Online recommendation phase

The online recommendation phase provides an intelligent and proactive learning service during the pre-visit, onsite-visit, and post-visit stages. The recommendation service has two modes, user specification and system recommendation. The user specification mode makes

The system recommendation mode makes recommendations by matching the individual usage pattern (dynamic preference), group usage patterns and global usage patterns with cluster content patterns. The similarity between the usage patterns and the cluster content patterns is measured by calculating the distribution of concept definitions in the concept tree according to the shared ontology. The algorithm for measuring the similarity between a usage pattern and content pattern is shown below:

U1, U2 represent the preference concept definition list specified by a learner and concept definition list specified by content experts U1i，U2j are two nodes in the same concept tree

A(U1i,U2j) is U1i & U2j’s co-ancestor If U1i=U2j then

During the pre-visit stage, a learner can specify topics from the recommendation list, and arrange a plan to meet his requirements for both modes. The recommended content focuses on exhibition items and education activities that the learner plans to see. During the onsite-visit learning stage, the learner can choose the plan-learning mode by downloading a registered plan, or choose the package-learning or free-learning mode. The extended content is proposed dynamically, while the refreshed individual usage pattern is similar to cluster content patterns.

The recommended content focuses on collection knowledge units, exhibition units and messages of education activities according to the up-to-date individual usage pattern. After visiting the museum, the learner can obtain further recommendations from the system based on the last learning activity or matching individual, group or global usage pattern with the cluster content patterns. The recommendations are extended to relevant areas in collection, exhibition and education, and are categorized according to content type in the multi-layer reusable content structures defined in Section 4.5.1. Figure 4.6 shows an example of the personalization service during the pre-visit, onsite-visit and post-visit stages. Figure 4.7 shows the implementation system to fulfill this ubiquitous personalization service model.

Figure 4.6 Example of ubiquitous personalization service

Figure 4.7 The implementation system for ubiquitous personalization service

Chapter 5

Ontological techniques for reuse and sharing knowledge

5.1 Background and Purpose

In the previous two chapters, the KBDM framework was constructed through the UCKM model and the knowledge-based ubiquitous learning service model. A unified knowledge system is created by integrating knowledge bases from various individuals, domains, projects, and applications. A knowledge-based digital museum is created by reusing the unified knowledge content; these contents and applications are diffused to various users proactively, adaptively and ubiquitously based on common and sharable ontological knowledge concepts.

However, knowledge bases should be reused and shared as broadly as possible among communities and institutes using common knowledge acquisition and representation tools.

Furthermore, implicit and innovative knowledge should further be automatically discovered and generated based on these knowledge bases. Consequently, knowledge sharing and discovery services can be incorporate into the KBDM framework to achieve these objectives.

Like most digital museums, NMNS aims to construct a virtual museum for the public by utilizing various information technologies. Though the content can be manually represented via query or through metadata schemas or hyperlinks, this study argues that the digital archive represents a promising model for providing "knowledge." Restated, the usability of current NMNS only focuses on providing explicit and static information. The current systems thus are insufficient for supporting advanced knowledge engineering, for example knowledge inference processes. To make provision for the future knowledge era, NMNS has surveyed

pioneered project based on ontological techniques for restructuring current digital contents into corresponding knowledge bases. Three main tasks involved in this project are:

(1) Constructing a vascular plant ontology which represents a prototype of the knowledge base of NMNS,

(2) constructing an herbal drug ontology which pretends another knowledge base from outside communities, and

(3) developing a knowledge system for verifying the usability of correlative knowledge bases.

5.2 XML-based and Ontological techniques

Various studies have utilized KBSs in various fields, including expert systems, artificial intelligence, and decision support systems. Additionally, various approaches and tools have been used over the years to develop KBSs. However, many studies have noted challenges of sharing and reuse among KBSs [12, 31, 81]. Swartout [76] have concluded that KBSs are generally based on embedded systems and utilize their own special terminology system.

Consequently, traditional knowledge technology and engineering restrict interoperable abilities among KBSs. Recently; XML technology has been introduced for data exchange and integration purposes in various application areas. Moreover, the XML-based plus ontological technique is an emerging technology that has been considered as an important solution for addressing the problem of reuse and sharing in KBSs.

