Download presentation
Presentation is loading. Please wait.
Published byEmory Black Modified over 9 years ago
1
1 September 2003 Brokering Services A Transformation-Based Perspective Stephen Potter & Marco Schorlemmer www.aktors.org
2
2 September 2003 Services and the Semantic Web l Semantic Web vision: a move toward augmenting information with well-defined and machine-processable semantics. l Semantic Web as a medium for: –service-providers to advertise their services; –service-requesters to publicise their needs. l Missing: some means by which available services are matched to stated needs. l A web service is a service that is visible on and invoked via the web (using web-based communications protocols).
3
3 September 2003 Some sort of brokering mechanism? l Service-providers advertise their available services to the broker using some specification language; l Service-requesters post their requests to the broker in the same language; l Broker attempts to find one or more available services (or sequences of services) that would appear to satisfy the request. l Broker then: –Sends these details to requester; –Or invokes the services directly and returns results; –(Or tells requester that no matching services were found.)
4
4 September 2003 Brokering Problem l Few brokers? –Centralise brokering services; –But: overheads of storing and reasoning with many advertisements (and is ‘general reasoning’ adequate?) l Or many brokers? –Specialised brokering services; –But: too many and the problem shifts to becoming one of finding the right broker… l Want broker to have reasonable coverage: need flexible brokering mechanism.
5
5 September 2003 Brokering and Ontologies l Application ontologies used to provide content of specifications (and define valid reasoning). l Two extremes: ‘ontology-independent’ and ‘ontology-dependent’ brokers. l ‘Ontology-independent’ broker: –General applicability but relies on simple pattern- matching reasoning to find services. l ‘Ontology-dependent’ broker: –Sophisticated ontological reasoning but limited applicability (must re-engineer for new domains).
6
6 September 2003 Information Transformations l So: want a brokering mechanism that lies somewhere between the two extremes. l Unifying concept of a service as providing an information transformation provides this? –Some transformation of input-information into output-information. l Insist on the presence of these abstract concepts in the application ontology: –Relating the services described using the ontology to the broker’s abstract idea of an information transformation. –Gives the broker a ‘semantic hook’ into the ontology. –Gives provider- and requester-agents some idea of the appropriate content of service descriptions.
7
7 September 2003 A Brokering Environment l Assume environment includes: –Agents (have unique identifier/location (URI), plus ability to communicate with environment (basic mechanism + ACL): Service-providers; Service-requesters; One or more brokers (locations of which are known a priori to providers and requesters); –Brokering ontologies (locations known a priori to (some) agents.) l An agent has the ability to process and ‘understand’ certain ontologies.
8
8 September 2003 Broker ‘Capability’ l Broker makes known its ‘ontological capability’ to the environment in terms of the ontologies that it can handle: –Either explicitly, by, for example, giving particular application ontology URIs; –Or, more generally, by specifying the ontology language it can deal with (e.g. RDFS, OWL) along with certain facts it expects to find in any ontology. l Assume this capability is also known to other agents a priori. l Example of an ontological capability…
9
9 September 2003 Example Ontological Capability l Ontology language: –class(C) defines a class (concept), here considered to be a collection of individuals in the domain of discourse. A class is a unary relation, C(x), which is interpreted as meaning that individual x is an instance of class C. –subclass-of(C,P) : class C is a subclass of class P iff every instance of class C is also an instance of class P. –relation(R) refers to a set of tuples that represents a binary relationship among individuals in the domain of discourse. Hence R(x,y) is interpreted as meaning that individual x stands in relation R to individual y. –subrelation-of(R,P) : R is a subrelation of P iff every tuple of R is also a tuple of P. –implies(A,B) : material implication (if A then B). –and(A,B,...) : logical and, can take any number of arguments. –thing, a term indicating the most general class and relation, a term indicating the most general relation. l Ontology facts: –Must provide axioms defining classes input_information, output_information and relation transformation, plus the following axiom: implies(information_transformation(I,O), and(input_information(I), output_information(O), transformation(I,O))).
