Weitere ähnliche Inhalte
Ähnlich wie The Many Faces of Reputation: Towards a discipline of Web 2.0 reputation system design
Ähnlich wie The Many Faces of Reputation: Towards a discipline of Web 2.0 reputation system design (20)
Kürzlich hochgeladen (20)
The Many Faces of Reputation: Towards a discipline of Web 2.0 reputation system design
- 1. The Many Faces of Reputation:
Towards a discipline of Web 2.0
reputation system design
Prof. Chris Dellarocas
Robert H. Smith School of Business
University of Maryland
cdell@umd.edu
Copyright © C. Dellarocas 2008-09
Suddenly everything Web 2.0 became so sexy…
Copyright © C. Dellarocas 2008-09
1
- 2. Except, perhaps, reputation systems…
RepuNomics
How Online Reputation
Enables Markets
Sustains Communities
This
and Creates
New Business
book
Opportunities
does not
exist
Copyright © C. Dellarocas 2008-09
But reputation systems are everywhere
eCommerce
eBay
Amazon
Epinions
Web 2.0
Yelp
Slashdot
Crowdsourcing
Yahoo! Answers
iStockPhoto
Gaming
Xbox Live
Copyright © C. Dellarocas 2008-09
2
- 3. Reputation Systems
are the Unsung Heroes
of the Web
Copyright © C. Dellarocas 2008-09
Outline of this talk
Reputation Systems Serve Business Objectives
Four areas where practice runs ahead of
research
Our challenge: Developing a rigorous discipline
of real-life reputation system design
Copyright © C. Dellarocas 2008-09
3
- 4. Some Definitions
Reputation = a summary of one’s past actions
defined within the context of a specific community
Reputation system = a system that mediates
and automates the process of assessing one’s
reputation
Keeps track of a user’s actions
Aggregates and displays summary statistics
Copyright © C. Dellarocas 2008-09
Bird’s eye view of a reputation system
(Inter)
actions Inputs
Aggregation
Outputs
Users Artifacts
COMMUNITY
Copyright © C. Dellarocas 2008-09
4
- 5. Reputation Systems
serve a variety of
Business Objectives
Copyright © C. Dellarocas 2008-09
Why do communities need reputation
systems?
Trust
Encourage “good” and discourage “bad” behavior
Quality
Provide incentives for quality contributions
Recognize best contributors
Matching and Filtering
Assist users in finding suitable partners
Reduce information overload
Participation and loyalty
Give user reasons to join and stay in your community
Copyright © C. Dellarocas 2008-09
5
- 6. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
6
- 7. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
7
- 8. Copyright © C. Dellarocas 2008-09
Four examples
Trust Quality Filtering Loyalty
eBay +++ + + ++
Amazon ++ +++ + ++
Yelp ++ +++ +++ +++
Yahoo!Answers ++ ++ + +++
Copyright © C. Dellarocas 2008-09
8
- 9. The design space of
Reputation Systems
is very rich
Copyright © C. Dellarocas 2008-09
Four dimensions of reputation system design
where practice runs ahead of research
What inputs should be solicited
1.
What outputs should be presented
2.
How transparent should the rules be
3.
How should reputation evolve over time
4.
Copyright © C. Dellarocas 2008-09
9
- 10. What inputs should be solicited?
What internal actions to keep track of
What external feedback to solicit
Ratings of artifacts vs. ratings of people
Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
10
- 11. Determining inputs
Perform an audit of your system
List all actions a user can take
Assess how each action relates to each of the
four classes of objectives
Track/report actions that:
you want to encourage/discourage
provide the most information related to one or more of
your key objectives
Copyright © C. Dellarocas 2008-09
A story of a site that got it wrong:
Consumating.com
Copyright © C. Dellarocas 2008-09
11
- 12. An interesting question:
Should you allow ratings of users….
… separately from the artifacts they create?
Pluses:
Makes a person feel like a person… which might
increase site loyalty
Minuses:
Can distract from quality of one’s work
Might encourage personal attacks
Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
12
- 13. Four dimensions of reputation system design
where practice runs ahead of research
What inputs should be solicited
1.
How outputs should be presented
2.
How transparent should the rules be
3.
How should reputation evolve over time
4.
Copyright © C. Dellarocas 2008-09
What outputs should be presented?
