Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Commonsense knowledge support in database design expert systems Ding, Jie 1994

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_1994-0405.pdf [ 1.6MB ]
Metadata
JSON: 831-1.0087456.json
JSON-LD: 831-1.0087456-ld.json
RDF/XML (Pretty): 831-1.0087456-rdf.xml
RDF/JSON: 831-1.0087456-rdf.json
Turtle: 831-1.0087456-turtle.txt
N-Triples: 831-1.0087456-rdf-ntriples.txt
Original Record: 831-1.0087456-source.json
Full Text
831-1.0087456-fulltext.txt
Citation
831-1.0087456.ris

Full Text

COMMONSENSE KNOWLEDGE SUPPORTIN DATABASE DESIGN EXPERT SYSTEMSbyJIB DiNGB.Eng., Tsinghua University, 1990A THESIS SUBMITTED iN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF SCIENCEinTHE FACULTY OF GRADUATE STUDIES(Faculty ofCommerce and Business Administration)We accept this thesis as conformingto the required standardTHE UNIVERSITY OF BRITISH COLUMBIASeptember 1994© Jie Ding, 1994In presenting this thesis in partial fulfillment of therequirements for an advanced degree at the University of BritishColumbia, I agree that the Library shall make it freely availablefor reference and study. I further agree that permission forextensive copying of this thesis for scholarly purposes may begranted by the head of my department or by his or herrepresentatives. It is understood that copying or publication ofthis thesis for financial gain shall not be allowed without mywritten permission.(Signature)Department of__________________The University of British ColumbiaVancouver, CanadaDate S1’j7. 2]ABSTRACTConceptual database design is the most critical and difficult phase in designing adatabase centered application. It usually requires database design experts which are hard tofind and expensive. There has been some effort in building expert systems, including the ViewCreation System (VCS), to support this design process. However, all of these systems lackcommonsense knowledge human experts have and therefore can not provide effective supportto the user. A prototype commonsense knowledge base, the Commonsense Business Reasoner(CBR), has been built for the VCS. But it is not filly implemented and not connected to theVCS. The usefulness of commonsense knowledge in a database design expert system has notbeen studied. In this paper, a new domain of relevance ontology was proposed to storedomain of relevance for commonsense knowledge. Combined with the Naive Semanticsontology used in the CBR, a new commonsense knowledge base structure was built andimplemented. This was integrated into the original VCS and can provide interactivecommonsense knowledge support during the design process. Other improvements were alsodone to the VCS to make it more user friendly. A fill scale empirical study was conducted totest the effectiveness of the commonsense module. The results indicated subjects perceivedthe system with commonsense knowledge easier to use and finished the task in less time.However, there is no statistically significant difference in the design quality. Explanations tothese results are discussed as well as future research directions.11TABLE OF CONTENTSABSTRACTTABLE OF CONTENTS iiiLIST OF TABLES viiLIST OF FIGURES viiiACKNOWLEDGEMENTS ix1. INTRODUCTION 11.1 Objective of the Thesis 21.2 Outline of the Thesis 32. CONCEPTUAL DATABASE DESIGN 42.1 Overview of Conceptual Database Design 42.2 E-R Model 42.3 Relational Model 52.4 Process Model of Conceptual Database Design 62.5 Design Process of Expert vs. Novice 83. EXPERT SYSTEMS FOR CONCEPTUAL DATABASE DESIGN 103.1 Overview 103.2 The View Creation System (VCS) 103.2.1 Creation of the Initial E-R Model 103.2.2 Creation of the Final Relational Model 103.3 Problems with the VCS 113.3.1 Lack of Commonsense Knowledge 113.3.2 Poor User Interface 124. COMMONSENSE KNOWLEDGE 131114.1 Overview 134.2 Commonsense vs. Expert Knowledge 144.2.1 Structure 144.2.2 Generality 144.2.3 Availability 154.2.4 Usability 154.3 Naive Semantics 154.4 Commonsense Business Reasoner 174.4.1 Has Attribute 174.4.2 Has Key 184.4.3 Synonym 184.4.4Ako(AKindOf) 184.4.5 Subject 184.4.6 Object 194.5 Domain of Relevance for Commonsense Knowledge 194.5.1 Organization of the Domain ofRelevance Knowledge 204.5.2 Dimensions of Relevance 214.5.3 Some Examples 295. THE NEW VCS 315.1 Architecture of the New VCS 315.2 Consultation of Commonsense During Design Process 315.2.1 Stage One 325.2.2StageTwo 335.3 Other Improvements 335.3.1 Direct Input Method 345.3.2 Smart Cursor Positioning 345.3.3 Context Sensitive Message 345.3.4 Configurable Features 356. HYPOTHESIS DEVELOPMENT 366.1 The Research Model 366.1.1 Dependent Variables 366.1.2 Independent Variables 376.2 Commonsense Knowledge Effect 396.3 Task Complexity Effect 406.4 Interaction Effect 41iv7. RESEARCH METHODOLOGY 427.1 Experimental Design 427.2 Subjects 437.3 Independent Variables 447.3.1 Commonsense Knowledge 447.3.2 Tasks 457.4 Dependent Variables 457.4.1 Time Taken to Finish the Task 457.4.2 Design Quality 467.4.3 Perceived Ease ofUse 467.5 Experimental Procedure and Data Collection 477.5.1 Pre-Test Questionnaire 477.5.2 Introduction and Tutorial 477.5.3 Understanding the Test Case Description 487.5.4 Creation of the E-R Model 487.5.5 Post-Test Questionnaire 488. DATA ANALYSIS 498.1 Commonsense Knowledge Effect 508.1.1 Time and Perceived Ease OfUse 508.1.2 Design Quality 508.2 Task Complexity Effect 538.2.1 Time 538.2.2 Design Quality 538.2.3 Perceived Ease OfUse 548.3 Interaction Effect 559. CONCLUSION AND FUTURE RESEARCH 569.1 Limitation of the Study 569.2 Future Research 5710. BIBLIOGRAPHY 59APPENDIX A GRADING SCHEME FOR DESIGN QUALITY 62APPENDIX B STANDARD DESIGN 63VAPPENDIX C STATISTICAL RESULTS 66APPENDIX D TUTORIAL 69APPENDIX E TUTORIAL TASK 72APPENDIX F EXPERIMENT TASKS 73viLIST OF TABLESTable 1 Time taken to complete task by groups (minute) 49Table 2 Quality of design by groups (percentage) 49Table 3 Perceived ease of use by groups (total score on the EOU questionnaire, ranges from -21 to 35) 49vi’LIST OF FIGURESFigure 1 Process model of conceptual database design (Batra and Davis, 1992) 7Figure 2 Structure of the new VCS 31Figure 3 Research model 39Figure 4 Average number of extra entities and relationships entered by each subject 53yinACKNOWLEDGEMENTSI own an unpayable debt of gratitude to Professor Robert Goldstein, my thesisadvisor. His academic and personal support made this thesis possible. I also want to thank thetwo other committee members, Professors Carson Woo and Dean Uyeno for their insightfulcomments and suggestions. Professor Izak Benbasat gave many valuable suggestions to theexperimental design and data analysis. Professor Andy Trice provided feedback on the domainof relevance ontology. Special thanks to fellow graduate students Becky Lau, Kai, Hao Zhao,and Jiye Mao for testing the system and commenting.Finally, I want to thank my wife Jennifer, who has given me tremendous emotionalsupport throughout the whole program. Without her encouragement and sacrifice, this paperwould not have been possible.ixINTRODUCTION1. INTRODUCTIONDatabase design is the process of developing database structures from userrequirements for data (McFadden and Hoffer, 1988). It usually includes requirements analysis,conceptual design and physical design. Among them, conceptual design is the most critical anddifficult. In this stage, the designer needs to represent the system structure and userrequirements in a set of conceptual data constructs (Date, 1990). The conceptual design isusually accomplished by a database design expert working together with a domain expert.With the advance of end user computing, some expert systems (Storey and Goldstein, 1988;Choobineh et al., 1988; Dogac et al., 1989) were developed to allow the end user, who can beregarded as a domain expert but a novice database designer, to design a conceptual datamodel independently.However, the results of studies on effectiveness of these expert systems and otherdecision support systems in general have been mixed (Sharda, et al., 1988; Gilbert, 1993).Some external factors are proposed to explain the results, e.g., differences in task, application,personality and experimental design. These external factors certainly to some extent affect theresearch results, but internal characteristics of the expert system should be examined as well.One explanation is that while these expert systems possess a high degree of expertisein the database design domain, they usually know very little about the application domain andthe ‘real world’ (Goldstein and Storey, 1991). This causes several problems. First, thecommunication between the user and the expert system is often impeded because the user hasto spend much time entering ‘trivial’ information. This leads to low efficiency and poor userINTRODUCTIONsatisfaction. Second, current expert systems completely rely on the user to conceptualizedomain structures. This often leads to inferior design quality because of the user’sinexperience with conceptual data constructs. A human expert, on the other hand, already hassome commonsense knowledge and does not need to ask the end user every detail regardingthe real system. He can also use his knowledge about the application domain to quicklydiscover underlying structures and match them to conceptual data constructs.To address these problems, Goldstein and Storey (1991) developed a CommonsenseBusiness Reasoner (CBR) for their database design expert system, the View Creation System(VCS). CBR can store commonsense business knowledge and reason about it. Although CBRprovides a framework for further studies in commonsense knowledge reasoning, it was notfblly implemented and not connected with the VCS. So far, no other commonsense knowledgemodule has been developed for any database design expert system. As a result, the usefulnessof commonsense knowledge in a database design expert system has not been studied.1.1 Objective ofthe ThesisThe objectives of this thesis are to develop a new VCS with a commonsenseknowledge module and to evaluate the effectiveness of this commonsense module. Theorganization of the commonsense knowledge base is based on Naive Semantics and domain ofrelevance ontology. This knowledge base was integrated into the original VCS. Theeffectiveness of the new system was evaluated in a lab experiment focusing on efficiency,quality of design and perceived ease of use.2CONCEPTUAL DATABASE DESIGN1.2 Outline ofthe ThesisThe next Chapter describes some basic concepts in conceptual database design andtwo of the most often used conceptual design models. The process model for conceptualdatabase design is also presented and adopted as the research framework. Chapter 3introduces expert systems developed for conceptual database design and their limitations.Chapter 4 describes the concept of the comnionsense knowledge and provides the theoreticalfoundation of the commonsense knowledge base developed in the subsequent Chapters. Alsoincluded is a discussion of domain of relevance ontology. Chapter 5 introduces the structureand implementation of the new VCS. Chapter 6 to Chapter 8 describe the experiment. Chapter6 proposes the research hypotheses. Chapter 7 discusses and justifies the researchmethodology chosen. The experimental design is also described in Chapter 7. Chapter 8reports data analysis results. The conclusion and future research suggestions are