of the paraprofessional model online
A “closed” system where people anonymously seek and receive
emotional support online
7cot is successful, and quickly growing:
From 12/5/13 – 11/18/14,
130k+ registered users (33k are volunteer listeners)
Over 1.27M one-on-one conversations shared
465K unique connections between members and listeners
10. Research Questions
We explore 7cot to understand how people use such emotional support
services, and how effective ones may be designed in the future:
What kinds of activities are explored by different sets of users?
How do members and listeners connect and interact?
What aspects of 7cot encourage engagement?
Disclosure: Slides offer aspects of analysis, key takeaways
Please see our paper for full details!
11. Member/Listener Interactions
Interactions among members and listeners structurally represented as a
bipartite network
Bipartite projections characterize members and listeners who connect with
common others
Members seek a large number
of listeners to speak toVirtually no pairs of members
and listeners are exclusive to
each other
Small path length: we have a core of members
and listeners that are hubs: people who con.
to a very large, very diverse set of others
12. Member/Listener Interactions
Listener ProjectionMember Projection
Further Evidence of Hubs
Insight: Users thrive on connecting to very diverse sets of others
- New people, new ideas, new opinions, new breakthroughs
p = 0.362; α = 2.34 p = 0.985;α = 2.51
H0: We have a power-tail
13. Interaction Network Structures
(10k edge samples shown to aid visualization)
Hotter nodes: larger clustering coefficients
Insight: groups of members tend to be
connecting with the same subsets of listeners
Members may be affiliated by common
ailment, or attraction to a listener’s profile
Contrast: Listener clustering coefficients are
skewed
A large clique: listeners tend to connect to
a diverse number of members
Hypothesis: a listener is a volunteer, willing to
help anyone, if the request comes through
Members are selective, listeners less so
14. Understanding User Engagement
Longstanding clinical problem: Getting people to see a psychologist, and
to return for follow on visits
How can we design online emotional support systems in a way that
encourages people to return?
We model engagement as the message rate of users: for the days they are
active, the average number of one-on-one conv. Messages sent
We consider two questions:
What behavioral attributes best predict strong engagement?
Can we predict if new users will become a long-term one? If so, what
are the most predictive behavioral features?
15. Behaviors that anticipate engagement
We consider a random forest model that predicts the engagement
(message rate) of a user using 15 behavioral features
Features reflect gamification, user interactions, and user characteristics
Regression model fits data very well, R2 > 89%
Pairwise feature correlations
16. Feature importance
For each feature of the fitted model,
perturb its values across the dataset
Measure the % change in MSE between
the real and perturbed model
Intuition: a feature whose values are
randomly distributed that has little impact
on model error is not impactful
Group chat encourages
one-on-one chat
Forum popularity discourages
one-on-one chat
Gamification matters!
17. Predicting long-term active
Active user: One registered > 2 weeks ago, >= 2 actions over last 2 months
52,803 members registered on 7cot during the time period considered, of
which 11,117 (21%) became active and 41,686 (79%) became inactive
Can we predict if a new user will become long-term active?
Yes: 92.5% classification accuracy; moderate FP rate
18. Conclusions
7cot: a large-scale online social network to connect those needing with
those offering support. An example “Social Web” emotional support service
Insights highlight the positive features and potential of “Social Web” online
emotional support platforms
Members prefer to access large volumes of others
Listeners willing to speak to anyone, no matter their concerns. Members
are selective, listeners less so.
groups of members tend to connect with the same subsets of listeners
Gamification mechanisms encourage positive return and reuse
(A major clinical psych concern)
“more public” modes of real time communication (Group chat)
encourages others to participate in one-on-one conversations
19. “Emotional Support Processes on the Social Web”
Youngest members of society may prefer to communicate online vs. face to face.
What are the implications when they need to seek others out for emotional support?
Are they more comfortable speaking online vs. off?
If so, how do we create effective spaces for them to get the support they need?
Online social support systems may well be the shape of things to come
Analysis of 7cot revealed design considerations, highlighted positive ways other connect
Just the beginning:
User outcomes in private social web vs. public social web
Preference of different “modes” given condition
User privacy and discover in emotional support platforms
Quality of user outcomes: proxies for measurement, and comparison with
alternative online support systems
20. Thanks very much!
Questions and Discussions
Derek Doran, Samir Yelne
Dept. of CSE, Wright State University, USA
Luisa Massari, Maria-Carla Calzarossa
Dept. of Electrical, Computer, and Biomed Eng., University of Pavia, Italy
LaTrelle Jackson
Dept. of Professional Psych, Wright State University, USA
Glen Moriarty
7 Cups of Tea, Inc.
http://knoesis.wright.edu/doran