14.2 Technological Concepts 307 Table 14-1 RDF Schema constructs (excerpt) RDF/RDFS Construct Description Comment/Example Class Definition rdfs:Resource All things described by RDF; entities are resources. This is the root class from which all others are derived. rdfs:Class rdfs:Literal Definition of a class (using rdf:type). rdfs:Class is an instance of rdfs:Class. rdfs:Literal is an instance of rdfs:Class. rdf:type rdf:XMLLiteral Predicate used to allocate typification. Predicate used to designate XML literals. rdf:type is an instance of rdf:Property. rdfs:subClassOf Predicate used to specify an inheritance hierarchy. rdfs:subClassOf is an instance of rdf:Property. be found at (http://www.w3.org/2000/01/rdf-schema). Table 14-1 lists some of the important RDF Schema constructs. If we want to express that automobile is a class in a user-defined schema (with mymodel as prefix), we form an RDF-3 tuple (triple): mymodel:automobile rdf:type rdfs:Class To derive a subclass, lorry, from automobile, we write the following 3-tuple: mymodel:lorry rdfs:subClassOf mymodel:automobile We can now allocate associations to the schema for automobile. First, we use a type assignment, weight-in-kg, to declare an association (attribute), and then we specify that weight-in-kg is assigned to the automobile class: mymodel:weight-in-kg rdf:type rdf:Property mymodel:weight-in-kg rdfs:range mymodel:automobile OWL uses RDF Schema to form class hierarchies and to declare associations between classes. Table 14-2 lists the most important OWL constructs. RDF Schema, in turn, uses RDF constructs, like rdf:type or rdf:Property, to describe a schema.
Note: If you are looking for good and high quality web space to host and run your jsp application check Lunarwebhost jsp web hosting services
14.2 Technological Concepts 305 The next things RDF distinguishes are resources and literals. A resource is designated by its Uniform Resource Identifier (URI), e.g., (http://www.SemanticWeb.org/schema-daml01/ #hasHomepage). Subjects and predicates are always resources, while an object can be either a resource or a literal. A literal has no resource allocated to it in a URI; instead, it is a character string to which the predicate is allocated as a value. The schematic view shows resources as ellipses and literals as rectangles. Figure 14-7 uses the example from Figure 14-6, additionally modeling the statement …and the homepage was created by Stefan Decker. http://www.SemanticWeb.org/ http://www.daml.org/ projects/ #11 schema-daml01/ #hasHomepage http://www-db.stanford.edu/ OntoAgents http://purl.org/ dc/elements/1.1/ Creator Stefan Decker Source: http://cse.hanyang.ac.kr/~jmchoi/class-old/ 2002-2/cse995/INTRO.ppt Figure 14-7 RDF graph with a literal allocated as a value. RDF is primarily designed for machine-assisted Web navigation, and therefore, it is hard to read for humans in its basic form, i.e. a series of URI-based statements. This was the reason why the notion of namespaces was subsequently borrowed from XML to improve readability. The convention is to associate a prefix with the respective URI. In our current example, such a definition is used to abbreviate (http://www.SemanticWeb.org/schema-daml01/#)to sw . The result is that the predicates are shortened to sw:hasHomepage or dc:Creator, respectively (see Figure 14-8). Figure 14-9 summarizes the entire RDF code for these statements. RDF itself uses namespaces to define other useful constructs, which are then fitted with the rdf prefix, including: rdf:type : RDF predicate for type allocation. rdf:Bag : RDF predicate for an unordered set. rdf:Seq : RDF predicate for an ordered set. rdf:Alt : RDF predicate for a set of optional choices. We can see that RDF offers a vocabulary that we can use to attach relationships between data sources on the Web. These data sources are called resources in RDF. The naming convention
Note: If you are looking for good and high quality web space to host and run your jsp application check Lunarwebhost jsp web hosting services
304 The Semantic Web The Network of Meanings in the Network of Documents unequality relations, and transitivity and symmetry conditions for associations. One important restriction and limitation of OWL-Lite is that cardinalities can only be specified as 0 or 1 (expressing can exist / must exist , respectively). The reason is said to be that systems with this restriction are easier to implement for the reason cited above, that full treatment of cardinality leads to higher computational complexity. OWL-DL (DL stands for description logics) and OWL-Full enhance OWL-Lite. Though OWL-DL and OWL-Full share the same syntax, some constructs in OWL-DL have a restricted formal semantics compared with OWL-Full. For example, in OWL-DL classes cannot be used as instances due to the strict separation of instances and classes in all description languages. For the practical use of technologies for the Semantic Web, it is important to understand how XML, RDF, RDF Schema, and OWL engage with one another, though opinions are divided about how meaningful this structure is one might argue that a lot of new notation has been introduced for little additional semantics, compared with the KRLs that have existed for many years, such as KL-One (Brachman