presented inChapter 9.3CONCEPTUAL DATABASE DESIGN2. CONCEPTUAL DATABASE DESIGN2.1 Overview ofConceptualDatabase DesignConceptual database design is the process of representing the information content inconstructs independently of the way in which the data is physically stored (Goldstein, 1985;McFadden and Hoffer, 1988; Elmasri and Navathe, 1989; Date, 1990). The purpose ofconceptual design is to understand the system’s basic elements, their relationships andconstraints. The input to a conceptual database design is the user requirement specificationand the result is a global data model.In the business domain, a conceptual data model represents the entities in anorganization and the relationships among them. One of the advantages of using a conceptualdata model is its independence of implementation consideration (Korth and Silberschatz,1986). This allows a conceptual model to be used in many application programs, and on theother hand, allows an application domain to be modeled using various conceptual datamodels. A conceptual model’s absence of implementation detail also makes it easier tounderstand and forms the basis for communication with end users.2.2 E-R ModelThe most widely used conceptual data model is the Entity-Relationship (E-R) model(Chen 1976, 1977). The E-R data model is based on the assumption that the real worldconsists of a set of basic objects and relationships among these objects.4CONCEPTUAL DATABASE DESIGNThere are three basic constructs in the E-R model: entity, relationship between entitiesand attribute. An entity is defined as “a thing which can be distinctly identified” (Chen, 1976).It is usually represented by a noun. For example, in a library database, book and borrower canbe treated as entities. A relationship is defined as “an association among entities” (Chen,1976). It corresponds to a verb in natural language. For example, borrow is a relationshipamong borrower and book in the above library database. An attribute is a kind of property ofthe entity or relationship it is related to. All the instances of an entity or relationship share itsattributes. For example, if employee number and name are defined as attributes for entityemployee, then all the employees would have an employee number and a name.One advantage of using the E-R model as a conceptual design tool is its simplicity andnaturalness (Brodie, 1984 and Konsynski, 1979). There are only three constructs in the modeland they can be easily understood by novice users. The E-R model provides a communicationmedium between the users and the designer.The E-R model was chosen as the basis for the VCS’s design approach for exactly thesame reason (Storey, 1988).2.3 RelationalModelWhile the E-R model is regarded as natural to the real world, the relational model(Codd, 1970) is the model used by most commercial products. The relational model representsthe database using a collection of tables (Codd, 1970; Korth and Silberschatz, 1986). Eachtable is called a relation and has a unique name and set of attributes. The relation is the onlyconstruct in the relational model.5CONCEPTUAL DATABASE DESIGNRelation models are based on relational algebra and relational calculus which givethem a solid theoretical foundation (Codd, 1970). Most of the current commercial databaseproducts use the relational model. The VCS, although using the E-R model as its design basis,generates a relational model in 3rd Normal Form (3NF) as its final result. This can be theneasily implemented using commercial products.2.4 ProcessModel ofConceptual Database DesignAs discussed above, although there has been a large amount of research done inbuilding the theoretical foundation of conceptual database models, relatively little is knownabout the process whereby a novice designer applies these concepts and rules. Because of thislack of research in the design process, the factors that might affect a novice designer’sperformance are not identified and the dynamics of the process are not studied. To study ifand why commonsense knowledge might help in the design process, the process itself needs tobe understood.Batra and Davis (1992) studied the verbal protocol and written trace of several expertsand novices designing a data base. Based on this data, they developed a three level processmodel of conceptual database design (Figure 1).6CONCEPTUAL DATABASE DESIGNThe specific subject activities at each level are the following:Enterprise Level During the enterprise level, the subject reads, contemplates therequirements, comments, elicits user requirements, seeks clarifications or establishesconnections. The focus at this level is on developing a reasonable understanding of theproblem domain.Recognition Level At this level, the subject focuses on some specific aspect of theuser requirements and tries to understand the sub-problem at hand, and this triggers theappropriate knowledge structures in his repertoire.Representation Level The representation phase constitutes the operationalization ofthe subject’s understanding of the structure into a conceptual data representation using aFigure 1 Process model ofconceptual database design (Bafra andDavis, 1992)7CONCEPTUAL DATABASE DESIGNspecific data model. This phase also includes the verification of the representation to ensure itsatisfies the user requirement.It was found that a subject would typically stay at one level for some time beforemoving on to another. However, there were substantive differences in the pattern.The three level process model, building on previous work on information processtheory and empirically tested, is adopted as the research framework for the experiment on theeffectiveness of commonsense knowledge module.2.5 Design Process ofExpert vs. NoviceBatra and Davis (1992) found human experts use commonsense knowledge in all thethree levels. At the enterprise level, experts start using commonsense to understand basicobjects and relations in the problem domain. Commonsense knowledge at this level helps anexpert quickly establish a general picture of the application domain. For example, when anexpert is told that he will be designing a database for a school, he will immediately know thatthere are basic objects like teachers, students, courses and so on.The recognition phase sees the biggest difference between experts and novices.Experts tend to spend much less time at this stage than novices. During this phase, experts usecommonsense knowledge to help identifjing entities and attributes. For example, ‘tudenttakes course” is a commonsense fact which should be modeled as a relationship because bothstudent and course are entities and the mapping ratio between the two entities are many tomany. An expert who has this commonsense knowledge can match it with the patterns in hisdatabase design knowledge base and represent it accordingly. An expert system without this8CONCEPTUAL DATABASE DESIGNcommonsense fact will rely on the user to identify both ‘student’ and ‘course’ are entities andthere is a relationship between them.It is often observed at the recognition stage, that an expert starts with a fuzzy pictureof the real world obtained from the previous stage and uses his commonsense knowledge tomap it to a data structure. Then the expert will actively elicit information from the user to tryto verify and complete the selected data structure. Commonsense knowledge and databasedesign knowledge together guide the expert during the recognition phase. For a novicedesigner, although he has commonsense knowledge for the problem domain, the lack ofdatabase design expertise makes it hard for him to use it.At the last stage, the designer generates the data constructs identified in the previouslevel and ensures that they satisfy the syntax and constraints of the data model.Based on the above findings, it is argued that the recognition phase is the most difficultphase and where commonsense knowledge can play an important role. For an expert systemto be as effective as a real expert designer, it must be able to provide both strong databasedesign knowledge and commonsense knowledge support to the user.9EXPERT SYSTEMS FOR CONCEPTUAL DATABASE DESIGN3. EXPERT SYSTEMS FOR CONCEPTUALDATABASE DESIGN3.1 OverviewBecause of the complexity of conceptual design, a lot of effort has been put into thestudy of the conceptual design process and especially ways to improve the design result ofcomputer-novice end-users. One major approach is the development of expert systems thatautomate the design process or help the designer in the design process (Storey and Goldstein,1988; Choobineh Ct al., 1988; Dogac et al., 1989).3.2 The View Creation System (VCS)The VCS is a conceptual database design expert system developed by Storey andGoldstein (1988). It is based on a formalized methodology for generating and integrating userviews. It uses the E-R model discussed in Section 2.2 as the view modeling tool. The designprocess with the VCS system can be divided into two stages:3.2.1 Creation of the Initial E-R ModelIn this stage, the user expresses his view of the database in terms of entities andrelationships. For each entity, the user enters its name, attributes, and keys, and for eachrelationship, its cardinality and attributes.3.2.2 Creation of the Final Relational ModelIn this stage, the system checks the user’s initial E-R model, identifies problems, seeksmore information to fix the problems and converts it into a relational model.I0EXPERT SYSTEMS FOR CONCEPTUAL DATABASE DESIGNAccording to Batra and Davis’ process model (1992), the VCS facilitates the designprocess basically at the representation level. Instead of directly representing the systemstructure using a relational model, which is a less direct representation of the real world, theuser can start with the more natural E-R model and later let the VCS convert it to a relationalone. However, the user has to capture and model the real system before he can represent it inan E-R model. So the enterprise and recognition level processes are essentially not supportedin the original VCS.3.3 Problems with the VCSGilbert (1993) conducted a lab experiment comparing users’ performance workingwith the VCS and human experts. Several problems of the VCS were identified. Among them,the lack of commonsense knowledge and poor user interface are the most important.3.3.1 Lack of Commonsense KnowledgeAs discussed in Section 2.5 about the difference between experts and novices, expertsuses commonsense knowledge in the enterprise and recognition phases to help understand theuser requirements and recognize data constructs. But so far, no expert system has usedcommonsense knowledge in the design process. The lack of commonsense reasoning ability inexpert systems in general has been criticized by many researchers (Kolata, 1982; McCarthy,1983; Meltzer, 1985; Buchanan, 1988). The VCS also suffers from this problem (Goldsteinand Storey, 1991; Gilbert, 1993). It has rules and facts about database design process butnothing about the application domain. This creates an obstacle in the communication between11EXPERT SYSTEMS FOR CONCEPTUAL DATABASE DESIGNthe user and the system. The user has to routinely enter ‘trivial” information and answer“obvious” questions. The system’s image as an expert is damaged as a consequence.3.3.2 Poor User InterfaceThe user interface is a another major factor affecting the