The ontologies have long been used to express a shared understanding of information by human beings. Gruber [30] defined ontology as "a specification of a conceptualization." A conceptualization is an abstract, simplified view of the world that is used for representational purposes. That is, the ontology is a formal description of the concepts, attributes, and

relationships involved in constructing common understanding for cognitions of real world events. A conceptualization is an abstract, simplified view of the world that is used for representational purposes. ‘Explicit’ means that the type of concepts used and the constraints on their use are explicitly defined. The ontology is a formal description of the concepts, attributes, and relationships involved and should be machine-readable. The ‘shared’ means ontology is used in constructing common understanding of a domain that can be communicated across several parties and application systems. For museums, ontology acts as the common understanding concept between knowledge specialists as well as the communication agent between knowledge specialists and users.

In the KBSs community, ontological techniques are used by defining a set of terminologies, the universe of discourses, and axioms. Consequently, as Gruninger [31] noted, ontology is useful for defining the common vocabulary used for representing shared knowledge. Since the advantages of XML are obvious, for example machine readability, automated parsing, and self-descriptive ability, there has been strong development of the XML-based ontological languages including RDF, DAML+OIL, and OWL.

Resource Document Framework (RDF) and RDF Schema (RDF-S) are the first solution from the World Wide Web Consortium (W3C) that provides a way for defining structured sets of terms involving class hierarchies and constraints [19]. RDF is a format for making descriptive assertions and statements by using two roots: metadata and representation. In library communities, electronic catalogs have been improved via the RDF-based Dublin Core format. For example, Amann [4] utilized RDF and RDFS to integrate ontologies and thesauri that can be used to query metadata. Additional background on RDF and RDF-S can be found in (http://www.w3c.org/RDF).

Because RDF/RDF-S has a limited ability to describe relationships with respect to

Language) has been developed as an extension to XML and RDF. The recent release of DAML plus Ontology Interchange Language (OIL) provides a rich set of constructs for creating ontologies and marking up information to make it machine readable and understandable [21, 73]. More details of DAML+OIL can be found in the Web site (http://www.daml.org).

Ontology Web Language (OWL) is the newest XML-based ontological language to be developed by the W3C. OWL inherits most features from DAML+OIL and has now become the recommended version (official and formal standard). According to the OWL specifications, the standard has three increasingly expressive sublanguages for different levels of usability:

OWL-Lite is designed for a classification hierarchy and simple constraint features; OWL-DL supports those users who want the maximum expressiveness while retaining computational completeness and decidability; and OWL-Full has useful computational properties for reasoning systems with maximum expressiveness, but without computational guarantees.

Because of the need to take advantage of the inference process, this study chooses OWL-DL as the ontology language. Further background of OWL and relative specifications can be found in the Web site (http://www.w3c.org/2004/owl).

5.3 Develop Ontological KBSs

To demonstrate the scope of KBS, this study creates the vascular plant ontology that belongs to the botany domain. Additionally, herbal drug ontology is created as a correlative domain. In practice, both ontologies are created using the OWL-DL format and are published on the Web. The herbal drug ontology integrates the vascular ontology during the execution. A developed application system then implements user requests to retrieve knowledge contents by implementing inference process.

Figure 5.1 illustrates the design architecture used in this study. The right block of the figure shows the multi layer reusable content architecture in the NMNS digital museum and the corresponding vascular plant ontology. Meanwhile, the left part of the figure illustrates an herbal drug ontology that is logically located outside NMNS. Both ontologies will be implemented in this study. Knowledge engineering procedures have been proposed for constructing ontological KBSs. Since there is different scope and content of various knowledge engineering procedures, this study has surveyed other research for further clarification [58, 62, 74, 75]. This study concludes and redefines knowledge engineering procedures for constructing ontological KBSs as follows:

„ Knowledge acquisition deals with how knowledge engineers gather expertise from experts.

„ Knowledge representation involves how knowledge engineers formalize expertise into a knowledge base format.

„ Knowledge retrieval involves the query process between applications and knowledge bases.

Figure 5.1 Overview of ontological architecture design

As Devanbu and Jones [20] and Baader and Nutt [6] observed, a KBS can be denoted as

=( , )

K = T A . The expression represents that a knowledge base (K =) can be derived from TBox (T ) and ABox (A ). The TBox contains intentional knowledge in the form of a terminology, and the ABox contains extensional knowledge or so called assertional knowledge. The key role in both the TBox and the ABox is description logic that offers inference capability based on terminology and assertions. Figure 5.1 shows that both TBox and ABox are the outcomes of constructing ontologies. The figure also stresses that both ontologies are logically divided into TBox and ABox respectively. Additional studies of KBS can be found in the research [6].

