Transcript Towards an Automatic Non-Deterministic Web Service Composition

Towards an Automatic Non-Deterministic
Web Service Composition Platform
Markou, G. & Refanidis, I.
Dept. of Applied Informatics, University of Macedonia, Greece
NWeSP 2012 - Poster Session
Related Work
Literature for Web Service Composition (WSC) suggests that:
• Most widespread method of choice: AI planning
• Use of standards is diverse:
o OWL-S / PDDL widely used
o Other standards are also common(e.g., WSMO)
o Proprietary formats
• Very few cases of online WSC systems
o Especially, currently active
• Gap in the evaluation process of the current WSC systems
o Many simply evaluate
their methodology on a single case study, without
referring to quantitative criteria
o There are a few notable exceptions, especially in the past few years
NWeSP 2012 - Poster Session
2/12
System Goals – Functionalities (1/2)
Our work:
• Makes use of AI planning as our method of choice
• Uses OWL-S and PDDL*, the “de facto” WS /Planning standards
• Able to cope with the domain’s non-determinism
• Planner-independent
• Rigorous evaluation process based on
o Quantitative criteria
o Use case scenarios
o Additional manual WSC module as a comparison
• Will be the 1st online system supporting various stages of WSC
*Specifically, NuPDDL
NWeSP 2012 - Poster Session
3/12
System Goals – Functionalities (2/2)
Online (Sub)Modules:
Translator between AI and
Web Service languages
Also:
• Manual WSC
module
• WS descriptions’
editor
Registry / Advertisement /
Semantic search of Web
Services
Automatic WSC module
NWeSP 2012 - Poster Session
4/12
Proposed Approach - Application
Inline
operations
XML Editor
Semantic
with syntax
search
highlighting
alternatives
Pre-defined
OWL-S
templates Initial screen
Web
User email
Service
registration
details
Based on Google
Web Toolkit (GWT):
NWeSP 2012 - Poster Session
5/12
Registry
• Core of the application is based on iServe
o Open platform for publishing
and discovering services
o WSs are semantically described in OWL-S, not Minimal Service Model
• Used its web-based application for
o Browsing
o Querying
o Uploading
services
• Populated with version 4.0 of OWL-S Test Collection (OWL-S TC)
• Made several improvements to its interface and functionality
NWeSP 2012 - Poster Session
6/12
Translation between AI & WS Languages
•
AI planning language: PDDL, the de facto planning language
o Specifically, NuPDDL
 Compatible with PDDL2.1, retaining most of it
 Capable of modeling non-deterministic action effects (one of, unknown)
•
WS semantic description language: OWL-S, most widely used
o Heavily
•
influenced from planning languages, such as PDDL
Partial mapping from OWL-S to PDDL exists
o Part of it straightforward
o Conversion of non-physical knowledge
from OWL-S to PDDL problematic
 Introduction of new predicate in PDDL domain?
NWeSP 2012 - Poster Session
7/12
Automatic WSC Module
•
Future work – Basic directions:
o WSs may fail to execute correctly, e.g.,
 Unavailability of an atomic WS involved in the plan
 Output of its successful execution
is not the expected one
o Generate plans tackling the most influential
and likely contingencies
o Proposed solution based on contingent planning
 Produce seed plan
 Examine for significant/ likely points of failure
 Add conditional
branches
 Repeat until we reach a plateau or run out of time
 Re-planning module will be integrated
NWeSP 2012 - Poster Session
8/12
Evaluation
•
Currently based on the manual WSC module
o Modification of an open source BPMN 2.0 editor (PetalsBPM)
•
Making use of OWL-S TC Semantic Web Services v. 4.0
o Each scenario uses a subset of descriptions of an OWL-S TC domain
o Several semantic WS descriptions added/ modified for testing purposes
•
•
Each scenario with increasing non-determinism & complexity
Quantitative criteria to be considered:
o Number of Web Services considered for WSC
o Preprocessing time (parsing of ontologies’
concepts, etc)
o Transformation time (of WS domain to PDDL one)
o Planning time (to output a successful plan)
o Optimality of the outputted plans
NWeSP 2012 - Poster Session
9/12
Use Case Scenario 1/3: Movie Database
1) Movie database search using a (partial) movie title
o Based on Communication domain: 58 semantic WSs in total
o Fully deterministic
- Serialized composite WS output
o Retrieve comedy films with titles similar to user input
o Output:
 Movie titles
 Pricing
information (tax-free price, recommended price, etc)
Graphical workflow representing the movie database use case scenario
NWeSP 2012 - Poster Session
10/12
Use Case Scenario 2/3: Online Bookstore
2) Online bookstore product purchase
o User wants to purchase a book from an specific online bookstore with a
preferred method of payment
o Preferences regarding method of payment
o Based on Education domain: 285 semantic WSs in total
o Non-deterministic output
 Different outcomes:

Output if book in stock:
 Addition of book in shopping cart
 Purchase with the specified method of payment
 Output
information
regarding
such as its author
Graphical workflow
representing
the
onlineit,bookstore
use case scenario

No output if book not in stock
NWeSP 2012 - Poster Session
11/12
Use Case Scenario 3/3: Camera Search
3) Camera search and purchase
o User want to purchase an analog SLR model; willing to
settle for other ones if it is not in stock
 User preference towards specific model
 More than one sellers exist
o
o
Based on Economy domain: 359 semantic WSs in total
Possible iteration in output Web Service
 Since more than one sellers exist, if product not available at one, check
another
 If product not found in any of the sellers, try with other versions (digital
SLR, simple compact cameras)
Graphical workflow representing the camera search use case scenario
NWeSP 2012 - Poster Session
12/12