end-user’s performance onthe VCS (Gilbert, 1993). The input of data is not direct and intuitive. For example, to enter anew entity or relationship into the system, the user first goes to the main ‘Entity andRelationship” screen where all the entities and relationship are displayed. Then instead oftyping directly, the user has to press [iNSERT] key first to open a separate input box andenter the new entry there. When finished, he has to press [ESC] to go back to the main menu.Procedures like this make it hard for novice users to remember and often discourage themfrom learning how to use it at all.To address above problems, especially the lack of commonsense knowledge, variouseffort was made. This will be introduced in the following Chapters.12COMMONSENSE KNOWLEDGE4. COMMONSENSE KNOWLEDGE4.1 OverviewThe idea of incorporating commonsense knowledge into computer programs has along history (McCarthy and Hayes, 1969; Hobbs and Moore, 1985; Lenat et al., 1986).There has been no common definition for commonsense knowledge. Dahigren (1989)describes commonsense knowledge as ‘a set of naive belief; at times vague and inaccurate,about the way the world is structured’. Ein-Dor (1985) defines commonsense as ‘vhat anyparticipant in a culture expects any other participant in that culture to know when meeting forthe first time and before any exchange takes place between them” Goldstein and Storey(1991) define commonsense as ‘general knowledge about the way the world works’. It can besummarized that commonsense knowledge is eveiyday knowledge everybody knows about theworld.Because of the popularity of expert systems in recent years, there has been a lotattention paid to expert knowledge. Commonsense knowledge, although considered importantto overcome the brittleness of these systems (Winograd and Flores, 1987), remains largelyuntouched. A comparison between commonsense knowledge and expert knowledge willillustrate why it is so hard to model commonsense knowledge.13COMMONSENSE KNOWLEDGE4.2 Commonsense vs. Expert KnowledgeAlthough they are both important components of a human being’s knowledge of theworld, commonsense knowledge and expert knowledge have some distinguishingcharacteristics.4.2.1 StructureThe accumulation process of commonsense knowledge is not structured and oftenunintentional. It is gathered through one person’s personal interaction with the environmentaround him. This interaction is constant, informal, and incomplete. Usually no effort is madeto analyze and organize the information gathered and memorize it as formal knowledge.Expert knowledge, on the other hand, is developed through intentional cognitiveprocess (Winograd and Flores, 1988). An expert learns directly from his own experienceand/or indirectly from other people’s knowledge. During the learning process, all of the inputinformation is processed and used to enhance or modify the existing knowledge framework.Expert knowledge is processed in an effort to better understand and form a clear view of thesubject. Therefore, expert knowledge has formal structures lacked by commonsenseknowledge.This unstructured nature makes commonsense knowledge very difficult to representusing formal language.4.2.2 GeneralityCommonsense knowledge comes from people’s everyday life and is easily shared by alarge number of people through communication.14COMMONSENSE KNOWLEDGEOn the other hand, an expert usually has only in-depth knowledge about his owndomain. Because the effort required to acquire expert knowledge is so great, it isunderstandable that expert knowledge is not shared by most people.4.2.3 AvailabilityBecause commonsense knowledge is obtained from everyday life, it is not confined toa small number of experts. It is readily available from ordinary people. In some sense,everybody can be treated as an expert on commonsense knowledge.4.2.4 UsabilityCommonsense knowledge is used in almost any place where human thinking isrequired. Analogy is used to bridge the more general commonsense knowledge and thespecific question at hands.Expert knowledge is usually generated to solve problems in a specific domain. Thisfocus limits its use. Furthermore, the representation of the expert knowledge in most expertsystems is domain dependent, which makes the transferring of knowledge across domainseven harder.4.3 Naive SemanticsBecause of the above characteristics of commonsense knowledge, it is understandablethat so far there is little success in the Al community in developing a simple, elegant formalismto represent it. One of the most ambitious project in this area is the CYC project (Lenat et al.,1986; Lenat et al., 1990; Guha and Lenat, 1990). They are trying to build a comprehensivecominonsense knowledge base that can be accessed by various expert systems. Although the15COMMONSENSE KNOWLEDGECYC project may eventually accumulate enough knowledge and provide general knowledgesupport to domain specific expert systems, the feasibility and the cost effectiveness of thus asystem can not be proved.Naive Semantics (Dahlgren, 1989) provides a much more concise and implementableframework. The most important feature of Naive Semantics is that it represents wordmeanings as commonsense knowledge and not sets of primitives. In traditional AT research,word meaning has been treated as critical. For example, if ‘flying’ is defined as an attribute for‘bird’, then all the creatures that can not fly would be automatically disqualified. NaiveSemantics, representing word meaning in commonsense, treats ‘flying’ as a typical rathercritical description of ‘bird’. In other word, Naive Semantic representations are probabilisticand non-monotonic.Because Naive Semantics tries to include all commonsense knowledge into eachword’s meaning, the number of features needed is theoretically limitless. To keep the size ofthe knowledge base manageable, Dahlgren (1989) developed a Kind Type system which usesan ontology to organize and represent commonsense knowledge. A set of 54 feature types ofnouns and verbs is compiled based on psycholinguistic experiments in prototype theory(Rosch et al., 1976; Asheraft, 1976; Fehr and Russell, 1984; Dahigren, 1985). NewSelectorSystem, an expert system built by Lord and Dahlgren (1990) stores business and financialknowledge using the Kind Type system.16COMMONSENSE KNOWLEDGE4.4 Commonsense Business ReasonerBased on Naive Semantics theory and ontology, Goldstein and Storey developed aCommonsense Business Reasoner (CBR) (1991) as a framework to capture commonsenseknowledge in the business domain to facilitate conceptual database design.In the CBR, two distinct types of knowledge are represented: classificatory andgeneric. The first type provides meanings of words and the second type relationships amongobjects. The ontology from Naive Semantics is used to provide a framework for the genericcommonsense knowledge stored.The 54 feature types proposed by Dahigren (1989) provide a comprehensive list forcommonsense knowledge encoding. Obviously, not all the feature types can be populated andthere are some additional feature types specific to database design need to be added. Some ofthe most important feature types used in the CBR are discussed here.4.4.1 Has AttributeAttributes have a very important role in E-R models as they actually store theinformation about entities and relationships. Entering attributes for each entity is a routine taskwhen using the VCS. The ‘Has Attribute’ feature type is created to store common attributesfor a given word, saving input time and improving the human computer communication.For example, the commonsense knowledge base may indicate that ‘person hasattributes name and address’17COMMONSENSE KNOWLEDGE4.4.2 Has KeyDetermining the key for an entity sometimes is obvious, other times tricky. It would bevery usefhl if the commonsense knowledge base can store some possible candidate keys andprompt the user when needed. As other pieces of commonsense knowledge, the key stored inthe ‘has key’ feature type may not be the best key in a particular case, but it can still help theuser by providing an example of a common key.One example of the ‘Has Key’ feature type is that a person’s Social Insurance Number(SIN) is a key.4.4.3 SynonymSynonym is included to reflect the similarity between different expressions. Forexample, a ‘teacher’ may be called ‘instructor’, ‘professor’ or ‘lecturer’. These differentphrases share most attributes and commonsense knowledge about one of them often apply toothers as well.4.4.4 Ako (A Kind Of)Inheritances exist in everyday life and in computer languages. It’s also an importantconcept in database design. Feature type ‘Ako’ captures inheritance by specifying that oneentity is a sub-class of another entity.4.4.5 SubjectVerbs usually represent relationships in the E-R model. A relationship in the VCSinvolves two entities. The first entity is the subject of the relationship and the second entity isthe object of the relationship. To store this information, two feature types are used for verb in18COMMONSENSE KNOWLEDGEthe CBR. Among them, the subject feature type is included to store the information that aword is the subject in a relationship.For example, ‘university’ is the subject of relationship ‘university offers course’.4.4.6 ObjectThe object feature type is included to store the information that a word is the object ina relationship.For example, ‘course’ is the object of relationship ‘university offers course’.Using the subject and object feature types, both parties involved in a relationship areidentified.4.5 Domain ofRelevancefor Commonsense KnowledgeWhile Naive Semantics helps to provide a framework for storing commonsense, theontology is still limited in that it provides only a rigid classification of terms. The majorweakness of this approach is that the “relevance” of facts is not accounted for. Because aperson learns commonsense knowledge of the world from his experience, the context and therelevance of the facts he observes become part of the knowledge. To fhlly capture hiscommonsense knowledge, the relevance of a fact must be ‘tagged’ to the fact. For example,the fact ‘providing pension plan’ may be related to the term ‘organization’. However, this onlyapplies to certain industries, or in certain countries. Although Naive Semantics’ ontologyattempted to provide a useful tool in organizing people’s commonsense, the application islimited in terms of its ability to fully capture the relevance of facts.19COMMONSENSE KNOWLEDGEBy examining the relevance of terms, it is possible to define “dimensions of relevance”that are pertinent for enhancing the original ontology in the database design. For example,“sales tax” may be different in terms of relevance according to differences in political body. Insome states or provinces, no sales tax is collected, but in some other places sales tax iscollected. If the data base design system has this domain of relevance knowledge stored, it willprompt the user about this commonsense knowledge only when it is relevant. Obviously thiswill make the design process more natural and efficient.Before we identifS’ possible dimensions for the domain of relevance knowledge, theorganization of this knowledge needs to be defined.4.5.1 Organization of the Domain of Relevance KnowledgeThe domain of relevance knowledge is different from normal commonsense knowledgein a sense that it is