Specification of Policies for Web Service Negotiations

1
Specification of Policies for Web Service
Negotiations

• Steffen Lamparter and Sudhir Agarwal

University of Karlsruhe (TH)
Semantic Web and Policy WorkshopGalway, November
7th
2
Outline

• Motivation
• Modeling preferences Utility theory
• Preferences and Policies
• Policy Ontology
• Preference Modeling
• Conclusion
• Open problems / Outlook

3
Motivation

• Web services are highly configurable products
• Attribute value pairs are insufficient to
  describe offers and requests

encryption key 512 bits response time
5sprice 3 Euro
WS Provider I
Agent
I need a service with encryption key 128 bits,
response time lt 10s andprice lt 5 Euro
encryption key 128 bits response time
3sprice 4 Euro
WS Provider II

• Automatic selection as well as negotiation
  requires
• Preference information within the valid range
• Cardinal preferences to make multi-attributive
  decisions

4
Representing Preferences

• Multi-attribute utility theory
• Scoring function maps attribute values to a
  numerical measure
• This measure is comparable and can be aggregated
• ?Classical optimization algorithms can be used
• ? Allows realizing trade-offs (good expensive
  vs. bad cheap)
• Allows weighting of attributes
• Allows aggregation and weighting of preference
  functions for one attribute

5
Policies vs. Utility Functions

• Policies express preferences!
• Policies specify the allowed attribute range
  (e.g. encryption key lt 512 bits)
• Which attribute value is preferred (e.g. 128 bits
  or 512 bits)?

128 encryption key 512
u(x)
1
128 encryption key 512longer keys are
preferred
128
512
bits
-8
6
DOLCE-based Policy Framework
Privacy Policy
store
WS Provider
Private data
Storage Duration
1,2,,14
7
WS Invocation

• DOLCE used as modeling basis
• Reuse of modules Description and Situation,
  Ontology of Plans, Ontology of Information Objects

7
Modeling Utility Information
pv
yl
? Adding primitives for utility modeling
?
degree
8
Modeling Utility Information

• ? represents the points (x,y) that form the
  utility function
• Change Policy Value to a subclass of ?
• ? restricted to piecewise linear functions
• Satisfiability defines the degree a Situation
  Value satisfies the Policy Value
• YL contains an instance for each line in the
  function ?

9
Policy Evaluation

• Aggregation functions such as SUM, MIN, MAX, etc.
  are required ? Ontology formalism ALC(?)
  Baader,Sattler 03
• Deriving utility for a concrete Situation Value

P(satisfies degree, ? (yl ?))
10
Policy Evaluation

• Calculation of the overall utility according to
• Weighted degree of satisfaction (wds) is
  calculated by P(wds degree, satisfies
  degree , ?ij)
• True iff wds degree (satisfies degree)
  weight holds
• wds of attributes are aggregated to the overall
  utility P(degree, ? aj wds degree)
• GoodService v Service u 9 gt(0.7,degree)

11
Conclusion

• Bringing together two powerful paradigms
  Policy-based computing and utility theory
• Enables automated selection of services and
  negotiation of service parameters
• Preference information is modeled using DL
• Facilitates interoperability in open and
  heterogeneous environments
• Reuse of existing DL-reasoners
• Preference information can be used within the
  reasoning process

12
Open Problems / Outlook

• Checking for satisfiability and subsumption in
  ALC(?) may lead to undecidability Baader,Sattler
  03
• Specifying policies gets even harder
• Approximate preferences from existing policies
  Lamparter et. al. 05
• There are 30 years of work in the field of
  decision analysis and preference elicitation
  Keeney, Raiffa 76
• ? Support policy specification by reusing of
  existing preference elicitation techniques

13
References

• Baader, Sattler 03 Franz Baader, Ulrike
  Sattler Description logics with aggregates and
  concrete domains. Information Systems 28(8)
  979-1004 (2003)
• Keeney, Raiffa 76 Keeney, R.L. Raiffa,
  H.Decisions with Multiple Objectives
  Preferences and Value Tradeoffs. J. Wiley, New
  York, 1976
• Lamparter et. al. 05 Lamparter, S., Eberhart,
  A., Oberle, D. Approximating service utility
  from policies and value function patterns. In
  6th IEEE Int. Workshop on Policies for
  Distributed Systems and Networks, IEEE Computer
  Society (2005)

14

• Thank you!