IR APPLICATION: Intelligent Agents

1
IR APPLICATIONIntelligent Agents

• (????????)

2

Intelligent Agent

• Simply put, agents can be considered personal
  software assistants with authority delegated from
  their users.
• Agents are computer programs that simulate a
  human relationship, by doing something that
  another person could otherwise do for you.an
  agent is a software thing that knows how to do
  things that you could probably do yourself if you
  had the time--
• Ted Selker from IBMs Almaden Research Center

3

Essential features of Agents

• Autonomous
• Personalized (user-centered)
• Task/goal-oriented
• Communicative
• Collaborative
• Reactive to environmental changes
• Evolutionary/adaptive

4
Autonomous
The search agent must be able to operate without
interaction with a human agent. It must control
over its own internal states and make independent
decisions. This implies a search capabilities to
traverse information sites based upon
pre-established criteria collecting potentially
relevant information.
5
Communicative - communications ability
The search agent must be able to communicate with
the information sites as it traverses them. This
implies a universally accepted language defining
the external interfaces (e.g., Z39.50).
6
Collaborative - capacity for cooperation
This concept suggests that intelligent agents
need to cooperate to perform mutually beneficial
tasks.
7
Capacity for Reasoning

• There are three types of reasoning scenarios
  Roseler 94
• Rule-basedwhere user has defined a set of
  conditions and actions to be taken
• Knowledge-based where intelligent agents have
  stored previous conditions and actions taken
  which are used to deduce future actions
• Artificial evolution based where intelligent
  agents spawn new agents with higher logic
  capability to perform its objectives.

8
Adaptive Behavior
Adaptive behavior permits the intelligent agent
to access its current state and make decisions on
the actions it should take.
9
Trustworthiness
The user must trust that the intelligent agent
will act on the users behalf to locate
information that the user has access to and is
relevant to the user.
10
Implementations of Intelligent Agents
There are many implementation aspects of
Intelligent agents. They include communications
to traverse the Internet, how to wrap the agent
in an appropriate interface shell to work within
an Internet server, and security and protection
for both the agent and the servers. However, the
major focus for information retrieval is how to
optimize the location of relevant documents as
the agent performs it tasks. This requires
expansion of search capabilities into conditional
and learning feedback mechanisms (i.e. automatic
relevance feedback).
11
Automatic Relevance Feedback
As an intelligent agent moves from site to site,
it is necessary for it to use similar techniques
to learn the language of the authors and
correlate it to the search need of the user. Some
basic groundwork is being laid by the work on
incremental relevance feedback. It will also need
capabilities to normalize ranking values across
multiple systems. The quality of possible
information being returned necessitates a merged
ranking to allow the user to focus on the most
likely relevant documents first.
12

Agent???

• Internet Agent
• Web search agent
• Information filtering agent
• Information retrieval agent, e.g. PointCast
  Network
• Mobile agent, etc
• Intranet Agent
• Process automation agent
• database agent, etc
• Desktop Agent
• Application agent
• Email agent
• Search agent, e.g. Web Compass

13
The problem of search engines
771654 Web pages
Information overload!!
14
An Agent Model
Agents send the requests on behalf of the user
User sends the request
User
Agents
Information Sources
Information sources respond to the requests
Agents supply the responses to user
The Three Layer Model
15
An Agent Model
User
Agents
Information Sources
The main concept inside the three layer model is
that the agents are responsible to koowthe
users preference, and make the users
information need can be satisfied based on the
users preference. Besides,the agent should
work as intermediaries without introducing any
interference to the normal activities of the
user.
16

Agent??????-??? (personalization)??

• ???(personalization)???????????????????,??????????
  ???,???????????,??
• ?Smart News Reader?? ?NewsHound?,?????????????????
  ??????

17

Personal profile techniques

• ???personal profile techniques????Agent
  applications?
• Users persistent information interests
• by means of enabling users to specify information
  profiles using a form type interface
• Users dynamic information interests
• current users interests can be tracked by
  monitoring the topic of a followed link, the
  bookmarking of a link, or the time spent on a
  particular Web page

18
The Agent Community
Task Agent
Group Agent
Personal Agent
Personal Agent
Personal Agent
User A
User B
User C
19
The Agent function capturing users preference
???????,????????(personal Agent),???????????,????
? ?????????????????????, ???Agent?????,????????? (
learning)???????????????? ?????????(information
retrieval)?? ?????????????????,??? ?????????
20
Case study Smart News Reader
Smart News Reader is an experimental
software application developed by Intel. It
claims that it uses an Information Evaluation
technology to read articles. It scores each
article based on the user interests. After click
the preferences on as few as 50 articles, the
system can sort articles by the user identified
preferences and put those articles at the top of
the list. The agent can therefore according to
the user feedback obtained to adjust its
behavior for offering users preferred articles.