a kind of meta knowledge, i.e. knowledge about knowledge. The NaiveSemantics’ ontology introduced in Section 4.3 was created only to store conimonsenseknowledge itself A new ontology is required to store domain of relevance knowledge. Onenatural way to organize the new ontology is to divide it along several relevance dimensions.For example, geographic, industry, organization and application are frequently used to definethe relevance of a term, and can be treated as relevance dimensions. Under each of thesedimension, there are many nodes representing sub-classes in that dimension. For example,under the geographic dimension, there are Canada, US, Mexico sub-classes. Under Canada,there are different provinces, etc.20COMMONSENSE KNOWLEDGEAfter the framework of the ontology has been set up, the actual relevance knowledgecan be put it. The facts in the original ontology will be attached to different nodes in thesedimensions in the new ontology based on their relevance. For example, the fact ‘GST’ will beattached only to node Canada in the geographic dimension, and to all industries except charityindustry node in the industry dimension. For other dimensions that are not relevant to this fact,it will simply not be attached.Using this structure, the domain of relevance information about commonsenseknowledge is stored completely. There are two ways this information can be used. First, givena specific domain, all the relevant commonsense knowledge can be retrieved. This is donethrough searching all the dimensions for the given domain and recording the facts attached tothem. If a fact is attached to all of the dimensions, then it is relevant in the given domain.Secondly, for a particular fact, its domain of relevance can be easily identified by searching allthe dimensions for the attachment of this fact and the product of these domains is the result.4.5.2 Dimensions of RelevanceGoldstein and Storey (1991) identified some contexts such as functions andorganizations. The organization context can be further divided into industry, organization anduser. Following this line of argument, more dimensions of relevance are proposed. Thesedimensions are part of the foundation upon which we construct our ontology of domain ofrelevance.21COMMONSENSE KNOWLEDGEThese dimensions are grouped into several categories. It should be noted that theseclassifications, like other commonsense knowledge, are somehow vague and insufficient. Inaddition, because of the parallel existing in business world, cross-classification is used.We categorize all the dimensions into three groups, ENVIRONMENT, BUSINESSand PEOPLE. They correspond to three levels of domain of relevance.4.5.2.1 Relevance ofEnvironmentBesides the relevance related to a business itself; the economy and geographyenvironment where the business exists also provide relevant common sense information.4.5.2.1.1 EconomyThere are still centralized economies where most of the planning and controls are doneby a central authority. The mechanism is so different that many economic terms such as price,tax and profit have different meaning. For example, the price of a product will be set by theauthority instead of the market forces. Much of the business commonsense knowledge in afree market economy will not apply.Dimension Stage ofDevelopment (Developed Developing)The stage of development also affects the relevance of knowledge. For example, indeveloping countries, the fact “using credit card” may be not relevant.Dimension Type (Centrally Planne4 Free Market)22COMMONSENSE KNOWLEDGE4.5.2.1.2 MarketMarkets can be classified into monopoly, oligopoly or competition markets. Thoughthe real world doesn’t fit in perfectly, this classification still reflects some distinct facts aboutmarkets. For example, the price is determined by the monopoly and the profit is higher in amonopoly market, while the price is more determined by the supply and demand in acompetition market. Considering that the government may step in and introduce regulations,the regulated market was added.Dimension Type ofMarket (Regulated Monopoly, Oligopoly, Competition)Market can be divided by their scopes, too. There are global market, domestic marketand regional market.Dimension Scope ofMarket (Global Domestic, Regional)4.5.2.1.3 Political BodyDifferent regions have different regulations regarding business activities. This can bereflected in tax, minimal wage, pension, etc. Examples of political bodies are:Dimension Political Body (Canada (Provinces), US (States))There is no need to start with a complete list of political bodies in the world. Jhstead,they can be added to the ontology whenever needed.4.5.2.1.4 GeographyThe geography environment is less important today than before. However, it still hasstrong direct or indirect impact on a business. Some of the geography dimensions are listedbelow.23COMMONSENSE KNOWLEDGEDimension Topography (Coastal Mountain, Prairie, Deseri, Metropolitan, Town,Countryside, West coast, East coast)Dimension Orientation (North, South, Easfl West)Dimension Continent (America, Asia, Africa, Europe)4.5.2.2 Relevance ofBusinessThe first category of dimensions deals directly with the particular business. It includesindustry, company and product.4.5.2.2.1 IndustryOne of the most obvious and important dimensions identified in this category isindustry. There are many different ways to classif,’ industries (Krahn and Lowe, Mitchell,1987; 1988; Reinecke and Schoell, 1977). However, they tend going into detail instead ofcapturing the general knowledge about industries.One of the intention of our study is to build a frame that is general and flexible enoughto facilitate further expansion. After carefully examining several industries, we identifj thedivision of service and manufacturing industries as the top nodes in the industry dimension.These two kinds of industries are different in several ways. While service industries provideservices that are usually non-physical processes, manufacturing industries produce products inboth physical and non-physical forms. Service industries usually don’t need raw materials, butmanufacturing industries will need raw material to be transformed to the products. These twokinds of industries may also pay different taxes.24COMMONSENSE KNOWLEDGEService industries can be divided into transportation, communication, information,medical, tourism, insurance, banking and accounting. They can be further divided whennecessary. The structure of the industry domain is listed below:Dimension Industiy (Service (Transportation (Airline, Shipping, ...),Communication(Mail Telephone, ...), Information, Medical, Tourism, Insurance, Banking(Accounting, ...), Manufacturing (Automobile, Computer, ...))The lower level details are not fully defined because they are less important and can beadded whenever that particular industry in used.4.5.2.2.2 CompanyAnother important aspect of relevance of business is company. Companies can becategorized by the ownership into sole proprietor, partnership and corporation. Corporationscan be further divided into private and public owned corporations. These companies aredifferent in terms of their employees and owners. Sole proprietor, partnership companies’owners usually are the managers themselves. A corporation will pay tax on its profit while asole proprietor or partnership company doesn’t need to pay.Dimension Ownership (Sole Proprietor, Partnership, Corporation Privateownea Public owned)Corporations can be distinguished by whether they sell stock to the public. Thisdimension will be useful in deciding the control in the company.Dimension Stock (Open, Close)25COMMONSENSE KNOWLEDGECompanies can also be classified by their goals. Profit and non-profit companies differin their products, employees, taxes and profits.Dimension Goal (Prof14 Non-Profit)One interesting dimension about the company may be the existence of trade union. In acompany with strong trade union, the employees wages may be higher, so the cost ofoperation will be higher.Dimension Union Existence (Unionize4 Non-unionized)4.5.2.2.3 Department/DivisionInside almost every companies, there are several different departments or divisions thathave different but related functions. They played different roles in the operation of thecompany. Although we identified this as an dimension, we will not provide any value or sub-dimension because of the variety across companies. As in Section 4.1.1, they can be addedincrementally.Dimension Department/Division4.5.2.2.4 Functional AreaDifferent functional areas have different terminology and different processes. They areusually clearly defined.Dimension FunctionalArea (Finance, Accounting, Marketing, Personnel)26COMMONSENSE KNOWLEDGE4.5.2.2.5 ProductProducts differ greatly across industries, but they still can be roughly divided by theirusers: a product is either used by consumers directly or it may be input of another industry.Products also may different in their taxation, some products like food are tax free, andothers not. Products imported and exported are subject to duties and quota.Dimension Product (End-user (Food Clothe ...), Intermediate Product (Miningproduct, ...)4.5.2.3 Relevance ofPeoplePeople play a central role in business. There are many dimensions upon which peoplecan be classified. We will only select those relevant to business commonsense.4.5.2.3.1 OccupationA person’s occupation is an important aspect of his characteristics. For example,professionals are usually paid on a salary basis, while unskilled workers are usually paidhourly. Maria Hirszowicz (1981) roughly categorized occupations into four groups:Dimension Occupation (White-collar (Professional and technical Managers andproprietors Clerica Sales), Blue-collar (Skilled worker.s Semi-skilled workers Unskilledworkers), Service workers, Farm workers)Obviously, one person’s role, function and position in a business setting are closelyrelated with his occupation. The domain of relevance information will help to construct themodel of the business from human resource’s perspective.27COMMONSENSE KNOWLEDGEThere are many more detail defined occupations (Reinecke and Schoell, 1977; Krahnand Lowe, 1988; Mitchell, 1987). However, as mentioned before, we are only interested inidentiing a frame which will be the basis of further expansion.4.5.2.3.2 Personal AttributesPeople’s personal attributes differ greatly and the facts in the ontology may only applyto certain kinds of people. Following are some personal attributes that representing differentdimensions.Dimension Sex (Male, Female)Dimension Age (Chila’ Adult, Senior)Dimension Marriage Status (Single, Married Divorced)Among above dimensions, the gender difference exists in most of today’s business interms of salary, occupation, etc. Age affects a person’s role in the business and the societyand marriage status has direct impact on taxes.4.5.2.3.3 Personal BackgroundLike personal attributes, personal background affects a person’s position in a business,but its effect is more indirect.Dimension Education (high school college, university)Dimension Income (below average, average, above average)28COMMONSENSE KNOWLEDGE4.5.3 Some ExamplesIn the following, some business terms is examined according to the above dimensionsto show the usefhlness of the domain of relevance knowledge.Tax Taxes are relevant to most of the dimensions