Edward Brent, Idea Works, Inc'

1
Representing Metadata with Intelligent Agents
An Initial Prototype

• Edward Brent, Idea Works, Inc.
• BrentE_at_missouri.edu www.ideaworks.com
• Albert F. Anderson, Public Data Queries, Inc.
• afa_at_pdq.com www.pdq.com
• and G. Alan Thompson, Idea Works, Inc.
• PDQ-Explore is being developed by Public Data
  Queries, Inc., with funding, in part, from Small
  Business Innovation Research (SBIR) and Small
  Business Technology Transfer Research (STTR)
  grants from the National Institute of Child
  Health and Human Development (NICHD) and the
  National Institute on Aging (NIA) of the National
  Institutes of Health (NIH). The awards are R41
  HD32222, R41 HD32220, R43 HD33633, R43 AG13832,
  R43 HD37311, R43 HD37738, and R43 HD38216
  (pending).

2
Introduction

• Digital representations of data make it possible
  to have intelligent interactive social science
  data capable of helping users formulate
  questions, specify analyses, and interpret
  findings.
• This presentation describes an intelligent user
  interface now under development for the
  PDQ-Explore information system that tries to
  achieve some of these objectives.

3
Topics to be Covered

• The PDQ-Explore System
• Design Strategies
• Varieties of Help Offered
• Two Prototype Modules

4
PDQ-Explore

• The PDQ-Explore information system combines
  paralleled high performance processors, data
  cached in random access memory, and efficient
  retrieval algorithms capable of processing tens
  of millions of records per second.

5
PDQ-Explore (continued)

• Complex queries can be defined and executed in
  real time to produce tabulations, summary
  statistics, correlation matrices, and data
  extracts.
• The system is structured as a client-server
  architecture with a graphical user interface
  accessible over the World-Wide Web.

6
PDQ-Explore InterfaceWorkspace Window
7
PDQ-Explore InterfaceQuery Setup Window
8
PDQ-Explore InterfaceQuery Results Window
9
PDQ-Explore InterfaceQuery Details Window
10
PDQ-Explore InterfaceCustom Item Setup Window
11
PDQ-Explore InterfaceCustom Item Assignment
Window
12
A Demonstration Version

• A demonstration version of PDQ-Explore with a
  preliminary Web-based interface is accessible
  from the Public Data Queries, Inc. home page at
  www.pdq.com.
• Example queries and the graphical-based client
  program can also be downloaded from that site.

13
Design Strategies

• The Vision Data as Agents
• Case-Based Reasoning
• Machine Learning
• Representation Metadata, XML, and Ontologies

14
The Vision Data as Agents

• Agents as Intelligent Interface Managers
• Agents as Personal Assistants
• Agents Behind the Scenes
• Agent-to-Agent Communication

15
An Agent-Enabled System Architecture
16
Agents

• a computer program capable of acting on behalf of
  the user to carry out tasks that have been
  delegated to it
• does not require the user to specify the task in
  all of its detail

17
Agents (continued)

• able to take an admittedly vague description of
  the task from the user and infer what the user
  means
• translates a general description into what may be
  many individual steps or tasks to perform
• Bradshaw (19976)

18
Case-Based Reasoning

• Find an old problem that is close in nature and
  expected solution to what we anticipate for the
  new problem based on similarities in
• substantive problem
• data set
• specific items referenced
• methods used (tabulations, graphics, etc.)
• source of the query
• Then help the user tailor the solution to fit the
  new problem

19
Machine Learning

• Create a system that can grow over time and
  evolve to meet the changing needs of users
• Identify successful queries and incorporate them
  into the knowledge base as new examples for use
  by future users

20
Representation Metadata, XML, and Ontologies

• Make digitized information readily accessible to
  a wide range of users and intelligent autonomous
  agent programs
• Represent data using the extensible markup
  language (XML) and a standardized ontology

