Transcript agents98

Intelligent Agents
Conference Notes
ISE Workshop at NASA
16/17-SEP-1998
Topics
1. Information Retrieval
– Why Surf Alone? (Lieberman, MIT)
– Knowledge Representation (Hendler, UMD)
2. Agent Interaction
– Conversational Agents (Finin, UMD)
– Conscious Agents (Franklin, U.Memphis)
3. Organization
– Coordinating Intelligent Agents (Decker, UDE)
– Agent Based Computing and the Open Agent
Architecture (Cheyer & Moran, SRI)
4. Practical Engineering Applications
– Role of Intelligent Agents in Advanced
Information Systems (Kerschberg, GMU)
– Multiagent Systems, WWW, and Networked
Scientific Computing (Joshi, UMD)
5. Design Using Agents
– Improving Design w/ Agents (Brown, Worcester Tech)
– Decentralized Decision Making in Concurrent
Engineering (Birmingham, UMI)
– Standards Activity in Agent Based Learning and
Visual Consultation for Manufacturing (Aparicia,
IBM)
6. Multiagent Programming & Future
– Multiagent Oriented Programming (Huhns, USC)
– Future Directions (general discussion)
Topic 1.1
Why Surf Alone?
Reconnaisance Agents
Henry Lieberman, MIT
Two Modes of Info Retrieval
• Browsing
– exploring, can follow tangents, vague ideas
– fun but easy to get lost
– “chunky”- stick w/ clump of topic, then move on to
something else (e.g., paris, music)
• Searching
– precisely targetted retrieval
– computers can search pre-indexed docs quickly
– what if you don't want to search for exactly something
for which you know a keyword?
Why not something in between?
• People have Interests
– sometimes expressed as keywords, sometimes not
– can recognize information in which they have an
interest [sometimes better than a computer]
• IR should involve cooperation between an
intelligent agent and the user
– the user is better at deciding what has value to him
– the computer is better at searching prexisting indexes
• Evaluation Function
– provides a mapping between the user’s idea of merit
and computer’s idea of merit
– an agent can learn an evaluation function by observing
the user’s browsing behavior
Depth vs. Breadth First Searching
• Depth First
– traditional search engines used by Netscape, etc.
– presents options and the user picks one path to follow
– not optimal since the engine returns one batch of results
per query; there are no more chances
• Breadth First
– searching in parallel
– an agent can perform this type of search
– every time the user goes from one page to another, the
agent can refocus its breadth first search
Analyzable Qualities of
Users’ Browsing Behavior
• Every click is an expression of interest
– content [keywords] of documents browsed
– sequence of browsing operations
• Other expressions of interest
– dwell time
– downloads
– email
– bookmarks, links to page
• Agents would help in:
– maintaining areas of interest
– finding areas of dual interests (e.g. french jazz concert)
Let’s Browse
• Multiple users browsing together, agent:
– takes into account interests of passive participants
– can represent interests for each user (from home pages)
– dynamically performs freq analysis to to discover areas
of common (intersecting) interest
– “hey, I didn't know you played tennis” - icebreaker
– interesting result - had to keep more terms per user than
in single user case (had to dip lower into [minor]
personal interests in order to find common interests)
LETIZIA
• What is it?
– Lieberman’s example of a reconnaisance agent
– an “advance scout for web browsing”
– a “channel surfing interface”
• What does it do?
– provides temporal sampling of different info sources
– observes user’s browsing behavior
– infers user’s interests from its observations
LETIZIA, notes...
– works w/ interactive interface (unmodified netscape),
doesn't replace
– doesn't take any extra effort to communicate to agent
– anticipates user's interests
– explores links from home page to see if anything
interesting
– agent looks through links, determines which are most
interesting
– agent also looks at results from search engine
– also uses netscape history to keep track