identified earlier in this Section. Acompany making profit must pay taxes, and taxes are different from one place to anotherplace. Some countries don’t collect taxes while other do. People with different income will paydifferent taxes.GST GST’s relevance can be first easily identified in the POLITICAL BODY domain,i.e., only Canada collects GST. It is also has relevancy to most of the products except food inthe PRODUCT domain.Pension Pension’s relevance can mainly be identified by INDUSTRY,OCCUPATION, POLITICAL BODY and UNION.Interest Interest is usually the same across INDUSTRIES and COMPANIES, butdifferent in different countries. In different types of ECONOMY, knowledge about interestwill be different, too.Salary Salary varies greatly in almost all the dimensions in term of amount and form.In some iNDUSTRIES and OCCUPATIONS, salary is more hourly calculated. The amountof salary will also be different across COMPANY, POLITICAL BODY, ECONOMY andPEOPLE.29COMMONSENSE KNOWLEDGESo far, a new framework has been developed to store both commonsense knowledgeand its domain of relevant information. In the next Chapter, we will see how a new VCS isbuild on the framework developed here.30THE NEW VCS5. THE NEWVCSThe new VCS was developed aimed at fixing the problems of the original VCS. Itincorporates the commonsense knowledge base described in the previous Chapter. Majorchanges are also done to improve the user interface.5.1 Architecture ofthe New VCSBecause both the original VCS and the CBR module are relatively self containedsystems, it was decided that some run-time message passing mechanism would be establishedbetween the two systems instead ofmaking structural changes to either of them.Figure 2 shows the structure of the new VCS enhanced with CBR. At present, thecontrol flow is from the VCS to the CBR and the information flow from the CBR back to theVCS. The VCS sends a request to the CBR module requesting commonsense knowledge onone or a group of terms. After the CBR module receives this request, it searches in theconmionsense knowledge base and return all the relevant information it finds./5.2 Consultation ofCommonsense During Design ProcessRequestViewCreationSystem(VCS)CommonsenseBusinessReasoner(CBR)ResultFigure 2 Structure ofthe new VCS3’THE NEW VCS5.2.1 Stage OneAs discussed in Chapter Two, the recognition phase is the most difficult part for anovice user (Batra and Davis, 1992). To support the user in this stage, the commonsenseknowledge base stores some of the most important and universal structures in the applicationdomain and their mappings into conceptual data constructs. By accessing the commonsenseknowledge base, the user can quickly identify the underlying structure and modify thesolutions provided by the system to fit the task’s specific requirements.Commonsense knowledge can also be used in the enterprise and representation phasesby providing relevant information. When the user reads a term from the case or enters anentity or relationship into VCS, he can access the commonsense knowledge base throughCBR to get related facts. This will facilitate the communication between the user and thesystem because it reduces the need to enter trivial information and makes the system look‘smart’. The user’s attitude towards the system is also likely to be changed. The reasoningprocess in these two stages is largely based on commonsense knowledge stored.Commonsense reasoning in this context can be defined as the process of retrievingrelevant commonsense knowledge based on a given term. ‘Relevant’ means the knowledgeretrieved should be related to the term in the database design context. A term can be the nameof either an entity or a relationship. Obviously, the simplest reasoning process in the CBR isretrieving the attributes and keys given entity or relationship names. For example, it is often atedious task for the user to add the attribute ‘name’ to every ‘person’ related entity. Withcommonsense reasoning ability, the system can automatically add ‘name’ as an attribute32THE NEW VCSwhenever the user creates such an entity. This reasoning pattern can be incorporated into allother reasoning processes.All the entities must be related by relationships, and all the relationships involveentities. So the next reasoning process is retrieving relevant relationships for the entities andrelevant entities for the relationships. This will further reduce the user input needed. By nowthe user only needs to enter a name for his entities or relationships. Then the system willsearch through the commonsense knowledge base and add attributes and related relationshipsand entities.5.2.2 Stage TwoIn this stage the E-R model is translated into a relational model. Commonsenseknowledge plays a less important role here because most of the knowledge used in this stageis database design expert knowledge. However, conimonsense knowledge still can help toprovide missing attributes and keys and identifS’ multiple values of entities. For example, theuser doesn’t need to tell the system that a person cannot have multiple SIN’s.5.3 Other ImprovementsBesides lack of commonsense knowledge, the original VCS also suffers from poorinterface design (Gilbert, 1993). Users found it unintuitive and incompatible with othersoftware on the market. To identify interface problems for the original VCS, a similar pilotwas done at UBC. The user was trained to use the VCS. All of the users in the testing haddifficulties with the interface and some people even could not complete the design process. Itwas observed that the interface’s effect is so big, it may confound the result of an experiment33THE NEW VCSon the commonsense knowledge effect. A great amount of work was done to improve theuser interface for the VCS.5.3.1 Direct Input MethodThe ‘Entity and Relationship’ box used to be a display-only screen, which means thesystem displays current entities and relationships in the box but the user can not changeanything in the box. It is now converted to a fully functional editing box. This means the usercan enter, delete and modilj any entity or relationship directly in one box. This saves thetrouble of pressing [iNSERT] to go to the edit box every time a new entry is entered.5.3.2 Smart Cursor PositioningWhen the user wants to enter a new entity, he first enters the name of the entity in the‘Entity and Relationship’ box, and then presses [ENTER] to invoke the detailed entity screento enter attributes and keys for the entity. In the original VCS, when the user got into theentity screen, the cursor stopped at the entity name field, the first field on the screen. The userthen had to press [TAB] to go to the attributes field. In the new VCS, when the entity screenis first displayed, the cursor is automatically positioned in the attributes field, and the user cantype immediately.Similarly, when the VCS asks for a missing key from the user, the cursor is nowpositioned at the key attributes field directly.5.3.3 Context Sensitive MessageThere are three kinds of message in the system. All of them have been made contextsensitive.34THE NEW VCS5.3.3.1 Help MessageIt can be accessed by pressing [F 1]. This feature was available in the original VCS, butthe help message contained only the usage of function keys. The new help message providesexplanations of the choices available, depended on the current procedure.5.3.3.2 System MessageThere is always a system message box on the screen. While the user spends most ofthe time in the entity and relationship screen, the original VCS did not display the attributes,key and mapping ratio of the entries on the screen. The user had to constantly go into thedetailed entity or relationship screen to check this information. In the new VCS, the systemmessage box is used to display the complete information about the current cursor entry.5.3.3.3 Dialogue MessageThe dialogue message displayed during the solving session now contains contextsensitive information retrieved from the commonsense knowledge base.5.3.4 Configurable FeaturesTo allow maximum flexibility, all the new features and some of the old features weremade configurable through a configuration menu. The user can define whether to consult thecommonsense knowledge base, the reasoning level, whether to check for problems aftermodification. This allows the empirical study of different features of the system to be easilycarried out.35HYPOTNESIS DEVELOPMENT6. HYPOTHESIS DEVELOPMENT6.1 The Research Model6.1.1 Dependent VariablesEvaluation is an integral part of the system development cycle. As a result, systemeffectiveness has been an important issue over the years. Effectiveness of an informationsystem is generally defined as the extent to which the system accomplishes its objectives(Hamilton and Chervany, 1981). Therefore, assessing system effectiveness is first to determinethe objectives of the system, then to develop measures to evaluate how well the objectiveshave been achieved. However, different systems tend to have different objectives. As Melone(1990) suggests, “the bases for identifying criteria of effectiveness are many and diverse”.Melone (1990) divides effectiveness into an output-oriented component and an affect-oriented component. The output-oriented component emphasizes system performance interms of the quantity and quality of outputs and the efficiency of the process. This line ofmeasurement was widely used in system effectiveness studies (Hamilton and Chervany, 1981).User perceptions are now widely used as a surrogate for system effectiveness (Melone, 1990).In this experiment, the goal is to test the effectiveness of the commonsense knowledgebase in the new VCS system. Both output-oriented and affect-oriented measurements areneeded to flilly evaluate the changes of effectiveness caused by adding CBR. The intendedrole of the VCS is to support the conceptual design process of the end user. When interactingwith the user, the VCS improves productivity by producing better conceptual designs andreducing the user’s information processing requirement. The commonsense knowledge36HYPOTHESIS DEVELOPMENTmodule further enhances VCS’s effectiveness by providing commonsense knowledge neededin the design process. Based on this observation, user performance was chosen as thedependent variable. This is supported by past studies on system effectiveness (Gilbert, 1993).To filly measure the effect of the independent variables, the construct of user performancewas operationalized by three variables -- the time taken to finish the task, the quality of thedesign and the perceived ease of use.6.1.2 Independent VariablesAs discussed in Chapter 2, the major constructs in conceptual database design areentities, relationships and attributes. The key task in the design process then is to identifythese constructs from the user requirements and map them to the formal modeling language.This is defined as the recognition level in Batra and Davis’ process model (1992). Because ofthe lack of experience of the end user, the design is often error prone and time-consuming.Although many researchers have criticized the lack of commonsense knowledge of existingexpert systems (McCarthy and Hayes, 1969; Hobbs and Moore, 1985; Lenat et al., 1986;Goldstein and Storey, 1991) and many empirical studies have been done to investigate userperformance using different data models (Hoffer, 