The document discusses using game theory to model user behavior in social media networks. It presents a model where users decide whether to voluntarily contribute to tasks based on the individual costs and common gains. The model finds that as more users participate, the probability of task completion decreases due to free riding. However, the document suggests system designers can address this by offering virtual rewards to users for contributions, increasing both individual benefits and overall task completion rates.
Understanding User Behavior in Social Media Using Game Theory
1. Vivek Singh1, Ramesh Jain1 & Mohan Kankanhalli2
1University of California, Irvine
2National University of Singapore
2. Web 2.0
◦ Flickr
◦ Youtube
◦ Wikipedia
User contributions for tasks
◦ iReporter on CNN
◦ NewYorkTimes citizen journalism
◦ Twitter
◦ Google image labeler
◦ Wikipedia
3. Multiple advancements made in terms of:
◦ Data sharing
◦ Image tagging
◦ Mashups
◦ Ontologies
◦ Retrieval, …
How about the User?
Media
4. We need tools for modeling user behavior
Theoretical grounding into understanding
user motivation.
Create multimodal applications that
understand
◦ crowd sourcing,
◦ human computation,
◦ collective intelligence…
5.
6. User contributions for tasks
◦ iReporter on CNN, NewYorkTimes citizen
journalism, Twitter, Google image
labeler, Wikipedia, …
Distinct features
◦ Common gain
◦ Individual cost
◦ Contribution is totally voluntary
◦ No central coordination possible
Individual utility and motivation are
extremely important.
7. User perspective:
Optimal contribution strategy i.e. “when (and
when not) should she undertake the social
media task”
System designer perspective:
“Finding the optimal incentive levels to
influence these selfish end-users so that the
overall system utility is maximized”
8. Case Participation Taxation Reward Career Usage
currency benefits pattern
Physical societies Mandatory Enforceable Real - -
Open source software Voluntary - - Yes Partially
and scientific regulated
communities
P2P/ networking Voluntary - Virtual - Monitored &
regulated
Social media Voluntary Not Virtual Rare Public-good
enforceable
9. Human computing [vonAhn06]
User behavior study on Youtube [Maia08]
Sociological analysis[Schroer09][Sun06]
◦ What types of feature do users like?
Our focus:
◦ Motivation
◦ Quantitative (not qualitative)
10. Game
◦ Zero-Sum
◦ Non-zero-sum
Nash equilibrium
◦ A point where each agent knows other agent's
strategy options, and from which no agent has any
gain associated with changing their own strategy
unilaterally.
Mixed Strategy Nash Equilibrium
11. Problem motivation
◦ Citizen journalism task:
◦ Suspicious bag reporting
Parameters
◦ N agents walk past
◦ Cost to report is individual (Ci)
◦ Gain is common (G)
Problem: As an individual user, how often
should I report it myself (Pi) ?
12. Other agent
Agent
Do Dont
Do G-c, G- G-c, G
c
Dont G, G-c 0, 0
Solution concept: Nash equilibrium
Occurs at choice indifference:
13. Other agent
Agent
Do Dont
Do G-c1, G-c1, G
G-c2
Dont G, G-c2 0, 0
Solution concept: Nash equilibrium
Occurs at choice indifference:
14. Other agents
Agent
Do Dont
Do G-ci G-ci
Dont G 0
Solution concept: Nash equilibrium
Occurs at choice indifference:
16. Unattended bag reporting scenario
◦ G=100, c1=50, c2=60, c3=70
Probabilities of undertaking task:
◦ P1=0.084, P2=0.346, P3=0.445
17. Results
◦ EU1 =50, EU2 =40, EU3 =30
◦ Same as that obtainable by doing the task all-by-
themselves
◦ More than that by changing your strategy unilaterally
18. Varied C, G, and N
◦ Larger N-> Utility maintained, But
◦ Overall Probability(Task completion) decreases exponentially.
Larger N Lower task completion
Alarming result !
o From a system
designer‟s
perspective
o Free rider-problem !
19. Insights so far:
◦ Large „N‟ is bad for task completion
◦ The users are selfish
Can I exploit user selfishness to maximize
system gains
◦ Think loyalty coupons, Reward points, Frequent
flyer miles…
20. Designers perspective
◦ Gain from each task completion (Gs)
◦ Extra benefit (b) to all those who complete the task
Maximize system designer‟s utility
In social media systems, most rewards are virtual
Additional bandwidth, enhanced weaponry, titles/badges,
Farmville elephant…
Perceived benefit by user >> cost to system designer
, for now lets consider
21. Based on discussion in previous section
Maxima can be obtained by standard calculus
or numerical methods
22. Large set of homogenous users which can
report the bag
◦ c= 60, G=100, Gs = 50, N=1000
◦ b ϵ [0,60] e.g. chances at winning an iPhone, mobile phone
charge, reputation points, Game level unlocks, …
Results
◦ EUs maximizes at
b=42
◦ EUs = 29.87
◦ Eus, NoBonus=20.0
23. Results
◦ Probability (task completion) = 0.82
◦ Probability (task completion)NoBonus = 0.4
◦ Eui = 82
◦ Eui,NoBonus = 40
Higher
◦ Individual gains
◦ System gain
◦ Task completion
24. Modeling Humans
◦ „Bounded rationality‟ [Kahneman, Rubinstein]
◦ „Predictably irrational‟ [Areily]
Explicit Quantification
◦ Second life – Yes, but difficult in general
No numerical data
◦ Nascent field
◦ But lets work towards this
Virtual costs to designers
◦ True in most social media applications
25. K contributions required (rather than just 1)
◦ 2 images for stereo, 5 spam flags, …
◦ Graded utility of each contribution
„J‟ tasks to be completed by „N‟ users, each
requiring minimum „K‟ contributions
Patterns in very large numbers
Better modeling required…
◦ But more importantly:
◦ Food-for-thought for social media designers and
developers charged with creating crowd-
sourcing, media applications
26. Game theoretic framework to study user
behavior in social media networks
◦ Selfish agents
◦ Domain semantics (voluntary participations, virtual
reward)
Optimal user contribution strategy
◦ Probability (task completion) decreases with large N
◦ Free rider problem made explicit
System designer can exploit user selfishness
by virtual reward structure
◦ Improve system performance
◦ Potentially still benefiting individual users
Editor's Notes
Highlight that Little or no focus has been given on understanding the user or the social aspect of social media?
Before this slide I will showshort anime video about ants carrying sugar cubes(0:30 to 1:15) which shows 2 things:1. Ants when present in large numbers can do amazing things.2. Given the correct sugar candy you can attract large number of ants. Then will try to pose the question: Do these things hold true for humans?