Illustrating Digital Traces: Visualizations of patterns generated by computermediated collective act

1
Illustrating Digital Traces Visualizations of
patterns generated by computer-mediated
collective action systems

• Marc A. Smith
• Microsoft Research
• Community Technologies Group

2
An overview of Microsoft Research, Redmond, WA
3
Email (and more) is from people to people
4
Patterns are left behind
5
Whyte, William H. 1971. City Rediscovering the
Center. New York Anchor Books.
6
Email, Email lists, Chat, Buddy Lists/Instant
Messenger, Usenet, Web Boards, Forums, Ebay,
Blogs, Wikis, MUDs, MOOs, Graphical Worlds,
MMORPGs, Napster, Kazaa, Gnutella,
BittorrentFolksonomies, MoSoSo
Social cyberspaces are created by many kinds of
network interaction media
7
Youse.
Yall.
Yes, youse.
8
Computer-MediatedCollective Action

• Lots of collective action happening in
  computational media
• Newsgoups, wikis, tagging systems, massively
  multiplayer online role playing games, geospatial
  annotation systems, and more!
• Information foraging and organizing behaviors are
  rampant
• Formerly a professional activity, now widely
  amateurized and global

9
Netscan A tool for studying threaded
conversation repositories
Newsgroups
Newsgroups

• Data mining Usenet and other conversation
  repositories

Authors
Threads
Authors
Threads
The Message
10
Internet Scale THREAD -o- SPHERE
Groups.Google
Enterprise THREAD -o- SPHERES
Usenet
Microsoft QA
Technical
Social
Personal THREAD -o- SPHERES
Yahoo Groups
TechNet
MSDN
Medical
MSNBC
Political
Entertainment Communities
MSN Forums
Blog -o- Sphere
Web Boards
Recreation Communities
Financial Communities
XBOX
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
Mapping Newsgroup Social Ties
Microsoft.public.windowsxp.server.general
Two answer people with an emerging 3rd.
16
Distinguishing attributes

• Answer person
• Outward ties to local isolates
• Relative absence of triangles
• Few intense ties
• Discussion person
• Ties from local isolates often inward only
• Dense, many triangles
• Numerous intense ties

17
Distinguishing attributes

• Answer person
• Outward ties to local isolates
• Relative absence of triangles
• Few intense ties
• Reply Magnet
• Ties from local isolates often inward only
• Sparse, few triangles
• Few intense ties

18
From Page Rank to People Rank

• People Rank is critical component of an effective
  community strategy.
•  
• Communities are composed of a relatively small
  set of roles.
•  
• Technology to identify these roles is critical
  for selecting high quality content in a vast and
  diverse sea of material.
•  
• Social Accounting Metadata is the raw material of
  social sorting, a people rank that brings high
  quality content to the surface of an online
  community.
•  
• Reputations and profile are central to the
  effective management of a community.
•  

19
User Roles in Online Communities
20
User Roles in Online Communities
21
Netscan http//netscan.research.microsoft.com
22
Do you know who I am?
Taken from Connections by Sara Keisler
23
Explicit vs. implicit reputation systems

• Explicit
• Statements about behaviors and relationships
• eBay
• Amazon
• Slashdot
• Digg
• MySpace
• Facebook
• YouTube
• flickr
• Issues
• Provisioning not enough rating
• Latency ratings not fast enough
• Bias susceptible to initial reactions
• Collusion easily shilled
• Inflation disincentives to accuracy

• Implicit
• Observations about behaviors and relationships
• Google
• Amazon
• Flickr, MySpace, Facebook
• del.icio.us
• Technorati
• Netscan
• Issues
• Ambiguity Behavior is not endorsement
• Collusion Subject to manipulation
• May be subject to herding or positive-feedback
  loops

24
(No Transcript)
25
(No Transcript)
26
Visit Netscan
Netscan is on the Web at http//netscan.research.
microsoft.com Academic Data Sample Licenses
Available "Visualizing the Signatures of Social
Roles in Online Discussion GroupsThe Journal of
Social Structure.  8(2) Picturing Usenet The
Journal of Computer Mediated Communication You
are who you talk to HICSS 2007
27

• Hardin, Garrett. 1968/1977. The tragedy of the
  commons. Science 162 1243-48. Pp. 16-30 in
  Managing the Commons, edited by G. Hardin and J.
  Baden. San Francisco Freeman.
• Wellman, Barry. 1997. An electronic group is
  virtually a social network. In S. Kiesler (Ed.),
  The Culture of the Internet. Hillsdale, NJ
  Lawrence Erlbaum.

28
Related WorkData Mining Community Data

• Cross-postings information can be used to
  identify related communities and associate
  newsgroups
• C. Borgs, J. Chayes, M. Mahdian and A. Saberi,
  Exploring the Community Structure of
  Newsgroups, KDD04, Seattle, Washington, USA,
  August 2004.

