Semantic Web Cased Based Reasoning AIST Meeting

Semantic Web & Cased Based Reasoning AIST Meeting JPL, CA 2003 Mehmet S. Aktas maktas@cs. indiana. edu

Outline n n Semantic Web Overview ¨ Semantic Web ¨ Motivations ¨ Ontology Languages ¨ Semantic Web and Cased Based Reasoning Overview ¨ Cased Based Reasoning ¨ CBR Process ¨ Conversational Cased Based Reasoning AIST Meeting JPL, CA 2003

Semantic Web Overview “The Semantic Web is a major research initiative of the World Wide Web Consortium (W 3 C) to create a metadata-rich Web of resources that can describe themselves not only by how they should be displayed (HTML) or syntactically (XML), but also by the meaning of the metadata. ” From W 3 C Semantic Web Activity Page n “The Semantic Web is an extension of the current web in which information is given well-defined meaning, better enabling computers and people to work in cooperation. ” n Tim Berners-Lee, James Hendler, Ora Lassila, The Semantic Web, Scientific American, May 2001 AIST Meeting JPL, CA 2003

Motivations Difficulties to find, present, access, or maintain available electronic information on the web n Need for a data representation to enable software products (agents) to provide intelligent access to heterogeneous and distributed information. n AIST Meeting JPL, CA 2003

The Semantic Stack and Ontology Languages B A The Semantic Language Layer for the Web From “The Semantic Web” technical report by Pierce A = Ontology languages based on XML syntax B = Ontology languages built on top of RDF and RDF Schema AIST Meeting JPL, CA 2003

Resource Description Framework (RDF) - I Resource Description Framework (RDF) is a framework for describing and interchanging metadata (data describing the web resources). n RDF provides machine understandable semantics for metadata. This leads, ¨ better precision in resource discovery than full text search, ¨ assisting applications as schemas evolve, ¨ interoperability of metadata. n AIST Meeting JPL, CA 2003

Resource Description Framework (RDF)- II n RDF has following important concepts ¨ Resource : The resources being described by RDF are anything that can be named via a URI. ¨ Property : A property is also a resource that has a name, for instance Author or Title. ¨ Statement : A statement consists of the combination of a Resource, a Property, and an associated value. Example: Alice is the creator of the resource http: //www. cs. indiana. edu/~Alice. AIST Meeting JPL, CA 2003

The Dublin Core Definition Standard n RDF is dependent on metadata conventions for definitions. n The Dublin Core is an example definition standard which defines a simple metadata elements for describing Web authoring. n It is named after 1995 Dublin (Ohio) Metadata Workshop. n Following list is the partial tag element list for Dublin Core standard. ¨ ¨ ¨ ¨ Creator: the primary author of the content Date: date of creation or other important life cycle events Title: the name of the resource Subject: the resource topic Description: an account of the content Type: the genre of the content Language: the human language of the content. AIST Meeting JPL, CA 2003

Example Alice is the creator of the resource http: //www. cs. indiana. edu/~Alice. Property Resource Property Value creator = http: //purl. org/dc/elements/1. 1/creator http: //www. cs. indiana. edu/~Alice • Property “creator” refers to a specific definition. (in this example by Dublin Core Definition Standard). So, there is a structured URI for this property. This URI makes this property unique and globally known. • By providing structured URI, we also specified the property value Alice as following. “http: //www. cs. indiana. edu/People/auto/b/Alice” Inspired from “The Semantic Web” technical report by Pierce AIST Meeting JPL, CA 2003

Example Why bother to use RDF instead of XML? Alice is the creator of the resource http: //www. cs. indiana. edu/~Alice. <rdf: RDF xmlns: rdf=”http: //www. w 3 c. org/1999/02/22 -rdf-syntax-ns##” xmlns: dc=”http: //purl. org/dc/elements/1. 1” xmlns: cgl=”http: //cgl. indiana. edu/people”> <rdf: Description about=” http: //www. cs. indiana. edu/~Alice”> <dc: creator> <cgl: staff> Alice </cgl: staff> </dc: creator> </rdf: RDF> • Information in the graph can be modeled in diff. XML organizations. Human readers would infer the same structure, however, general purpose applications would not. • Given RDF model enables any general purpose application to infer the same structure. Inspired from “The Semantic Web” technical report by Pierce AIST Meeting JPL, CA 2003