Simple statistics
Star ratings
Score
Named levels
Achievement badges
Social network
Copyright © C. Dellarocas 2008-09
13
- 14. Copyright © C. Dellarocas 2008-09
Simple Statistics
Copyright © C. Dellarocas 2008-09
14
- 15. Statistics
Score
Copyright © C. Dellarocas 2008-09
Social
Network
Copyright © C. Dellarocas 2008-09
15
- 16. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
16
- 17. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
17
- 18. Copyright © C. Dellarocas 2008-09
To what extent should the system allow
comparison across users?
Display of relative performance statistics
Top N lists
“Leaderboards”
Copyright © C. Dellarocas 2008-09
18
- 19. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
19
- 20. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
20
- 21. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
21
- 22. To compare or not to compare?
Pluses
Increases incentives to contribute
Enhances filtering role of reputation
Minuses
Instills a culture of competition
Obsession with rankings might lead to manipulative,
disruptive behavior
Low ranked users might be discouraged and exit
Copyright © C. Dellarocas 2008-09
A cautionary tale
Digg.com
Copyright © C. Dellarocas 2008-09
22
- 23. How you display reputation information
affects the spirit of your community
COMPETITIVE
CORDIAL
Named
Ranking
Statistical Levels Point-based score
Leaderboards
evidence Badges
Reputation system design must be consistent
with the community’s overall culture
Copyright © C. Dellarocas 2008-09
Some ideas
Implement multiple ways of ranking users
No single measure that people obsess on
Allows users with different qualities to feel good
Particularly relevant in systems where matching and
filtering is an important objective
Copyright © C. Dellarocas 2008-09
23
- 24. Copyright © C. Dellarocas 2008-09
Four dimensions of reputation system design
where practice runs ahead of research
What inputs should be solicited
1.
What outputs should be presented
2.
How transparent should the rules be
3.
How should reputation evolve over time
4.
Copyright © C. Dellarocas 2008-09
24
- 25. How transparent should be the aggregation
rules?
Transparency breeds trust
… but also facilitates gaming
Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
25
- 26. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
26
- 27. Copyright © C. Dellarocas 2008-09
Four dimensions of reputation system design
where practice runs ahead of research
What inputs should be solicited
1.
What outputs should be presented
2.
How transparent should the rules be
3.
How should reputation evolve over time
4.
Copyright © C. Dellarocas 2008-09
27
- 28. How should reputation evolve over time?
Accumulate and grow
Decay and refresh
Tradeoffs
Accumulation breeds complacency
Accumulation discourages new entrants
BUT
Accumulation promotes loyalty and lock-in
Copyright © C. Dellarocas 2008-09
Accumulation
Decay and Refresh Copyright © C. Dellarocas 2008-09
28
- 29. A case study:
Yelp
Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
29
- 30. Copyright © C. Dellarocas 2008-09
Copyright © C. Dellarocas 2008-09
30
- 31. Copyright © C. Dellarocas 2008-09
The challenge:
Mapping Business Objectives
to Design
Copyright © C. Dellarocas 2008-09
31
- 32. Observation #1: Everything should be driven
from a clear understanding of objectives
Trust building/Community Policing
Incentives for Quality
Matching and Filtering
Site Loyalty
It is important for the designer to be very clear with
respect to how these goals are prioritized
It is important to understand how every aspect of design
affects each of these goals
Copyright © C. Dellarocas 2008-09
Observation #2: The design of a reputation
system can affect a community’s culture
COMPETITIVE
CORDIAL
Named
Ranking
Statistical Levels Point-based score
Leaderboards
evidence Badges
Reputation system design must be consistent
with the community’s overall culture
Copyright © C. Dellarocas 2008-09
32
- 33. Observation #3: Intangible aspects of
reputation matter
Users care about their reputation beyond the
tangible benefits it confers to them
Reputation as pure status
Status is zero-sum
Users who are not recognized might get upset
So, introduction of a reputation system might make
some users worse off and cause them to leave
Copyright © C. Dellarocas 2008-09
Observation #4: A reputation system can be
a site’s competitive weapon
A way to lock-in users to our site
Lock-in properties of reputation system can
sometimes be in conflict with its other objectives
Understanding the design of competitors’
reputation mechanisms matters
Copyright © C. Dellarocas 2008-09
33
- 34. In summary
Much of what we know about reputation system
design in Web 2.0 communities is still anecdotal
We need research-driven guidelines on how to
design such systems
A lot of open areas where theoretical and
experimental work needs to be done
Copyright © C. Dellarocas 2008-09
In the meantime…
Reputation systems designers will be the poor
lonesome cowboys of the Web
Not necessarily such a bad place to be (hey, it can
get you to places like Gargonza)
Copyright © C. Dellarocas 2008-09
34