5.4 Knowledge Acquisition

Knowledge acquisition is the first phase in the development of a KBS. The main task involves how to gather cognized concepts from domain knowledge. This study invited

botanists and herbal pharmacologists as knowledge sources attempted to both vascular plant and herbal drug ontologies. The concept structure of vascular plant ontology utilizes the taxonomy of NMNS, which utilizes the specification of biological taxonomy (sometimes termed the Linnaean taxonomy due to its invention by Carl Linnaeus in 1735 [22].

On the other hand, pharmacologists have accumulated various resources such as literatures, herbal drug dictionaries, specifications, textbooks, and so on, to identify concepts related to herbal drugs. Figure 5.2 shows that knowledge acquisition involves gathering non-ordered and unstructured cognition into an identical ordered and structured concept. In the conceptual process, knowledge engineers must identify each concept that is comprised of a terminology, attributes, and relationships among terminologies. Numerous concepts can then be structured as taxonomy and become domain ontology.

Figure 5.2 Knowledge acquisition and concepts

To efficiently elicit concepts from knowledge and create ontology structure, this study trained both knowledge engineers and experts to utilize the formal concept analysis (FCA) method and supporting tools to facilitate acquisition. FCA is a mathematical approach that analyzes relationships among components and calculates their dependency [27]. First, this

the knowledge domain, including objects, attributes, and their relations. The components can be justified during the analysis. Second, mathematic algorithms, such as the Duquenne-Guigues algorithm, are used to calculate dependency rates and derive implications regarding their relations. Finally, a terminology hierarchy is proposed and a concepts hierarchy can be considered. Consequently, this phase provides a knowledge analysis mechanism for gathering common concepts and their attributes of domain ontologies. As knowledge scientists agree, FCA is a mature theory and tools exist for supporting concept analysis. Additional details of FCA have been provided [27]. Some useful FCA tools include the following: Vogt [79] and http://sourceforge.net/projects/conexp. To extract ontological concepts from real world, the formal concept analysis (FCA) [27] was adopted. FCA has been borrowed from lattice theory and mainly used for the analysis of data for filtering explicitly given information and the detail definition for formal contexts and concept lattices is defined as follows [27]:

1. A triple (G, M, I) is called a formal context if G is a set of objects, M is a set of attributes, and I is binary relations between G and M.

2. Let g be an object and m be an attribute. For deriving A⊆B andB⊆G,

The idea of FCA lattices comprises objects, attributes and the incidence relations into a table. The lattices can be represented by a line diagram to help designers to classify objects in different categories. To get ontological concepts precisely, the analysts can divide objects independently by adding more attributes in a lattice. For example, a simple set of vascular plants include objects, such as Pinacene, Thelypteridaceae, Aceraceae, and Araceae. The domain expert initially noted several attributes, such as seed, fern seed, herb, and arborous tree. Figure 5.3 illustrates objects in the left side and attributes in the top. The cross symbol

indicates that the specific plant have the corresponding attribute. All empty cells indicate both plants and attributes don’t have any relations.

As shown in Figure 5.3, the object Thelypteridaceae and Pinacene have same attributes that will further grouped in the same category. To identify unique object, domain expert may give more attributes to separate them. For examples, the attribute seed can be further divided into Angiospermae and Gymnospermae. The expert then assigns Angiospermae into Thelypteridaceae and Gymnospermae into Pinacene respectively. The updated lattice context can also be represented by a line diagram for better visual verification.

Figure 5.3 An example of vascular plants in lattice context

Generating implicit concepts is an important feature of formal concept analysis [44]. We adopt Ganter’s definition of formal concept analysis to explain this feature. Let A and B be two objects, thenA→B for A B, ⊆M holds if A'⊆B', that is all attributes from A also has all attributes from B. Many FCA tools embed above attribute exploration mechanisms to facilitate object-attribute verification and consistency. As illustrated in Figure 5.4, the added attributes contribute each object independent. Consequently, FCA is useful for analyzing and extracting the ontological concepts.

Figure 5.4 An example of vascular plants in line diagram

5.5 Knowledge Representation

Knowledge representation involves representing expertise (i.e. concepts) into information systems with specific formats. This study utilizes OWL-DL as the format for representing ontologies that are further considered as knowledge bases. Ontology normally comprises three components including classes (concepts), properties (attributes), and individuals (instances). The description logic (DL) is a describable fragment comprising classes, properties, and logic notations that express attribute or class relations [6]. OWL-DL supports automated reasoning by calculating logical relations and classifying ontology hierarchy.