and Schmolze 1985), Telos (Mylopoulos et al. 1990, Nejdl et al. 2001), KIF (Genesereth 1998), and conceptual graphs (Sowa 1976, Lukose et al. 1997). In order to get a grasp of the notations and their interdependencies, the next two subsections will briefly describe how OWL is embedded in RDF and RDF Schema, using XML as the general representation format. Subsequently, we will use an example to show how these nested Web languages can be used in practice. RDF The Resource Description Framework (RDF) has been designed to connect Web-based data sources to the emerging Semantic Web at relatively low cost (Decker et al. 2000). According to its inventors, the RDF data model is similar to the object-oriented model, but we think that it is actually close to the functional data model (Shipman 1981). RDF distinguishes between entities, which are represented by unique identifiers (UIDs), and statements, which are valid between entities. This means that a statement connects a subject (source entity) and an object (destination entity) via a predicate/property. Figure 14-6 shows a schematic view of the RDF statement The OntoAgents project has a home page at (http://www-db.stanford.edu/OntoAgents) . As can be seen, RDF can be thought of as using a distinction between subject, predicate, and object. http://www.SemanticWeb.org/ http://www.daml.org/ projects/ #11 schema-daml01/ #hasHomepage http://www-db.stanford.edu/ OntoAgents Subject Predicate Object Source: http://cse.hanyang.ac.kr/~jmchoi/class-old/ 2002-2/cse995/INTRO.ppt Figure 14-6 RDF graph with subject, predicate, and object.
Note: If you are looking for good and high quality web space to host and run your jsp application check Lunarwebhost jsp web hosting services
14.2 Technological Concepts 303 Person: Tom Person: Mary Expr Believe Thme Expr ThmeWant T Agnt Marry Thme Sailor Propositon: Situation: Figure 14-5 Conceptual graph for Tom believes that Mary wants to marry a sailor . aware that a computer system actually interprets only mathematical-logical relations, without understanding what believe , want , and being sure generally mean in human interactions. To simulate a behavior that somewhat corresponds to human behavior, both the concrete model and the underlying logic have to be very extensive. Our examples can only scratch the surface of this extensiveness. It should be noted that there is no difference between predicate logic, the Knowledge Interchange Format (KIF), and conceptual graphs (CGs), as far as the underlying formal semantics is concerned. However, conceptual graphs use a notation that is easier to read for humans, thus giving us an insight into the requirements which an application for the Semantic Web has to meet. 14.2.3 Semantic Markup on the Web The concepts available for semantic markup on the Web are based on the work on knowledge representation languages (KRLs; see above) during the past thirty years. The three main representatives for semantic markup are OWL, RDF, and RDF Schema. DAML + OIL = OWL The DAML (DARPA Agent Markup Language) description language was introduced as a KRL extension to RDF and XML in 2000 (Hendler and McGuinness 2000). Roughly at the same time, a European research project developed the Ontology Inference Layer (OIL) (Fensel et al. 2000). The protagonists of both KRLs agreed later on to continue developing a joint Web Ontology Language (OWL) (McGuinness and van Harmelen 2003). OWL currently exists in the form of three sub-languages with increasing expressivity: OWL- Lite allows specifying inheritance hierarchies with associations and type restrictions, equality and
Note: If you are looking for good and high quality web space to host and run your jsp application check Lunarwebhost jsp web hosting services
14.2 Technological Concepts 301 crawling, flying). These are some of the restrictions to be expected from ontologies that were built ad-hoc. Ontologies for Computer Applications For the Semantic Web, we use an explicit knowledge model that will serve to represent a knowledge space. According to (Gruber 1993), an ontology is an explicit specification of a conceptualization . In this context, we will use a terminology which, though not further analyzed, is described in a formally and logically substantiated Knowledge Representation Language (KRL; see next section). An ontology for computer applications is an attempt to explicitly represent the concepts of a knowledge space and the inherent dependencies and interplays between its concepts. Such an ontology is normally defined as the generally valid schema for all possible statements within this knowledge space (also known as the UoD -universe of discourse). Concrete specifics of this knowledge space can be maintained as instances of this schema. Formal Representation of Human Knowledge A Knowledge Representation Language (KRL) should be defined on a descriptive level on which any specific knowledge can be represented by specializing by the primitive terms of the representation language. The simpler the KRL, the more modeling will be required to describe even simple facts. The more powerful a KRL becomes, the more complex can be its use. In either case, every KRL has to deal with the problem that human language and human understanding of a language is