User Modelling and Groups

1
User Modelling and Groups

• Where?
• CSCW
• CSCL
• Collaborative filtering
• P2P, ubiquitous computing
• What?
• Explicit group model
• Interacting individual models
• Interacting UM fragments
• Why?
• Improve individual adaptation
• Facilitate collaboration
• Facilitate negotiation
• Motivate participation
• Models represent users (in MAS)

• How?
• Capturing parameters of group activity
• Reconsiliating individ. models with group
  model
• No group model interaction of individ.
  models

Winter, Gaudioso, GoodmanTang, Zhu
Jameson, Lock, Winter, Masthoff
Gaudioso, Goodman
Tang, Zhu
Sun, Kaminka
Jameson, Lock, Masthoff
Jameson, Winter, Tang Goodman, Gaudioso, Masthoff
Jameson, Gaudioso, Masthof, Sun,
Lock, Kaminka
Sun, Kaminka,
Kaminka, Tang McCalla, Zhu
Goodman, Gaudioso, Lock
Jameson, Masthoff
Sun
Jameson
2
My vision

• Loose groups and virtual communities
• Why?
• History

3
Ensuring Quality of Service in P2P File
Sharingthrough User and Relationship Modelling

• Lingling Sun, Yamini Upadrashta,
• Julita Vassileva
• MADMUC Lab, Computer Science Department,
  University of Saskatchewan

4
Loose Groups and Virtual Communities

• Ubiquitous P2P applications, newsgroups,
  chatrooms, bloggers
• Users are often anonymous
• No explicit common goal, no assigned roles goals
  and roles emerge
• Motivation to participate and contribute is
  crucial

5
COMTELLA
Helen Bretzke CRA-W and NSERC Summer 2002
project

• A P2P (Gnutella based) system for file sharing
  and service
• users share academic papers, code snippets, help
• non-centralized digital library for a research
  group / class
• motivation and community-building

Christopher Cox NSERC Summer 2002 project
Lingling Sun undergraduate project
Yamini Upadrashta Graduate student
6
Our approach

• Modelling relationships between pairs of users
  based on positive experience
• Every peer keeps a list of friends in each area
  of interest
• If a peer appears in many friends lists, s/he
  is important for this community

7
Relationships

• Not necessarily symmetrical!
• Reinforcement learning based on history of
  interactions between two peers
• Each relationship has strength, balance.

8
Adaptation

• Rewarding peers who are friends of many other
  peers with better QoS
• Transfers not interrupted, higher badwidht,
  priority in queues
• Having many good friends pays
• Faster search, more relevant documents

9
Evaluation results - simulation

• Figure 1 Graph comparing the round trip time
  obtained for queries without a
  friends list with the round trip time for
  queries with friends list

10
Evaluation results user experiment
11
Conclusion

• The simulation results show that peers obtain
  results faster when searching for files in
  categories for which they have friends
• The user evaluation still underway