Jo

1
Using an Agent-Based Framework and Separation of
Concerns for the Generation of Document
Classification Tools
2
The Problem

• Document classification
• Given a set of documents and a set of classes,
  decide which documents best match each class.

3
Questions

• Which format?
• Where to look for the documents?
• Which parameters are available?
• Is the document a copy?
• What is the output format?
• Which algorithm should be used?

4
Separating the concerns (1)

• Platform concerns
• Which format?
• Where to look for the documents?
• Which parameters are available?
• Is the document a copy?
• What is the output format?

5
Separating the concerns (2)

• Algorithm concerns
• Which algorithm should be used?

6
Separating the concerns (3)

• Wouldnt it be nice if someone with accurate
  skills in algorithm engineering were able to
  handle the algorithm concerns?

7
The Framework (1)

• Avestruz
• A multi-agent framework that generates tools to
  classify documents
• All stated issues are hot spots of the framework
• Platform concerns hot spots are totally detached
  from algorithm concerns hot spots

8
The Framework (2)

• The Strategy
• First, instantiate all the platform concerns hot
  spots
• Then, give someone with accurate algorithm
  engineering skills the opportunity to instantiate
  the algorithm concerns hot spots

9
The Benefits

• Ability to test different algorithms
• A specialist in algorithms has a powerful tool
  for making tests
• Ability to refine existing algorithms
• By taking into consideration characteristics of
  the classified documents domain, it is possible
  to improve existing algorithms

10
The Agents

• A classifier agent is constantly
• Searching for documents
• Choosing which will be processed
• Applying an algorithm to process a document
• Reporting the results to somebody

11
The Multi-Agent System

• To isolate the multi-agent system, an agent
  management layer was created
• Creation of the agents
• Execution control
• Initialization
• Finalization (premature or normal)
• Statistics

12
The Multi Agent System
Agent
Hot-Spots
13
The Multi Agent System
14
The Multi Agent System
Application
Multi-Agent System
Agent
Hot-Spots
15
The Multi Agent System

• Agent coordination is a frozen spot
• Blackboards to exchange messages
• Synchronization of the agents execution
• Infra-structure for the cooperation of the agents

16
The Multi Agent System

• The multi-agent system was as much encapsulated
  (hidden) as possible from the instantiation task
• The framework instantiator does not need to know
  it is based upon agent technology

17
Current Work

• Generate instances
• A document classification tool for SHOE documents
  (Semantic Web)
• A probe to look for relevant documents about
  information security over the Internet
• Integration with a knowledge management tool

18

• Thank you!