In this study, the vascular plant ontology begins from establishing its concepts hierarchy, which is identical to a pyramid or tree with seven levels: species, genus, family, order, class,

phylum, and kingdom. Each level has a specific description based on using the DL. For example, the plant "Gardenia Jasminoides Ellis", must be identified and located based on its corresponding levels, starting from Plantae, Angiospermae, Dicotyledoneae, Rubiales, Rubiaceae, to Gardenia. As object technology discipline, the child object inherits all features defined in its super objects. The following DL expressions are partial samples of the vascular plant which comprises Spermatophyte and Pteridophyta, and then the Spermatophyte comprises Angiospermae and Dicotyledons, and so on. The expressions can be represented using OWL-DL and are listed in Table 5.1.

Table 5.1 Partial expressions of the vascular plant using OWL-DL - <owl:Class rdf:ID="Vascular_Plant">

<owl:Class rdf:ID="Spermatophyte">

<owl:Class rdf:ID="Pteridophyta" />

</owl:unionOf>

</owl:Class>

</owl:someValuesFrom>

- <owl:onProperty>

<owl:ObjectProperty rdf:ID="hasVascular" />

</owl:onProperty>

</owl:Restriction>

</rdfs:subClassOf>

<rdfs:subClassOf rdf:resource="http://www.w3.org/2002/07/owl#Thing />

</owl:Class>

ontology and defines 266 named plant concepts in the vascular plant ontology. Some 313 herbal drug assertions and 722 vascular plant assertions are constructed respectively to provide facts. The ontologies described in this study have been published on the Web. The ontologies also can be reused and shared with other ontological knowledge bases. The establishing tasks are extremely time-consuming and require heavy loading; thus, the use of assisting tools is recommended. Some useful tools are available online, including Protégé (http://protege.stanford.edu/) and OilED (http://oiled.man.ac.uk/).

This study utilizes Protégé 2.1.2, OWL plugin, and Racer to establish ontological knowledge bases in vascular plants. Protégé is an integrate development environment (IDE) tools for building XML-based knowledge base systems. It provides simple and customizable editor and implementation functions. Protégé acts as a common architecture that allows extensible applications plug-in and provides add-on functionalities [50, 66]. The OWL plugin provides developers edit and visualize OWL classes and their properties. Moreover, OWL DL enables the definitions of description logic. Racer is a reasoner engine for developing inference capability. The Racer is especially designs for description logic computation based on TBox and ABox of knowledge base architecture. Consequently, the synergy of these supporting tools is helpful for creating ontological knowledge base.

To demonstrate the scope of ontological knowledge bases, this study creates the vascular plant ontology that belongs to the botany domain. In practice, ontologies are created using the Protégé and OWL plug-in editor and are published on the Web. A developed application system then implements user requests to retrieve knowledge contents by implementing inference process. To efficiently elicit concepts from knowledge and create ontology structure, this study trained both knowledge engineers and experts to utilize the formal concept analysis method and supporting tools to facilitate extracting concept structures. The conceptual structures further provide the inputs of establishing ontologies. As shown in the left window

of Figure 5.5, the conceptual tree view represents ontology class structure. When classes and properties of the ontology are established, the last step is giving formal definitions to describe each component. As shown in the middle bottom of Figure 5.5, for example, the asserted conditions window is the description of the "Rhododendron."

Each DL expressions can be associated with several items including notations, properties, and fillers. The filler can be a class, a set of classes, or a combination of another DL expression. For example, if the "Rhododendron" had a feature of the erect steam, we describe the expression using an existential notation( )∃ , a property (hasStem), and the filler (erectStem).

The DL expression therefore be defined as . The asserted conditions have two different categories: necessary (^ô ) and necessary & sufficient ( ). The necessary expressions represent a subsumption relationship and the necessary & sufficient expressions represent a bidirectional equivalent relationship. In this example, the "Rhododendron" only demonstrates necessary expressions.

hasStem ( erectStem )

$

≡

Figure 5.5. A Screen shot of using Protégé OWL plugin

5.6 Knowledge Inference and Retrieval

The protégé platform provides capabilities to integrate description logic reasoner engine, such as Jess, Clips, Racer, and so on. This study utilizes the Racer as a DL reasoner. The Racer provides several functionalities, such as consistency checking and taxonomy classifying for basic inference process. The consistency checking validates each class definition. If any inconsistent situation happened, the ontology has logic issue and can not be classified

The protégé platform provides capabilities to integrate description logic reasoner engine, such as Jess, Clips, Racer, and so on. This study utilizes the Racer as a DL reasoner. The Racer provides several functionalities, such as consistency checking and taxonomy classifying for basic inference process. The consistency checking validates each class definition. If any inconsistent situation happened, the ontology has logic issue and can not be classified