A Trainable Information Distribution System to Support Crisis Management

1
A Trainable Information Distribution System
to Support Crisis Management

• Niels Netten (Ph.D. student)
• Human-Computer Studies Laboratory
• University of Amsterdam
• 19-04-2005

2
Presentation Overview

• Project Collaboration
• Introduction
• Goal and Approach
• Current Status

3
Research Project Collaboration

• Task-Adaptive Information Distribution (TAID)
• Adaptive workflow modelling
• Prof. Robert de Hoog (University of Twente)
• Guido Bruinsma (Ph.D. student)
• Adaptive learning methods
• Dr. Maarten van Someren (University of Amsterdam)
• Niels Netten (Ph.D. student)
• Interactive Collaborative Information Systems
  (ICIS-CHIM)
• Adaptive Information Delivery (HCI aspects)
• Dr. Vanessa Evers (University of Amsterdam)
• Henriette Cramer (Ph.D. student)

4
Introduction
5

• Crisis Response and Management
• Characteristics
• Multiple collaborators
• Variety of information sources
• Complex organisations
• Highly dynamic
• Managing the information flow (i.e. presenting
  the right information to the right person at the
  right time) is crucial for a successful crisis
  management.

6
Firework disaster Enschede
Crisis management is for all also information
management (Rapport Oosting, 2000)
For the effectiveness of actions in fighting the
crisis it is essential that people involved
timely acquire the information they need to take
their decisions.
7
Other Examples
Mont Blanc Tunnel Disaster (1999)
Hercules Disaster (1996)
8

• Current options for information distribution
• Broadcasting leads to information overload
• Information search up to the user. User is
  unaware of which information is available.

9

• Research problem
• The problem of selecting and distributing
  information to users as a function of their tasks
  and the state of their workflow in a
  collaborative dynamic setting.
• Being able to
• Determine task-relevant information for users
• Adapt to changing information needs of users
• Anticipate future information needs of users
• Actively push the information to the users

10
Goal and Approach
11

• Goal
• Develop an information distribution prototype
  system that provides users with task-relevant
  information.
• Approach
• Trainable information distribution system
• Using an adaptive workflow model for role-task
  descriptions
• Get work context of actors
• Each role has a set of tasks associated
• Use features from tasks descriptions to fasten
  learning of method

12

• Text/Document classification (e.g. Recommending)
• From a collection of documents (or fragments)
  that are labelled relevant/irrelevant a system
  with machine learning can learn to classify new
  text messages accordingly.
• Normally a large collection of text messages are
  needed to create a good classifier.
• We deal with short messages and therefore take
  predefined descriptions of tasks into account.
• Automatically construct context-specific task
  profiles

13
Training/Review phase
(Off-line Training)

Information Distributor
Task profiles
?
Context-specific Task profile Learning
Domain Experts
Information flow Log file
Feedback
?
14
Information Distribution phase
?
Stream of structured information (e.g. sensor
values)

Structured unstructured
process


Task descriptions
?
Stream of unstructured information (e.g. text
messages)
Workflow Information
Task profiles
Information flow Log file
15
Current Status
16

• Basic information distribution prototype system
  to be able to test dynamic collaborative
  situations with information communication.
• Goal of this first test is
• To see if adding task descriptions to message
  representations of the learning method leads to a
  better information distribution.
• Replay crisis scenarios.

17
Open Question

• We have a problem with finding detailed
  descriptions of information flows (e.g. log
  files) of crisis/disaster situations to use for
  our scenarios.
• If you have any of this data or know how to get
  it please let us know!!

18
Questions?

• netten_at_science.uva.nl