1990), no study has ever addressed the rolecommonsense knowledge plays in the design process. The development of the new VCS withcommonsense knowledge represents a concrete step towards more intelligent expert systems.The experiment was designed mainly to test the effectiveness of such effort, so the existenceof commonsense knowledge is treated as an independent variable.37HYPOTHESIS DEVELOPMENTThe complexity of the design task also has a direct impact on the performance of thedesigner (Batra and Antony, 1992a; Gilbert, 1993). Studying the effect of task complexity willprovide valuable information on the generality of an expert system. If the system onlyproduces good results for simple tasks, then the usefi.ilness of the system in a real worldsetting will be seriously limited. On the other hand, if the system performs consistently wellunder both easy and difficult tasks in the test, it can be expected to deal with a wide range ofproblems. Therefore, the complexity of the task was treated as another independent variable.However, the relevance of task complexity does not stop here. It can also be apotential moderating variable for the relationship between the existence of commonsenseknowledge and user performance. The interaction between the adding of commonsenseknowledge and task complexity would be a valuable source of information when designingfI.iture expert systems. If it is proved that commonsense knowledge only helps in morecomplex tasks then the system should take this into account and try to consult commonsenseknowledge only when complicated structures are modeled or the size of the model exceedscertain limit.The research model used in this study is illustrated in Figure 3.38HYPOTHESIS DEVELOPMENTTaskComplexityUser PerformanceTimeQualityPerceived EOUCommonsenseKnowledgeFigure 3 Research model6.2 Commonsense Knowledge EffectThe commonsense knowledge base incorporated into the VCS contains commonsenseknowledge about the application domain expressed in entities, relationships and attributes. Byentering these constructs for the user directly instead of letting the user discover them himself;the new VCS facilitates the recognition of not only specific constructs, but also theunderstanding of the entire problem domain.Therefore, the following hypotheses are proposed (stated in null form):Hi There will be no dfference between using the VCS with the commonsense knowledgebase and using the VCS without the commonsense knowledge base in terms of time takentofinish the task.H2 There will be no dfference between using the VCS with the commonsense knowledgebase and using the VCS without the commonsense knowledge base in terms of the qualityof the design.39HYPOTHESIS DEVELOPMENTH3 There will be no dqference between using the VCS with the commonsense knowledgebase and using the VCS without the commonsense knowledge base in terms of theperceived ease of use.It is predicted that all the above hypotheses will be rejected.6.3 Task Complexiiv EffectTraditionally, there are two measurements of the complexity of a design task: the sizeof the problem and the nature of the relationships among entities (Date, 1990). The size of theproblem is usually represented by the number of entities and relationships in the model. Themore entities and relationships a model has, the more complex it is. Relationships as animportant factor in task complexity is confirmed by the empirical study by Gilbert (1993). Itshows that subjects had great difficulties modeling unary and category relationships, whilethey had little trouble with binary relationships.This leads to the following hypotheses (stated in null form):H4 There will be no dfference between using the VCSfor a complex task and using the VCSfor an easy task in terms of time taken tofinish the task.H5 There will be no dfference between using the VCSfor a complex task and using the VCSfor an easy task in terms of the quality of the design.H6 There will be no d7erence between using the VCSfor a complex task and using the VCSfor an easy task in terms of the perceived ease of use.It is predicted that all the above hypotheses will be rejected.40HYPOTHESIS DEVELOPMENT6.4 Interaction EffectIn the easy task, there are three entities and relationships and all of the relationshipsare simple binary relationships. In the complex task, there are seven entities and relationshipsand some are unary and category relationships. As described in Chapter 4, commonsenseknowledge is general knowledge about the world. Compared with expert knowledge, it is‘shallow’ and ‘common’. In designing the experiment, the amount of the commonsensesupport in the two tasks is made the same, i.e., same number of entities and relationships areprompted to the subject in each task. For the easy task, commonsense knowledge covers alarge portion of the design. But for the complex task, commonsense knowledge only helps inthe easier portion, leaving difficulties like unary and category relationships completely to thesubject.Therefore, it is expected that when the task becomes more and more complex,commonsense knowledge’s effect will become less obvious.This discussion suggests the following hypothesis (stated in null form):H7 There will be no interaction effect between the existence of commonsense knowledge andthe complexity of the task.will be rejected.41RESEARCH METHODOLOGY7. RESEARCH METHODOLOGYAs stated before, the main objective of this study is to test if the VCS’s effectivenesswill be affected by the existence of commonsense knowledge or task complexity. However,there are many other features besides these two factors that may affect the user’sperformance. Examples are subject’s understanding of the task, system design concepts andfamiliarity of the VCS. To find the real relationship between the independent variables anddependent variables, other factors’ effects must be ruled out. Because of this focus on causalrelationship, a laboratory experiment was selected as the research method (Emory andCooper, 1991). Benbasat (1989) states ‘high internal validity is one of the major advantages oflaboratory experiments’. By using a laboratory experiment, various kind of threats to theinternal validity can be reduced or avoided. Also more information can be captured in acontrolled manner by the experimenter.Z 1 Experimental DesignA 2x2 factorial design was used with existence of commonsense knowledge and thecomplexity of the task as the two factors. There are four treatment groups: the group workingon a complex task using the VCS with commonsense knowledge; the group working on aneasy task using the VCS with commonsense knowledge; the group working on a complex taskusing the VCS without commonsense knowledge; the group working on an easy task usingthe VCS without commonsense knowledge. This design allows the effect of commonsenseknowledge and the complexity of the task to be examined separately. The interaction betweenthese two factors can also be studied.42RESEARCH METHODOLOGYCurrently, commonsense knowledge support is mainly provided during the creation ofthe E-R model, which corresponds to the recognition level in Batra and Davis’s process modelintroduced in Section 2.4. This has been identified as the step that is most difficult for a novicedesigner and commonsense knowledge can possibly play an important role. It is therefore themain focus of this experiment.7.2 Subjects44 UBC students volunteered as subjects. With a=.05, effect size=.4, this sample sizegives a statistic power of .7 (Coven, 1977; Coven and Coven, 1983). The subjects wererandomly assigned to one of the four groups. The four groups are of equal size, with 11subjects in each of them.Monetary award was given to encourage participation. Extra money was given to thetop 5 performers.Although no formal database design knowledge was required, the subject did need toknow how to use the VCS, especially its interface. A tutorial was administered before theexperiment to familiarize the subjects with the system.One of the assumptions of the VCS is that the user may have no formal databasedesign knowledge but be an expert in the problem domain. The tasks used in the experimentwere both structured around the academic environment. Therefore, university students areappropriate subjects in this context.43RESEARCH METHODOLOGY7.3 Independent VariablesThe two independent variable were the existence of commonsense knowledge in theVCS and the complexity of the task.7.3.1 Commonsense KnowledgeBecause the experiment was conducted using university students as subjects andeducation as the application domain, the commonsense knowledge base used is in educationdomain as well. Based on the Naive Semantics discussed in Section 4.3, the followingGeneric Education Model was developed and used as the commonsense knowledge base inthe experiment.Obviously, the facts included in this model are ‘commonsense’ and trivial to mostpeople.Entities (Attributes):school (NAME, phone, address),course (COURSE_NUMBER, name, credit),person (SIN, name, phone, address),teacherprofessorstudent (STUDENT NUMBER) ,Relationships:teacher is_a person,professor is_a teacher,student is_a person,school employs professor,school enrolls student,school offers course,teacher teaches course,student takes course.44RESEARCH METHODOLOGYAlthough a domain of relevance knowledge base discussed in Section 4.5 can be builtand appended to the above knowledge base, it is not necessary in this experiment because ofthe focus of the experiment. All the knowledge stored in the commonsense knowledge base isin the education domain and therefore always available to the subject during the designsession.7.3.2 TasksTwo cases were developed based on the pilot study by Gilbert (1993). The universityselection case is the simple one, with only 3 entities and 4 relationships. The student advisorycase is more difficult and has 7 entities and 7 relationships (Appendix F).In the university selection case, the subject was asked to design a database to supportstudents in selecting their universities. All the entities in the design are easily identified and therelationships are simple binary relationships.The student advisory case required the subject to create a database to storeinformation such as courses and professors for an academic program. One of the difficulties inthis task is the category relationship between elective courses, core courses, and courses. Theother difficult point lies in the unary relationship between course and its prerequisites.7.4 Dependent Variables7.4.1 Time Taken to Finish the TaskThis is defined as the time from the moment the subject starts working on the modeltill the moment he finishes the E-R model.45RESEARCH METHODOLOGY7.4.2 Design QualityDesign quality can be regarded as how well the conceptual data model captures realityand conforms to the data model syntax. Measuring the design quality is a difficult task. Areview of previous literature (Batra and Antony, 1992a, 1992b; Shoval and Even-Chaime,1987; Batra, et. al., 1990, Gilbert 1993) indicates there is no standard grading scheme fordesign quality.It is generally accepted that the key elements of a E-R model are the basic constructs:entities, relationships and attributes (Chen, 1976; Date, 1990), so a grading scheme wasdesigned based on the design’s accuracy in using these constructs (Appendix A). A standarddesign (Appendix B) was