Docstoc

A Multi-Agent Referral System for Matchmaking

Document Sample
A Multi-Agent Referral System for Matchmaking Powered By Docstoc
					    A Multi-Agent Referral
   System for Matchmaking
                                             Leonard N. Foner
                                        MIT media laboratory
           The First International Conference on the Practical
Applications of Intelligent Agents and Multi-Agent Technology,
                                        London, UK, April '96.
  Hidehiko MASAKI
  TITECH, Higa Laboratory
  29th Jun 1999, M1 Seminar
       Introduction
Webhound, Ringo

     ITEMS        USERS’ TASTES



YENTA-LITE
     USERS        USERS
         Multi-Agents
Advantages

     Scaling, Fault Tolerance, etc

Disadvantage
     How are agents supposed to find
     each other?
   Applications
•Messaging into the group

•Introduction

•Finding an EXPERT
THE OVERALL GOAL

forming clusters of agents
whose users share similar
interests
                 Interests
Agent #1                             Agent #2


    grain   grain        grain    grain
   grain    grain         grain
                                  grain
            grain         grain


                              users’ interests are
                              capturable in some
       granule      cluster
                              computer based form
      Similarity
•Yenta-Lite at the moment
 scalar
•SMART
 vector
•WordNet
 a semantic net of words
•Future Implementations of Yenta
 Hybrid, SMART and WordNet
            How to Cluster
1)Intra-agent initialization, known as preclustering:
Combine grains into granules within a single agent


2)Inter-agent initialization, known as bootstrapping:
Find at least one agent with which to communicate


3)Walk referrals and cluster:
Form clusters of like-minded agents
              Preclustering
         email, news, and so forth
grain

          rating                    grain
                          grain
   SMART                             grain
                            grain
        clustering
                                     grain grain
                                    grain
 Agents determine
 users’ interests!
         Bootstrapping
Finding at least one other agent
with which to communicate




                       FIND!
                    (heuristics)
          Data structures
• A cluster cache
contains the name of all agents in same cluster

• A rumor cache
contains the name and other information from
last r agents

• A pending-contact list
a priority-ordered list of other agents that have
been discovered but never contacted
         Conclusions
            & Future Work
-conclusions-
   Yenta-Lite demonstrates that referral-based
   matchmaking can provide acceptable results
   without requiring any one agent to know about
   all other agents.
-future work-
  privacy safeguard, user interface,
  evaluating the algorithms in real environment,
  experimenting with different comparison metrics
      My Comments
• 多分、手法としてはめずらしくはないだろうけど、
  参考になった。
• IT-based Seminarでの設計に利用できる。

• 個人的に
ここにアプライしてみる。

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:2
posted:11/2/2011
language:English
pages:13