RDF Schema (RDFS ) n n RDF Schema is an extension of Resource Description Framework. RDF Schema provides a higher level of abstraction than RDF. ¨ ¨ ¨ n n n It resembles objected-oriented programming specific classes of resources , specific properties, and the relationships between these properties and other resources can be described. RDFS allows specific resources to be described as instances of more general classes. RDFS provides mechanisms where custom RDF vocabulary can be developed. Also, RDFS provides important semantic capabilities that are used by enhanced semantic languages like DAML, OIL and OWL. AIST Meeting JPL, CA 2003

Limitations of RDF/RDFS n No standard for expressing primitive data types such as integer, etc. All data types in RDF/RDFS are treated as strings. n No standard for expressing relations of properties (unique, transitive, inverse etc. ) n No standard for expressing whether enumerations are closed. n No standard to express equivalence, disjointedness etc. among properties AIST Meeting JPL, CA 2003

DAML, OIL and DAML+OIL - I n RDFRDFS define a framework, however they have limitations. There is a need for new semantic web languages with following requirements n They should be compatible with (XML, RDF/RDFS) n They should have enough expressive power to fill in the gaps in RDFS n They should provide automated reasoning support n Ontology Inference Layer (OIL) and DARPA Agent Markup Language (DAML) are two important efforts developed to fulfill these requirements. n Their combined efforts formed DAML+OIL declarative semantic language. AIST Meeting JPL, CA 2003

DAML, OIL and DAML + OIL - II n DAML+OIL is built on top of RDFS. n It uses RDFS syntax. n It has richer ways to express primitive data types. n DAML+OIL allows other relationships (inverse and transitivity) to be directly expressed. n DAML+OIL provides well defined semantics, This provides followings: n Meaning of DAML+OIL statements can be formally specified. n Machine understanding and automated reasoning can be supported. n More expressive power can be provided. AIST Meeting JPL, CA 2003

Example How is DAML+OIL is different than RDF/RDFS? Example: T. Rex is not herbivore and not a currently living species. n This statement can be expressed in DAML+OIL, but not in RDF/RDFS since RDF/RDFS cannot express disjointedness. n DAML+OIL provides automated reasoning by providing such expressive power. For instance, a software agent can find out the “list of all the carnivores that won’t be any threat today” by processing the DAML+OIL data representation of the example above. ¨ RDF/RDFS does not express “is not” relationships and exclusions. ¨ From “The Semantic Web” technical report by Pierce AIST Meeting JPL, CA 2003

Web Ontology Language (OWL) n n n Web Ontology Language (OWL) is another effort developed by the OWL working group of the W 3 Consorsium. OWL is an extension of DAML+OIL. OWL is divided following sub languages. n OWL Lite n OWL (Description Logics) DL n OWL Full – limited cardinality OWL Lite provides many of the facilities of DAML+OIL provides. In addition to RDF/RDFS tags, it also allows us to express equivalence, identity, difference, inverse, and transivity. OWL Lite is a subset of OWL DL, which in turn is a subset of OWL Full. AIST Meeting JPL, CA 2003

From Semantic Web to Cased Based Reasoning n Developing new tools, applications and architectures on top of the Semantic Web is the real challenge. n AI techniques should be used to utilize the Semantic Web up to its potentials. n CBR is an AI technique based on reasoning on stored cases. n CBR technique can be applied to do intelligent retrieval on metadata of codes related Earthquake Science. AIST Meeting JPL, CA 2003

Overview of Case-Based Reasoning What is CBR? n CBR is reasoning by remembering: It is a starting point for new reasoning n Problem-solving: CBR solves new problems by retrieving and adapting records from similar prior problems. n Interpretive/classification: CBR understands new situations by comparing and contrasting them to similar situations in the past n Case-based reasoning is a methodology of reasoning from specific experiences, which may be applied using various technologies (Watson 98) AIST Meeting JPL, CA 2003

What is CBR? Everyday Examples of CBR n Remembering today’s route from the place you live to campus and taking the same route. n Diagnosing a computer problem based on a similar prior problem. n Predicting an opponent’s actions based on how they acted under similar past circumstances n Assessing a hiring candidate by comparing and contrasting to existing employees AIST Meeting JPL, CA 2003

CBR Process n What is a Case? ¨ Input cases are descriptions of a specific problem. ¨ Stored cases encapsulate previous specific problem situations with solutions. ¨ Another way to look at it: n Stored cases contain a lesson and a specific context where the lesson applied. n The context is used to determine when the lesson may apply again. AIST Meeting JPL, CA 2003