highly context-sensitive, subject to conditions that change constantly, while even good knowledge-based systems have only as much flexibility as was programmed into them (if one views the formal definition of search spaces as programming , which is our chosen interpretation of knowledge-based programming). Our discussion of the Semantic Web assumes that the objective is to obtain explicit knowledge descriptions for the respective use purpose that can be interpreted by agents appropriately. The responsibility for this interpretation is vested in the agent and less in the ontology, provided the ontology is based on correct knowledge with regard to at least one such interpretation. We will introduce two KRL examples below. EER Diagrams Entity-relationship (ER) diagrams have been known as a conceptual language for the design of relational databases since the 1970s (Chen 1976). In the 1980s and 1990s, they were enhanced to support the modeling of inheritance hierarchies and aggregation relationships (Extended Entity- Relationship Diagram EER Engels et al. 1992). ER and EER models force the designer to make a distinction between entities and relationships that can arise between entities. One example from the tourism industry would be the entities guest and hotel, which can have a relationship through the term (guest) stays in (hotel). ER and EER also allow us to model quantitative restrictions in relationships. Such a restriction would be that a guest can stay overnight in a hotel only once at a time (but the guest can book several hotels for the same night). At the same time, many guests can stay overnight in one hotel (but with an upper limit). Both the differentiation of entities and relationships and the quantitative restrictions between the end points of a relationship are
Note: If you are looking for good and high quality web space to host and run your jsp application check Lunarwebhost jsp web hosting services
302 The Semantic Web The Network of Meanings in the Network of Documents characteristic properties of all data-centric KRLs, but are sometimes excluded from inference- centric KRLs because the consideration of numeric constraints can inflate the search space for the inference engine of a knowledge-based system in prohibitive ways. Conceptual Graphs The idea of conceptual graphs (CGs Sowa 1976, Lukose et al. 1997) is based on the existential graphs originally developed by Charles S. Peirce in the 19th century, but which Peirce never fully completed (see hrstr m 1997). A CG distinguishes between concepts and conceptual relations, where concepts can be related only through conceptual relations. The most important idea for our purposes is that CGs allow us to map complex language expressions, such as type hierarchies. Moreover, it is possible to describe nested contexts. The following examples serve to better understand this concept (see Sowa 2000, pp. 476 491). The sentence John is going to Boston by bus uses four concepts, namely John, Go, Boston, Bus. These concepts are attached to three conceptual relations: Agent (Agent), Instrument (Inst), and Destination (Dest), as shown in Figure 14-4. Person: John Agnt Go Dest Inst Bus City: Boston Figure 14-4 Conceptual graph for John is going to Boston by bus . The sentence Tom believes that Mary wants to marry a sailor has a nested conceptual structure, as shown in Figure 14-5. The outer level describes the proposition that Tom believes something. Inside that context is another context of type situation, which describes a situation that Tom believes Mary wants. The resulting CG represents the sentence Tom believes that Mary wants to marry a sailor . The proposition box has as theme (Thme) a situation that Mary hopes will come to pass. Inside the proposition box are three concepts: Person: Mary, Want, and the situation that Mary wants. Since those three are only asserted within the context of Tom s belief, the graph does not imply that they must exist in the real world. Since Mary is a named individual, one might give her the benefit of the doubt and assume that she exists; but her desire and the situation she supposedly desires exist in the context of Tom s belief. If his belief is false, the referents of those concepts might not exist in the real world. All of this would present a few problems for our agent: Should the agent assume that Mary really exists? And if so, does she really want to marry a sailor, or is Tom wrong and she actually wants to marry a computer scientist? We always have to be
Note: If you are looking for good and high quality web space to host and run your jsp application check Lunarwebhost jsp web hosting services