first developed by the researchers and the subjects’ scores weredetermined by matching entities, relationships and attributes in his design with this standarddesign. To make the results comparable across tasks, scores were converted to percentage ofcorrectness by dividing each subject’s score by the perfect score.7.4.3 Perceived Ease of UseUser perception has been increasingly used as a surrogate for system effectiveness(Melone, 1990). Davis (1989) developed and validated two measures of user perceptions:perceived usefl.ilness and perceived ease of use. Moore and Benbasat (1991) further refinedthese measures and established an instrument to measure the perceptions of adopting aninformation technology innovation. Because these measures have been widely used byresearchers and showed good convergence and reliability, we adopted Moore and Benbasat’sscale in this experiment. However, the perceived usefhlness would be hard to measure becausemost subjects don’t design any database on their job. The validity of the result would be46RESEARCH METHODOLOGYcompromised under this circumstance. Only the perceived ease of use measure was used inthis experiment.7.5 Experimental Procedure andData CollectionThe experiment was conducted in an office with a computer and a video camera. Thecomputer is a 486 IBM-PC compatible with color monitor. The experimenter started the VCSbefore the subject entered the room. The video camera was used to video tape the computerscreen during the design session. The complete design session was video taped except whenthe tutorial and questionnaires were administrated.7.5.1 Pre-Test QuestionnaireThe subject first filled out a questionnaire on his background, including databasedesign experience and computer experience. This information was used in the data analysis.7.5.2 Introduction and TutorialBefore the experiment, the subject read a one and half page introduction to the wholeexperiment, database design principles and, especially, the VCS system. This was followed bya tutorial. The goal of the tutorial was to show the subject how to apply the design principlesand how to use the system. The tutorial also helped to remove the learning effect. A samplelibrary information system was used in the tutorial (Appendix E). The experimenter explainedand demonstrated the usage of function keys through a complete design process. The subjectwas encouraged to ask questions about the use ofVCS during the tutorial.The tutorial took approximately 30 minutes.47RESEARCH METHODOLOGY7.5.3 Understanding the Test Case DescriptionIn this step, the subject read the case and was allowed to ask the experimenterquestions regarding the description in the case or the design requirements. No time limit wasspecified to allow the subject fully understanding of the case.7.5.4 Creation of the E-R ModelAfter the subject indicated he had finished reading the case and had a goodunderstanding of the task required, he was allowed to use the VCS to start to design thedatabase using the E-R model. No questions of any kind were allowed to eliminate humanintervention.The process was video taped and time stamped. The time to finish this step was latercalculated based on the video tape.After the subject finished the task, his design was saved as a file. The design was latergraded by the researcher using the pre-defined grading scheme.7.5.5 Post-Test QuestionnaireAfter the user finished designing the E-R model, his perceived ease of use of the VCSregarding the creation of the E-R model was measured by a questionnaire.48DATA ANALYSIS8. DATA ANALYSISThe experimental data collected were entered into computer files. There are threedependent variables in this study:1. Time taken to finish the design task;2. Quality of the design;3. User perceived ease of use.The summary of the results is shown in Table ito 3.N Time AverageNo Commonsense Easy Task H 8 28 23 13 13 27 47 40 25 27 22 25CommonsenseEasyTask H 17 20 12 21 10 22 9 19 25 18 10 17No Cominonsense Difficult Task 11 24 21 39 22 32 27 52 11 25 30 46 30Commonsense Difficult Task 11 25 14 24 30 13 39 28 7 18 26 37 24Table I Time taken to complete task by groups (minutes)N Quality AverageNo Commonsense Easy Task 11 78 85 66 85 63 66 29 68 73 76 61 68Commonsense Easy Task 11 76 73 85 61 78 56 93 68 71 83 49 72No Commonsense Difficult Task 11 53 63 66 69 81 56 63 57 51 53 59 61Conunonsense Difficult Task 11 87 78 69 62 82 44 76 46 82 79 78 71Table 2 Quality ofdesign by groups (percentage)N EOU AverageNo Commonsense Easy Task 11 28 23 23 7 15 18 13 24 22 8 25 19Commonsense Easy Task 11 23 28 15 23 21 9 25 26 19 26 14 21NoCommonsenseDifflcultTask 11 11 19 8 19 23 18 10 19 16 22 9 16Commonsense Difficult Task 11 28 19 20 25 29 18 19 26 21 19 19 22Table 3 Perceived ease ofuse by groups (total score on the EOU questionnaire, rangesfrom 0 to 35)49DATA ANALYSISBecause the experiment was a multiple dependent variable design, a MultivariateAnalysis of Variance (MANOVA) was conducted. The SPSSIPC 4.0’s MANOVA procedurewas used with time, design quality and user perceived ease of use as dependent variables.Existence of the commonsense knowledge module and the complexity of the task were thetwo independent variables. The printout from the MANOVA procedure is included inAppendix C.8.1 Commonsense Knowledge Effect8.1.1 Time and Perceived Ease OfUseThere are significant differences between the VCS without commonsense knowledgeand the VCS with commonsense knowledge in terms of time taken to finish the task andperceived ease of use (for time: F(1,40)=5.65, p<O.O2 and for perceived ease of use:F( 1 ,40)=5 .86, p<O. 02). Therefore, Hi and H3 are rejected. This suggests that subjects usingthe VCS enhanced with the commonsense knowledge module used less time to finish thedesign and perceived it to be easier to use. This was supported by the discussion in Chapter 6where the hypothesis is developed.8.1.2 Design QualityH2 is not rejected ( F(1,40)=2.07, p<O. 16). This shows adding the commonsenseknowledge base did not improve the design quality significantly.The content of the commonsense knowledge base provides a possible explanation whythe design quality is not improved. The commonsense knowledge base used in the experimentonly contains a minimal set of facts. This was intended to avoid slanting the results in favor of50DATA ANALYSIScommonsense reasoning. This approach, however, can also weaken the effect ofcommonsense knowledge base.Another possible explanation for this may be found in the way commonsenseknowledge is retrieved in the current system. When the commonsense knowledge modulereceives a term that the VCS wants to find more about, it only does a simple string search onthe text of the term. If the term appears in a fact in the commonsense knowledge base, thatfact is regarded as relevant to the term and returned to the VCS. For example, if the VCSwants to find out all the commonsense knowledge about ‘students’, it will pass the term‘student’ to the commonsense knowledge module. The commonsense knowledge module thenfinds all the facts it knows where this term appears. If the user enters a misspelled word or asynonym not stored in the commonsense knowledge base, the system will not be able torecognize it and will not retrieve any information. In this case, the system can not be of anyhelp to the user.Another reason is that some subjects accepted whatever was suggested by thecommonsense knowledge base without judging its relevance. The commonsense knowledgeretrieved in this experiment is always relevant to the education domain, but not alwaysrelevant to the specific case. Sometimes, the entities or relationships retrieved were evenincorrect from the case’s point of view. The user, as the designer, has to decide what isneeded and what is not. The VCS as an expert system can help the user in the process, but cannot replace the user. If the user relies completely on the system and its commonsense51DATA ANALYSISknowledge, then some irrelevant or even incorrect entities, relationships and attributes will beincluded.This was clearly shown in the number of extra entities or relationships entered by thesubjects during the experiment (Figure 4). Subjects using the VCS with the commonsenseknowledge module entered more extra relationships than their counterparts using the VCSwithout the commonsense knowledge module. However, the number of extra entities was thesame for the two group. The way most people work with the E-R model helps to shed somelight on this contradiction. It was observed in the experiment that most people started themodel with entities. They only moved to the relationships after they had finished entering allthe entities. For each user entered entity, the VCS’s commonsense reasoning module couldretrieve related relationships and add them to the design automatically. For each user enteredrelationship, the VCS only added the two entities mentioned in the relationship if they did notexist. Because most subjects started with entities, only extra relationships were added by thecommonsense knowledge module. It was very rare when a subject entered a relationshipbefore he entered the two entities.52DATA ANALYSIS8.2 Task Complexity Effect8.2.1 TimeStatistical analysis shows task complexity has an effect on the time required to finishthe task (F(1,40)4.06, p<O.05). Therefore, H4 is rejected. This means the subjects used moretime on the complex task, which is expected.8.2.2 Design QualityThe effect of task complexity on design quality is not significant (F(1,40)=0.29,p<O.4’7). Therefore H5 is not rejected.This was a surprise because the complex task contained not only more constructs butalso more difficult relationships. Novice designers usually have great difficulties modeling theNb Conirion• CommonExtra Bit Extra ReinFigure 4 Average number ofextra entities and relationships entered by each subject53DATA ANALYSISunary and category relationships as found by Gilbert (1993) and confirmed in the pilot testingat UBC. They have trouble both identi1,’ing and representing these constructs.A closer look at the experimental procedure helped to explain the result. In the tutorialgiven at the beginning of the experiment, the concepts of unaiy and category relationships andthe techniques to model them were introduced. The subjects had a fresh memory about thisknowledge when they started the experiment. They often looked actively in the experimenttask for unary and category relationships because they thought what was taught in the tutorialmust be used somewhere in the experiment. Under this circumstance, it was relatively easy forthem to recognize these relationships and apply the techniques they just learned. This helpedthe subjects given the complex case to perform better than they otherwise might have. In realdesign tasks, the identification and modeling of unary and category relationships depend onmany more factors than that those can be taught in a 30 minute tutorial, and the performanceof novice designers may not be as good as in the experiment.8.2.3 Perceived Ease Of UseThe effect on perceived ease of use is not significant (F(1,40)=0.22, p<O.