10
10 September 2003 Example Ontology - nlp-ont class(language_data). class(processed_data). … subclass-of(language_data,thing). subclass-of(processed_data,language_data). … class(representation_format). class(free_text). … class(language). class(english). class(chinese). … class(domain). class(business). class(government). … relation(language_process). relation(analysis). relation(generation). … subrelation-of(language_process, relation). subrelation-of(analysis, language_process). … class(input_information). class(output_information). relation(information_transformation). relation(transformation). implies(information_transformation(I,O), and(input_information(I), output_information(O), transformation(I,O))). implies(input_information(I), and(language_data(I),language(I), representation_format(I),domain(I))). implies(output_information(I), and(language_data(I),language(I), representation_format(I),domain(I))). implies(transformation(I1,I2), language_process(I1,I2)). relation(generation). … subrelation-of(language_process, relation). subrelation-of(analysis, language_process). … class(input_information). class(output_information). relation(information_transformation). relation(transformation). implies(information_transformation(I,O), and(input_information(I), output_information(O), transformation(I,O))). implies(input_information(I), and(language_data(I),language(I), representation_format(I),domain(I))). implies(output_information(I), and(language_data(I),language(I), representation_format(I),domain(I))). implies(transformation(I1,I2), language_process(I1,I2)).
11
11 September 2003 Service Descriptions l A Service is defined as: x y.information_transformation(x,y) input_information(x), output_information(y), transformation(x,y). and
12
12 September 2003 Example Services: l tokeniser(x,y) unprocessed_data(x), british_english(x), free_text(x), domain(x), tokenised_text(y), british_english(y), xml_encoding(y), domain(y), tokenising(x,y). l tagger(x,y) tokenised_text(x),english(x), markup(x), domain(x), pos_tagged_text(y), english(y), xml_encoding(y), domain(y), pos_tagging(x,y). l Advertising a service: l E.g., tokeniser service: l …and a request has the same form. input_information output_information transformation
13
13 September 2003 Broker Algorithm l With new request : 1.Search advertised services for those whose capabilities match the desired output_information and transformation of D. 2.If the input_information of these services also matches, then they represent solutions. 3.Else try to form sequences by finding other services whose output_information matches the input_information of these matching services and combining the two into a sequence… 4.…and repeat from 2.
14
14 September 2003 Brokering Examples l New Request: pquery(x,y) unprocessed_data(x),language(x),free_text(x),domain(x), parse_tree(y),language(y),markup(y),domain(y), analysis(x,y). l Matches: ms-parser(x,y) unprocessed_data(x),chinese(x),free_text(x),domain(x), parse_tree(y),chinese(y),xml_encoding(y),domain(y), parsing(x,y). l …and also the sequence: tokeniser(x,y) unprocessed_data(x),british_english(x),free_text(x),domain(x), tokenised_text(y),british_english(y),xml_encoding(y),domain(y), tokenising(x,y)...and then: tagger(x,y) tokenised_text(x),english(x),markup(x),domain(x) pos_tagged_text(y),english(y),xml_encoding(y),domain(y), pos_tagging(x,y)....and then: parser(x,y) pos_tagged_text(x),british_english(x),xml_encoding(x),domain(x), parse_tree(y),british_english(y),xml_encoding(y),domain(y), parsing(x,y).
15
15 September 2003 Discussion: General l Basic reasoning mechanism: subsumption –Broker must be able to ‘prove’ subsumption over the terms in its specified ontology language (but this need be neither complete nor sound…) –Trade-off: completeness/soundness of reasoning vs. expressiveness of language vs. efficiency? l Ontology requirements: –Ontology must contain ‘hooks’ to allow the broker to comprehend it: is this a reasonable request? l When is broker algorithm valid? –When transformations are independent (in some sense)… l How general is this notion of an information transformation?
16
16 September 2003 Discussion: Relationship with DAML-S/OWL-S l Service Profile inputs and outputs implicitly represent transformation. l But no consideration of preconditions/ effects here… –How are these expressed/reasoned with in distributed environments? l Profile describes what should be said, but not how to say it… –here we are interested in the expression of this content. l (Here we are trying to preserve generality as far as possible…)
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.