300 The Semantic Web The Network of Meanings in the Network of Documents description language, namely for the Knowledge Interchange Format (KIF), the Constraint Choice Language (CCL), and the Resource Description Framework (RDF). In addition, a possibility has been created for groups of agents to commit themselves to a common ontology ( ontological commitment ), which will be made available to them by a specialized ontology agent. All agents can make requests to the ontology agent. This mechanism supports a better interplay between ontology vendors and application-specific users (agents). The Ontology Service Specification defines how agents can load themselves with an ontology and then act according to this ontology. Despite this promising work, no implementations that seriously use this concept are currently known. 14.2.2 Ontologies When using the term ontology , we have to distinguish between the philosophical discipline of ontological analysis and the product of such an analysis, i.e., the description of a specific knowledge space. Philosophy, Ontology, and Language Criticism Before describing the technical implementation of ontologies, one is well advised to undertake a brief excursion into ontology as a field of philosophy. This little excursion is aimed at showing the potential difficulties that could result from implementing solutions for the Semantic Web, if one tackled ontology-building in a linguistically naive manner. Ontological analysis has long been an important part of philosophical reasoning. In the Western culture, ontology dates back to Aristotle. According to Aristotle, the subject of ontology is the study of categories of things that exist or may exist in some domain. Aristotle distinguished ten basic categories for classifying anything that may be said or predicated about anything, and devised rules for combining them to create new concepts or sub-categories. In the 18th century Kant proposed an alternative form of categorization, which will not be discussed here; interested readers are referred to (Sowa 2000). Since the middle of the 19th century, philosophers have tried to separate the categories of human thinking from its linguification.Thishas ledtoa language-critical philosophy which, for example, identifies very clear correlations between the grammar of Indo-Germanic languages and Aristotelian categories (K oller 1988). Now, did Aristotle find the basics of human knowledge, or only the grammar structures of the Greek language? After all, mathematical logic is also based on these categories! So, in view of working out ontologies that can be interpreted by Web applications and software agents, we must be aware that these ontologies may well be pragmatic and useful, but that all too often, they are based on a rather na ive language understanding. We have to understand that, although a knowledge model created in this way could be helpful for data exchange, it can hardly claim ontological generality. Let s look at an analogy about cars to better understand this problem: Though driving a car is faster than walking, we have to accept that cars generally work well only on roads, and that they are suitable neither for climbing stairs nor for swimming. This means that technically implemented ontologies can have a clearly defined and specific benefit, similar to cars, but, in the world of human meanings, the notion of the automobile is not suitable to make useful statements about all forms of motion (swimming, climbing stairs,
Note: If you are looking for good and high quality web space to host and run your jsp application check Lunarwebhost jsp web hosting services
14.2 Technological Concepts 299 research network for agent systems (AgentLink 2003). This chapter will be limited to the FIPA standard, which has found broad industrial acceptance, and which can be used to clearly represent the technical components of agent systems. The Foundation for Intelligent Physical Agents (FIPA) is a non-profit organization to promote the standardization and further development of agent systems (FIPA 2003). Figure 14-3 shows the general model of an agent communication, assuming that an agent,X, follows a goal, G. The agent will approximate goal G by an intention, I. Now, intention I has to be converted into a speech act, S. To realize speech act S, a message, M, has to be sent to agent Y. This message is decomposed into transport packets similarly to a remote procedure call (RPC), and eventually converted into an intention of agent Y on the meanings level. Agent YAgent X Convert to transport form Convert from transport form Message delivery / tranportation service Goal G Intent I Speech act S Msg M Message M Figure 14-3 The FIPA agent communication model (Foundation for Intelligent Physical Agents 2002). FIPA uses ACL (Agent Communication Language), which comprises five syntactic categories: Type of communication act (performative); Agents participating in the communication (sender, receiver, reply to); Content of message (content); Description of the content (language, encoding, ontology); and Control of conversation (protocol, conversation identifier, reply with, in reply to, reply by). The only parameter that is mandatory in all ACL messages is the performative, although it is expected that most ACL messages will also contain sender, receiver, and content parameters to be able to reproduce a speech act. However, a valid minimum conversation would simply consist of a request , followed by an accept or not understood reply. Agents and Their Ontologies Of course, the discussion about the Semantic Web has given the software agent issue an additional lift. Consequently, there are currently FIPA standardizations for three variants of the content
Note: If you are looking for good and high quality web space to host and run your java application check Lunarwebhost java web hosting services
