Assisting Domain Experts To Formulate

1
Assisting Domain Experts To Formulate
SolveConstraint Satisfaction Problems

• Stuart Chalmers, Derek Sleeman
• Computing Science Dept, The University of
  ABERDEEN, UK
• email sleeman, schalmer_at_csd.abdn.ac.uk

2
Outline

• Introduction to CSPs KA for CSPs
• Knowledge Acquisition MORE, MOLE SALT
• CSP formulation solving
• Capture / Acquire
• Transform
• Solve
• Conclusions/Future Work
• Using the semantic web to enhance CSP modelling

3
Introduction CSPs are GOOD

• CSPs (Constraint Satisfaction Problems) -
  powerful technique for problem solving, e.g.,
  able to solve large, combinatorial tasks
• User has to identify variables, ranges
  associated with variables, relevant relationships
  / constraints pertinent to task
• CSP solver seeks to find assignments to all
  variables such that each constraint is satisfied
• Constraint Programming represents one of the
  closest approaches computer science has yet made
  to the Holy Grail of programming the user states
  the problem, the computer solves it.
• E. Freuder

4
But

• Expertise / background knowledge needed to create
  CSPs the so called Modelling task. The analyst
  needs to
• Identify key variables, features and their
  domains
• Specify constraints over those entities
• Understand feedback from the CSP solver when
  tasks are over- and under- specified

5
How to get from thisto this?

• SEND
• MORE
• MONEY
• What numbers do the letters S,E,N,D,M,O,R,Y
• Represent?

• - use_module(library(clpfd)).
• mm(S,E,N,D,M,O,R,Y, Type) - domain(S,E,N,D,M,O
  ,R,Y, 0, 9),
• S0, M0, all_different(S,E,N,D,M,O,R,Y),
  sum(S,E,N,D,M,O,R,Y),
• labeling(Type, S,E,N,D,M,O,R,Y).
• sum(S, E, N, D, M, O, R, Y) - 1000S 100E
  10N D 1000M 100O 10R E 10000M
  1000O 100N 10E Y.
• Answer
• S 9, E 5, N 6, D 7,
• M 1, O 0, R 8, Y 2

6
Typical Brain Teaser

• The Car Games is a simple problem that can be
  describe as follow
• Three persons took part in choosing a car in
  Aberdeen at the weekend, each of them has a name
  (Arthur, Brendan, Cormick), a height (tall,
  medium, short) and the cars they chose have a
  ranking with respect to their top speeds (1st,
  2nd 3rd)
• The other constraints are
• The medium persons car is ranked two places
  higher than Arthurs. Arthurs car is blue, and
  Cormicks car is not green
• Arthur is the short person
• Cormick did not select the 1st ranking car
• Also, two persons cannot have the same name,
  height or car-ranking

7
Classical KA Systems

• MORE MOLE ask questions about the (diagnostic)
  classes the observables
• SALT (the Propose Revise PSM) asks for
• Updates to the current configuration
• Constraints to be imposed on the variables
• Fixes to be applied when a constraint is
  violated
• (Set of descriptors)

8
Knowledge Acquisition for CSPs

• Frisch, Miguel et al. define 5 CSP problem
  categories
• Scheduling, Configuration, Assignment,
  Construction, Positioning
• We want to
• Find specific characteristics of these categories
  create templates for each one
• Assist user in choosing appropriate template for
  their problem (OR use a series of linked
  templates)
• Help user to instantiate this/these templates
  (knowledge elicitation)
• Solve task as CSP (or make recommendations if no
  solution found or problem under-specified)

9
3 Stages of CSP Solving

• Capture
• Elicit knowledge from the user using KA
  interfaces
• Fit to pre-defined template
• Transform
• Represent CSP using OWL
• Use Semantic Information to enhance CSP
• Solve
• Solve CSP using standard finite domain techniques
• Provide feedback to user on over/under
  constrained problems

10
KA Interface n

Capture
Task Specific Information
SWRL/OWL converter
SWRL/OWL Representation
Acquire Additional Semantic Information
WWW
Transformation
Enhanced SWRL/OWL Task Representation
SWRL/OWL to CSP mapping
Results or feedback (To be used in further task
specification)
Internal CSP Representation
Solving
CSP Solver
CSP solving engine
11
Systematic approach to CSP task formulation

• Identify important objects / concepts
• Identify features associated with each object /
  concept
• Identify possible values of features
• Identify objects described in the task
• Identify other information / relationships
  provided about the task

12
Capturing User Information

• GUI Interface
• Split solving stages
• Add objects
• Add instances
• Add relationships between objects instances
  (rules constraints)

13
Transformation

• Constraints represented using CIF/SWRL
• Variables/Domains represented in OWL-lite
• CIF/SWRL
• Extends SWRL using CIF
• Allows fully quantified constraints to be
  expressed naturally
• Currently no W3C standard for expressing fully
  quantified constraints on the Semantic Web
• Allows us to use semantic web tools to aid in
  problem modelling

14
Transformation

• CIF/SWRL Example S 0

• /
• 
  
• rdfresource"http//www.w3.org/2003/11/swrlbgrea
  terThan"/
• rdfresource"svar"/
• rdfdatatype"http//www.w3.org/2001/XMLSchemaint
  "0

15
Implementation

• AXIS Web Service GUI interfaces
• Various interfaces implemented
• Map colouring, cryptarithmetic problem, room
  allocation problem
• CIF/SWRL intermediary language
• Represent Constraints using CIF/SWRL
• Represent variables domains using OWL
• CHOCO finite domain CSP solver
• Currently only binary constraints

16
Current Work

• Easy-of-use study (e.g., with KA UIs versus
  pen-and-paper for Crypt Arithmetic, Graph
  Colouring, Brain teasers)
• CSP Feedback
• Provide feedback on problem
• Over-constrained no solution, highlight
  constraints that are preventing solution being
  found
• Under-constrained too many solutions, show
  unconstrained domains to prompt user to enter
  more information

17
Enhancing a task description using the Semantic
Web

• Recall that the domain expert extracts variable,
  features etc from a task. Suppose he extracts Car
  colour, manufacturer, price
• The analyst uses a tool like ONTOSEARCH2
  (www.ontosearch.org) to search for ontologies
  which include concept car summaries their
  features as Car colour, manufacturer,
  price, speed,

18
Questions
CIF/SWRL
http//www.aktors.org
http//www.csd.abdn.ac.uk/research/akt/cif/
www.ontosearch.org
19
Adding semantic information
20
Adding semantic information
21
Adding semantic information
22
Adding semantic information

• Augment problem specification with semantic data
• Can we aid the user in formulating their problem?
• Can we highlight features of the problem they
  might not have recognised?
• Take current problem description
• Run through ontosearch (www.ontosearch.org)
• Find additional problem descriptors
• Allow user to revise problem specification