CBR Process When and how are cases used? Given a Problem Description (P. D. ) to be solved, CBR follows a cyclical process. ¨ REtrieve the most similar case(s) ¨ REuse the case(s) to attempt to solve the problem ¨ REvise the proposed solution if necessary ¨ REtain the new solution as a part of new case. n AIST Meeting JPL, CA 2003

CBR Process The CBR Cycle Problem Retrieve Case-Base Reuse Retain Revise Confirmed solution Proposed solution AIST Meeting JPL, CA 2003

Conversational CBR (CCBR) n CCBR is a method of CBR where user interacts with the system to retrieve the right cases. n System responds with ranked cases and questions at each step n Question-answer-ranking cycle continues until success or failure AIST Meeting JPL, CA 2003

Conversational CBR n CCBR facilities ¨ Question management facility ¨ Case management facility ¨ GUI for user-system interaction ¨ Facilities to display questions or cases AIST Meeting JPL, CA 2003

A Prototype CCBR Application AIST Meeting JPL, CA 2003

A Prototype CCBR Application n Purpose Intelligent retrieval on metadata describing codes written for earthquake science. ¨ Guidance on how to run the codes to get reasonable results. ¨ Guidance for inexpert users to browse and select codes ¨ n Casebase disloc - produces surface displacements based on multiple arbitrary dipping dislocations in an elastic half-space ¨ simplex - inverts surface geodetic displacements to produce fault parameters ¨ VC - simulates interactions between vertical strike slip faults. ¨ AIST Meeting JPL, CA 2003

A Prototype CCBR Application n Classification Initial effort – dummy cases created to classify the different codes ¨ A general approach is needed ¨ AIST Meeting JPL, CA 2003

A Prototype CCBR Application CCBR CASE Solution Problem Feature = <Question, Answer> AIST Meeting JPL, CA 2003

A Prototype CCBR Application How does Case Ranking take place in CCBR? n n n Retrieved cases are sorted based on their consistency with the query case. As the questions are answered more cases are eliminated. A case is ruled out only if there is a conflict between the case and the query case Consistency number for a case remains same if the case has no answer for the question. Consistency number for a case gets incremented if the case has the same answer to the question as the query case. AIST Meeting JPL, CA 2003

A Prototype CCBR Application CCBR CASEBASE A Case from CASEBASE A Feature 1 Feature 2 Feature 3 Feature 4 Case Query Case Feature 1 Feature 2 Feature 5 IF ((A. Feature 1. Solution = B. Feature 1. Solution) & (A. Feature 2. Solution = B. Feature 2. Solution)) THEN Consistency # = 2 Case = <Problem, Solution> AIST Meeting JPL, CA 2003 B

A Prototype CCBR Application How does question ranking take place in CCBR? n n Questions can be ranked based on their frequency factor Questions can be ranked based on predefined inference rules Only distinguishing questions are to be ranked Questions can be YES/NO questions, multiple choice questions or questions with numerical answers. AIST Meeting JPL, CA 2003

References n W 3 C Semantic Web Activity Page. Available from http: //www. w 3. org/2001/sw/. n T. Berners-Lee, J. Hendler, and O. Lassila, “The Semantic Web. ” Scientific American, May 2001. n Resource Description Framework (RDF)/W 3 C Semantic Web Activity Web Site: http: //www. w 3. org/RDF/. n D. Brickley and R. V. Guha (eds), “RDF Vocabulary Description Language 1. 0: RDF Schema. ” W 3 C Working Draft 23 January 2003. n The DARPA Agent Markup Language Web Site: http: //www. daml. org. n OIL Project Web Site: http: //www. ontoknowledge. org/oil AIST Meeting JPL, CA 2003

References n n n CBR on the web http: //www. cbr-web. org Case-Based Reasoning Resources http: //www. aaai. org/Resources/CB-Reasoning/cbr-resources. html AI Topics - CBR http: //www. aaai. org/AITopics/html/casebased. html A mailing list including announcements, questions, and discussion about CBR, managed by Ian Watson mailbase@mailbase. ac. uk Riesbeck & Schank, Inside Case-Based Reasoning, Erlbaum, 1989. Kolodner, Case-Based Reasoning, Morgan Kaufmann, 1993. AIST Meeting JPL, CA 2003