298 The Semantic Web The Network of Meanings in the Network of Documents semantic networks and rule systems based on predicate logic. AI has developed different description formalisms for these domain models, including specific description languages for ontologies. Today, ontologies serve two purposes. They are used either as global integration schemas , for example, when building portals and data warehouses, or as controlled vocabularies for document annotations (meta-data). Only very few true AI applications use ontologies based on data to construct further-reaching inferences (forward or backward chaining rule systems), which can ultimately serve in decision-finding processes. Opinions in practice differ about the necessary degree and method of systematizing ontologies. Many ontologies are actually no more than controlled vocabularies. Though some are built hierarchically in the form of taxonomies, the type of hierarchization can be quite varied, reaching to formally defined inheritance hierarchies in object-oriented models. Very few ontologies stand up to a philosophical-ontological analysis, in addition to the existence of formal semantics (If my dog is muzzled, is the muzzle part of the dog, and why or why not? If the dog is mine, is his muzzle also mine, and what consequences does it have?). Such questions are not obscure, because if you tell your software agent to get a dog muzzle, you d probably want to make sure it won t buy a dog only because it can logically deduce that when you buy a dog you also get yourself a muzzle. So, those who seriously deal with ontology-based agent systems are well advised to use trustworthy knowledge models. Issues relating to ontologically clean modeling have been addressed by Guarino and Welty (2002) and others. We can see that the Semantic Web is not only a technical challenge, but also a philosophical and socio-economic one, especially if we look at the goal the European Union (EU) set itself in 2000, that, by 2010, it should become the most competitive and dynamic knowledge-based economy in the world (the goals have recently -2005 -been toned down, but the aspiration is still there). The EU is currently funding significant research in the field of the Semantic Web and will probably continue to do so in the next three to five years. 14.2 Technological Concepts Based on the Semantic Web architecture briefly described above, this section discusses agent standards, ontologies, and markup languages for the Semantic Web. The FIPA standard is currently the most matured for agent systems (FIPA 2002). With regard to ontologies, we have to distinguish between the semantic level and the syntactic level. On the semantic level, it is important to know what can principally be expressed with a certain knowledge representation language. For example, we cannot describe countable sets in first-order logic, because the concept of cardinality is excluded from first order logic. On the syntactic level, it is important to represent different semantic sub-levels with the help of adequate and concrete description languages. We will show that the knowledge level (OWL) and the Web resources level (RDF) and the programming interfaces (XML) can be sufficiently represented by embedding OWL in RDF and XML to yield practicable systems. 14.2.1 Agents According to the FIPA Standard There are currently a number of academic agent projects and also a number of different development frameworks. A good source for various approaches is AgentLink, a European
Note: If you are looking for good and high quality web space to host and run your java application check Lunarwebhost java web hosting services
14.1 Fundamentals of the Semantic Web 297 Figure 14-2 Using SHOE for semantic markup of a Web page. 14.1.3 The Role of Ontologies Ontologies are conceptualizations that codify the knowledge of experts in a form that can be comprehended and reconstructed by experts. Ontologies are not necessarily subject to a formal semantics in the logical-mathematical sense. If their semantics is non-formal, their notion of truth is then only measurable as the degree of the informal consensus between the users of the ontology. If an ontology is available in the form of an axiomatized theory that enables a model-theoretic interpretation, then its semantics is formal, enabling statements about its logical truth to be checked. Dealing with ontologies has at least two roots in informatics, namely data modeling as part of database research, and knowledge representation as part of research in the field of artificial intelligence (AI). For example, databases have an explicit ontology , namely the database schema based on a data model. However, this schema normally maps a very limited section of the real world into the computational system, because the database application orients itself to the respective problem, thus being suitable for logical deductions only within narrow boundaries. In practice, such a schema is interpreted only by the specified applications, which use the schema to access data. AI takes a broader look by studying whole application domains for which it develops
Note: If you are looking for good and high quality web space to host and run your java application check Lunarwebhost java web hosting services