21
Varieties of Help Offered

• Interpret and Clarify the Query
• Identify Key Variables
• Identify Relevant Data Sets
• Point to Related Literature
• Similar Queries to Serve as Models
• Measurement, Indices, Recoding, and
  Transformations
• Appropriate Tables or Analyses
• Check Assumptions
• Structured Tutorials

22
Interpret and Clarify the Query

• Tell the user how the program is interpreting
  their problem statement
• Identify key variables and types of analysis
  implied by that
• Ask the user for further clarifications as
  required

23
Identify Key Variables

• Identify specific variables in existing data sets
  and link to terms used by the user
• Provide extensive information on each variable
  including its developmental history,
  characteristics, and examples of its use in the
  literature and in past queries

24
Identify Relevant Data Sets

• List available data sets that include the
  variables identified
• Provide at the users request extensive
  information on the data set including studies
  from the literature using the dataset as well as
  previous queries.
• Permit the user to browse through the datasets
  searching by various indices

25
Point to Related Literature

• Intelligent agents would scan the literature
  identifying studies on a broad range of relevant
  topics as well as studies using these specific
  data sets
• The literature would be retrievable by data set,
  by variable, by types of analysis, and by broad
  topics
• Users could scan the literature and examine how
  they analyzed the data and key methodological
  decisions they made.

26
Similar Queries to Serve as Models

• The system would automatically incorporate
  subsequent queries into its knowledge base
• Relevant queries would be displayed to the user
  to help them specify their own analysis
• Queries would be selected that examine the same
  variables, use the same data sets, or employ
  similar forms of analysis

27
Measurement, Indices, Recoding, and
Transformations

• The system would show users strategies that have
  been used in the literature and in past queries
  to handle common problems, such as
• developing indices for key concepts
• transforming data to assure normality
• handling missing data
• recoding

28
Appropriate Tables or Analyses

• The system will identify relevant analyses or
  tables that appear appropriate for the problem
• It would also point to examples of those in the
  literature and in previous queries

29
Check Assumptions

• Once the user selects the tables of analyses to
  perform, the system can automatically check
  important assumptions for the analysis such as
  normality, level of measurement, and the number
  of categories.
• It can also point out other data issues such as
  common transformations, missing data problems,
  and so on.

30
Structured Tutorials

• The system can point the user to structured
  tutorials available over the Internet including
  multimedia presentations.
• These might include
• Multimedia presentations by experts
• Internet-Based Instructional Materials
• Multimedia Interactive Tutorials

31
Module One Case-Based Reasoning to Identify
Relevant Example Queries

• Case-based reasoning strategies are used to
• 1) collect information from the user regarding
  their objectives,
• 2) identify and display existing queries that are
  similar, and
• 3)facilitate the user modifying the existing
  query to accomplish their objectives.

32
User Objectives

• User objectives are specified on this form.
• For example, lets indicate that the user will
  examine a tabulation of individuals in 1990 in
  all households, looking at educational
  attainment, comparing across groups defined by
  region.

33
A Similar Query is Retrieved Displayed
34
Modify the Query to Meet Objectives

• Note that this query is similar to the users
  objectives in that it also examines a tabulation
  of individuals in 1990 in all households,
  comparing across groups defined by region
  (cities).
• The user can then change a few parameters of this
  query to meet their objectives.

35
Module 2 Advice Regarding Recoding or
Transformations

• This module uses information from the user to
  determine whether recoding is indicated and what
  the objectives of the recoding should be.

36
The Recoding Control Panel

• Users select conditions that characterize their
  study using this control panel.

37
Clicking a phrase shows details
38
Recommendations Can Be Detailed or Brief
39
Example Recoding Race
40
Summary and Overview

• This framework provides a plan for developing an
  interface that takes full advantage of digital
  databases.
• These two prototypes illustrate how this program
  will work
• We are proposing to develop the complete system
  in a Phase II STTR grant from the National
  Institute of Child Health and Human Development
  to Public Data Queries, Inc.