Problems and Challenges in the Integration of Semantic Services

1
Problems and Challenges in the Integration of
Semantic Services
Stefano Ceri Dipartimento di Elettronica,
Politecnico di Milano Stefano.ceri_at_polimi.it
2
PROLOGUE

• AN HISTORICAL PERSPECTIVE
• Data management community focus performance
• Assessments Laguna, Lagunita, Asilomar, Lowell
• Focus shift semantics
• Data-driven abstractions, components, mappings
• Knowledge-driven ontologies, semantic services
• Search-driven incorporating domain knowledge
• Issue working in the large and together
  (interoperability)

3
The good news is

• Web Services !
• Many software systems expose a WSDL interface and
  can be called from integrator applications.
• Services can be registered at this time their
  semantic domain becomes known (even if they are
  NOT semantic web services).
• Services can be composed by invoking them
  according to some known pattern it is possible to
  respond to complex users needs

4
TALK PART 1 Integration of search services

• THE LOWELL CHALLENGE Find an ethnical
  restaurant in a nice place close to Milano
• Logically a composition of domains
• Restaurants (ethnical)
• Geo-locations (nice place close to Milano)
• This is now almost solved by local.google.com!
• But in general no system is capable of composing
  arbitrary semantic domains

5
Example Books written by researchers

• Books on Amazon
• Papers in DBLP

6
Example rock CDs high in iTunes ranking

• rock CDs on Amazon
• high in the iTunes ranking

7
Example same news in two newspapers

• News from Il Corriere della Sera e La
  Repubblica

8
The simplest patternJOIN of Web Services

• Input XML Items Resulting from two Web Service
  Calls
• Output Concatenation of matching XML Items
• Matching condition Between XML items, using
• value equality
• partial set matching
• term matching within a vocabulary
• external ontology
• ..
• This can be generalized to an algebra of web
  service operations
• Joint work with Daniele Braga, Alessandro Campi,
  Alessandro Raffio (Polimi)

9
Architecture
10
Join of two Web Services

• Given two services Si and Sj, a query q which is
  decomposed into two queries qi and qj.
• The join of the two services is obtained by
  composing the results xi and yj returned by qi
  and qj,, producing a sequence R of elements
• rk ?c(xi, yj), ?k?
• c composition operation
• ?k is the relevance index of each result item
• ?k ?i ? ?j ? mij
• ?i relevance index of the result produced by Si
• ?j relevance index of the result produced by Sj
• mij ranking of the comparison between xi e yj

11
Value-based matching
12
Search Space

• Tile tij blocks bxi e byj

13
Scenarios

• Simple matching the dominating cost is Web
  Service request-response
• Semantic matching the dominating cost is
  matching pairs of XML fragments
• Interactive search systems the emphasis is on
  the first block of good results (top k)
• Analysis-oriented systems the emphasis is on
  getting all the results in the right ranking
  order

14
Tile extraction methods

• Lazy
• Parallel execution of request-responses
• Ideal if matching costs are low

15
Tile extraction methods

• Nested Loop
• Ideal if one service has a step

• Merge Scan
• Ideal if matching cost dominate

16
Tile extraction methods

• TEO (Tile Extraction Optimal)
• Ideal if services return numeric relevance
  indexes

17
Tile extraction methods

• Heuristic TEO
• Improves by local search

18
Method comparison

• Methods are compared on the basis of the number
  of produced result items realative to visited
  tiles

19
Conclusion of part 1

• The message it is possible to bridge
  domain-specific search engines by a new algebra
  whose operands are homogeneous XML fragments and
  whose main operator is the JOIN
• This requires
• Set-up time - system organization
• Publishing time semantic description of
  services
• Matching time routing of the query to the
  appropriate servers

20
PART 2 Integration in the semantic web

• An extension of the current web in which
  information is given well-defined meaning, better
  enabling computers and people to work in
  co-operation'.
• The Semantic Web provides a common framework
  which allows data to be shared and reused across
  applications, enterprises, and community
  boundaries.
• Sir Tim Berners-Lee

21
Modeling semantics using ontologies, services,
goals, and mediators WSMO

• WSMO (Web Service Modeling Ontologies) describes
  the semantic web using four elements
• ontologies ( ) provide the terminology
  used by other elements
• goals ( ) define the problems
• web services ( ) provide components
• mediators ( ) bypass interoperability
  problems.

22
Designing Web Applications for WSMO

• Use of high-level design (methods and tools)
• Visual models
• Automatic deployment (code generation)
• Powerful verification techniques
• Use classical abstractions of Web applications
• Data modeling
• Process modeling
• Web application modeling
• Web service modeling
• Derive WSMO descriptions from such ingredients
• Joint work with Marco Brambilla, Federico Facca
  (Polimi), Irene Celino, Dario Cerizza, Emanuele
  Della Valle (Cefriel)

23
Model-driven design for WSMO
24
A typical WSMO scenario
25
Process View of the wwMediator
26
WebML view of thewwMediator
27
Mediation service extraction

• A draft WebML representation can be extracted by
  the BPMN model.
• The designer then completes the WebML unit
  descriptions by adding their properties.
• The visual WebML representation completely
  specifies the wwMediator semantics in terms of
• Web service invocations
• Lifting-lowering operations
• Content management operations affecting meta-data
  and case-isomorphic entity
• Control flows

28
Conclusion of part 2

• Our current approach from Model-Driven Software
  Engineering to the Semantic Web
• Presently a proof of concepts, some translators,
  no environment
• The rest is under construction (but the Web
  design technology is solid, see www.webml.org and
  www.webratio.com)
• Our future approach use WSMO concepts as first
  class citizens, and modeling software behavior
  using them
• But still with visual abstractions covering data,
  processes, and hypertexts - and with full code
  generation capabilities towards conventional and
  WSMO-empowered execution environments

29
The end