• Thread structure and participants activity can be
  used to improve search and retrieval of newsgroup
  messages
• W. Xi, J. Lind and E. Brill, Learning Effective
  Ranking Functions for Newsgroup Search,
  SIGIR04, Sheffield, UK, July 2004.

• Patterns of activity can tell a lot about the
  type of participants
• A.J. Brush, X. Wang, T.C. Turner and M. Smith,
  Assessing Differential Usage of Usenet Social
  Accounting Meta-Data, CHI05, Portland, Oregon,
  USA, April 2005.

29
Related WorkIdentifying Discussion Topics

• Social text mining - consider the social network
  of online communities to drive the topic analysis
• R. Agrawal, S. Rajagopalan, R. Srikant and Y.
  Xu, Mining Newsgroups Using Networks Arising
  From Social Behavior, WWW03, Budapest, Hungary,
  May 2003
• Social network was used to classify newsgroup
  authors who are for and against a given
  discussion topic
• Assumes that replies are mostly to express
  disagreement and social roles are not considered
• Temporal text mining - identify temporal topic
  patterns from the full-text of messages
• Q. Mei and X. Zhai, Discovering Evolutionary
  Theme Patterns from Text - An Exploration of
  Temporal Text Mining, KDD05, Chicago, Illinois,
  August 2005

30
(No Transcript)
31
The Ties that Blind?
32
The Ties that Blind?
Reply-To Network Network at distance 2 for the
most prolific author of the microsoft.public.inter
netexplorer.general newsgroup
33
Darwin Bell
34
The Ties that Blind?
Pajek without modification can sometimes reveal
structures of great interest.
35
(No Transcript)
36
Social Network Theory / Social Network Systems

• Most CMC systems are SNS
• replies create ties
• SNS are novel in that they make ties durable and
  more visible
• Ties are authored by more granular actions (not
  just reply), i.e. making an item a favorite or
  simply viewing an object
• Some ties remain too coarse
• Are you my friend Yes/No?

37
SNA Resources
38
SNA 101

• Node
• actor on which relationships act 1-mode versus
  2-mode networks
• Edge
• Relationship connecting nodes can be directional
• Cohesive Sub-Group
• Well-connected group clique cluster
• Key Metrics
• Centrality (group or individual measure)
• Number of direct connections that individuals
  have with others in the group (usually look at
  incoming connections only)
• Measure at the individual node or group level
• Cohesion (group measure)
• Ease with which a network can connect
• Aggregate measure of shortest path between each
  node pair at network level reflects average
  distance
• Density (group measure)
• Robustness of the network
• Number of connections that exist in the group out
  of 100 possible
• Betweenness (individual measure)
• shortest paths between each node pair that a
  node is on
• Measure at the individual node level

A
B
C
A
B
D
E
D
E
G
F
C
D
H
I
E
39
AnswerPersonSignatures
Discussion People
40
Discussion Starter
Spammer
Reply orientedDiscussion
Flame Warrior
41
Clear and consistent signatures

• Light touch to numerous threads initiated by
  someone else
• Most ties are outward to local isolates
• Many more ties to small fish than big fish

42
Roles Project

• Using Netscan data to derive social roles in
  Usenet
• Next steps quantify explore in more depth

Answer Person, microsoft.public.windows.server.gen
eral
Discussion, rec.kites
Flame, alt.flame
Social Support, alt.support.divorce
PUBLISHED in HICSS, JCMC, JoSS, IEEE Internet
Communications (special issue on Social Networks)
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)
47
A Social Media ThreadMillFrom Data Source To
Results
48
WikiPediaEdits HighlightsAmountDistributio
n Timing ContentSpecificity
49
(No Transcript)
50
SNARF Email AssistantSocial Network and
Relationship Finder

• Side bar for peripheral awareness
• Multiple Panes
• List of people with associated messages
• Sorted by relationship
• Yellow bars, for visual indication of scale
• Double-click on person to see their messages

Danyel Fisher, AJ Brush, Andy Jacobs, Marc Smith,
Adam Perer, Bernie Hogan
http//www.research.microsoft.com/community/snarf
51
(No Transcript)
52
(No Transcript)
53
(No Transcript)
54
(No Transcript)
55
A place apart
A part of every place
56
(No Transcript)
57
(No Transcript)
58
Trace Encounters http//www.traceencounters.org/
59
Scott Counts, Marc Smith, AJ Brush, Paul Johns,
Aaron Hoff
60
(No Transcript)
61
(No Transcript)
62
Slam Group-based communication
Slam location map
Slam UI
Privacy settings
Scott Counts, Jordan Schwartz, Shelly Farnham
63
Prototype sensor board that includes GPS,
accelerometer, pressure sensor, temperature,
Bluetooth, and battery.
64
SLAM XR
Scott Counts, Marc Smith, Jianfeng Zhang, Nuria
Oliver, Andy Jacobs
65
(No Transcript)