&1).Therefore, H6 is not rejected which means the subjects perceived the system to be equallyeasy to use regardless of the task.One explanation of this result is that the subjects were asked questions regarding thesystem, not the experiment task, so the complexity of the task did not affect their impressions.When the subjects evaluated the system, they very likely considered other aspects of the54DATA ANALYSISsystem like the user interface which was exactly the same for all groups. The strong effect ofthe interface might dominate the subjects’ perceived ease of use of the system.8.3 Interaction EffectThere is only one possible interaction effect, the interaction between the existence ofcommonsense knowledge base and the complexity of the task in this study. Statistical analysisshows this interaction effect is not significant (for time: F(1 ,40)=O. 11, p<O. 74; for perceivedease of use: F(1,40)1.47, p<O.23; for quality of design: F(1,40)=O.52, p<O.4’7).As discussed in Chapter 6, commonsense knowledge is only ‘shallow’ knowledgeabout the world, so it is not as effective in a complex task as in a simple task. However, eventhough commonsense can not provide a large percentage of the knowledge used in the finaldesign, it can be used as a framework by some users to construct their own model. Entitiesand relationships added by the system, although simple, may inspire the user to think about thetask from new directions and discover other structures. In that sense, commonsense can helpeven in complex tasks.55CONCLUSION AND FUTURE RESEARCH9. CONCLUSION AND FUTURE RESEARCHIn this study, a new database design expert system enhanced with commonsenseknowledge was developed and the effectiveness of the commonsense knowledge module wastested using a laboratory experiment. An ontology was also proposed to store domain ofrelevance information for the commonsense knowledge. One of the major contributions of thisexperiment is that it showed that even a minimal amount of commonsense knowledgeimproves the time taken to finish a design task and the perceived ease of use. Theestablishment of the causal relationship between commonsense knowledge and improveddesign process can be used as the basis for further research into other aspects ofcommonsense reasoning support.The results of this study suggest that commonsense knowledge can play an importantsupportive role in the database design process. The commonsense knowledge provided to theuser, however, should be relevant to the application domain and design task. The domain ofrelevancy information can be used to select facts from the commonsense knowledge base.Commonsense knowledge’s effect is apparent in both simple and complex situations. Thisshows that commonsense knowledge will be an important help in various design tasks in thereal world.9.1 Limitation ofthe StudyThe structure of commonsense knowledge in the commonsense knowledge base andthe retrieval process are still in the prototype stage, thus limiting the effectiveness of thecommonsense knowledge module.56CONCLUSION AND FUTURE RESEARCHAlthough a domain of relevance ontology was created, it was not tested in theexperiment because of the single problem domain. The Generic Business Model (Goldsteinand Storey 1991) was not implemented for the same reason.9.2 Future ResearchSeveral possible research directions come from this study. After the causal relationshipbetween comnionsense knowledge and system effectiveness has been established, protocolanalysis should be done to analyze the dynamics of the design process. By doing so, the exactrole played by commonsense knowledge can be identified. This should provide guidance tofurther development of an intelligent database design expert system which uses commonsenseknowledge to the best extent.The content of the commonsense knowledge base should be further refined. For thepurpose of the experiment, a very small amount of knowledge was put into the system. In thefuture, more commonsense facts and their domain of relevance information should beidentified and stored. An experiment with an expanded commonsense knowledge base willallow researchers to fully study the interaction between the VCS and the commonsenseknowledge module.The current commonsense knowledge module uses simple string search to retrieverelated commonsense facts. More reasoning patterns should be developed to provide morerelevant commonsense knowledge support.Learning ability will enable the system to learn new commonsense knowledge as it isused by different users. The system should be able to reorganize its comrnonsense knowledge57CONCLUSION AND FUTURE RESEARCHontology and domain of relevance ontology when knowledge is accumulated. This willsimplifj the maintenance of the commonsense knowledge base.58BIBLIOGRAPHY10. BIBLIOGRAPHYAshcraft, M.H. (1976). Property Norms for Typical and Atypical Items from 17 Categories: A Description andDiscussion. Memory and Cognition. 6:3, 227-232Batra, D. and Antony, S. R. (1992a). Effects of Data Model and Task Characteristics on DesignerPerformance: A Laboratory Study. Working Paper. Florida International University.Batra, D. and Antony, S. R. (1992b). Novice Biases in Conceptual Database Design. Working Paper. FloridaInternational University.Batra, D. and Davis, J. G. (1992). Conceptual Data Modeling in Database Design: Similarities and DifferencesBetween Expert and Novice Designers. International Journal ofMan-Machine Studies, 37, 83-10 1.Batra, D., Hoffer, J. A. and Biostrome, R. P. (1990). Comparing Representations with Relational and EERModels. Communications ofthe ACM Februaiy 1990, 33:2, 126-139.Benbasat, I. (1989). Laboratory Experiments in Information Systems Studies with a Focus on Individuals: ACritical Appraisal. Harvard Business School Research Colloquium. 23-47.Brodie, M. L. (1984). On the Development ofData Models, in M. L. Brodie, J. Mylopoulos and J. W. Schmidt(Eds.), On ConceptualModeling. 305-338Buchanan, B. (1988). Expert Systems: Working Systems and Research Literature, in Savory, S.E.(Ed.), ExpertSystems in the Organization: An Introductionfor Decision-Makers, Ellis Horwood. 34-75Chen, P. P. (1976). The Entity-Relationship Model - Toward a Unified View ofData. ACM Transactions ofDatabase Systems, 1, 9-36.Chen, P. P. (1977). The Entity-RelationshipApproach to Logical Data Base Design. Wellesley, Mass.Choobineh, J., Konsynski, B. R., Mannino, M. V. and Nunamaker, 3. F. (1988). An Expert System Based onForms. IEEE Software Engineering, Februaxy 1988, 242-253.Codd, E. F. (1970). A Relational Model of Data for Large Shared Banks. Communications ofthe ACM 13,377-387.Coven, J. (1977). Statistical PowerAnalysisfor the Behavioral Sciences. Academic Press, NJ.Coven, J. and Coven, P. (1983). AppliedMultiple Regression/CorrelationAnalysisfor the BehavioralSciences, 2nd Edition, Lawrence Erlbaum Associates, NJ.Date, C. J. (1990). An Introduction to Database Systems, Volume 1, Ffth Edition. Addison-Wesley, Reading,MA.Dahigren, K. (1985). The Cognitive Structure of Social Categories. Cognitive Science. Vol. 9, 379-398.Dahigren, K. (1989). Naive Semanticsfor Natural Language Understanding. Kluwer Academic PublishersDavis, F. D. (1989). Perceived Usefulness, Perceived Ease ofUse, and User Acceptance of InformationTechnology. MIS Quarterly, September 1989, 319-340.Dogac, A., Yuruten, B. and Spaccapietra, S. (1989). A Generalized Expert System for Database Design. IEEEtransactions on Software Engineering, 15:4, 479-49 1.Ein-Dor, P. (1985). Representing Commonsense Knowledge: A Survey. Working Paper No. 864/85, IsraelInstitute of Business Research, Tel Avi University, June 1985.Elmasri, R. and Navathe, S. B. (1989). Fundamentals ofDatabase Systems. Menlo Park, C.A. BenjaminCununings Publishing Co.Emory, C. W. and Cooper, D. R. (1991). Business Research Methods. Irwin, Homewood, IL.Fehr, B. and Russel, J.A. (1984). Concept ofEmotion Viewed From a Prototype Perspective. Journal ofExperimental Psychology: General. 113:3, 464-486.59BIBLIOGRAPHYGilbert, M. T. (1993). Expert systems and database design: the View Creation System. Working Paper,University of Rochester.Goldstein, R. C. (1985). Database: Technology andManagement, John Wiley & Sons.Goldstein, R. C. and Storey, V. C. (1991). Commonsense Reasoning in Database Design. Proceedings ofthe10th International Conference on the Entity-Relationship Approach.Guha, R. V. and Lenat, D. B. (1900). CYC: A Midterm Report. The AlMagazine, Fall 1990.Hamilton, S. and Chervany, N. L. (1981). Evaluating Information System Effectiveness - Part I: ComparingEvaluation Approaches. MIS Quarterly, September 1981, 55-69.Hobbs, J. R. and Moore, R. C. (1985). Formal Theories ofthe Common-sense World. Norwood, NJHoffer, J. A. (1982). An Empirical Investigation into Individual Differences in Database Models. Proceedingsofthe Third International Conference on Information Systems.Kolata, G. (1982). How Can Computers Get Conunon Sense?. Science. September 1982, 217:4, 1237-1238.Konsynski, B. R. (1979). Data Base Driven Systems. University of Arizona.Korth, H. F. and Silberschatz, A. (1986). Database System Concepts. McGraw-Hill, New York, NY.Lenat, D. B., Prakash, M. and Shepherd, M. (1986). CYC: Using Common Sense Knowledge to OvercomeBrittleness and Knowledge Acquisition Bottlenecks. The AlMagazine 4:65-85Lenat, D. B., Guha, R.V., Pittman, K., Pratt, D. and Shepherd, M. (1990). CYC: Toward Programs WithCommon Sense, Communications ofthe ACM. August 1990, 33:8, 30-49Lord, C. and Dahigren, K. (1990). Representation ofBusiness and Financial Knowledge in the NewSelectorSystem. Working Paper, IBM Los Angeles Scientific Center.McCarthy, J. and Hayes, P.J. (1969). Some Philosophical Problems from the Standpoint of ArtificialIntelligence. Machine Intelligence 4, Elsevier, NYMcCarthy, J. (1983). Some Expert Systems Need Common Sense. Annals of the New YorkAcademy ofSciences. Vol. 426, 129-137.McFadden, F. R. and Hoffer, 3. A. (1988). Data Base Management, SecondEdition, BenjaimnJCummings,Menlo Park, CA.Meltzer, B. Knowledge and Experience in Artificial Intelligence. Al Magazine. Spring 1985, 6:1, 40-42.Moore, G. C. and Benbasat, I. (1991). Development of an Instrument to Measure the Perceptions of Adoptingan Information Technology Innovation, Information Systems Research, 2:3, 192-222.Rosch, E., Mervis, C.B., Gray, W.D., Johnson, D.M., and Boyesbraem, P. (1976). Basic Objects in NaturalCategories. Cognitive Psychology. Vol. 8, 382-439.Sharda, R., Barr, S. H. and McDonnell, J. C. (1988). Decision Support System Effectiveness: A Review AndAn Empirical Test. Management Science, 34:02, 139-159.Shoval, P. and Even-Chaime, M. (1987). Data Base Schema Design: An Experimental Comparison BetweenNormalization and Information Analysis. Data Base, Spring 1987, 30-39.Storey, B.C. (1988). View Creation: An Expert Systemfor Database Design. ICIT Press, Washington, DCStorey, V. C. and Goldstein, R. C. (1988). A Methodology for Creating User Views in Database Design. ACMTransactions in Database Systems, 13:3, 305-338.Storey, V. C. and Goldstein, R. C. (1989). Some Findings on the Intuitiveness of Entity-RelationshipConstructs. Proceedings ofthe 8th International Conference on Entity-RelationshipApproach.Todd, P. and Benbasat, I. (1987). Process Tracing Methods in Decision Support Systems Research: Exploringthe Black Box. MIS Quarterly, December 1987, 493-512.60BIBLIOGRAPHYTodd, P. and Benbasat, I. (1992). The Use of Information in Decision Making: An Experimental Investigationof the Impact of Computer-Based Decision Aids. MIS Quarterly, September, 1992, 373-393.Winograd, T. and Flores, F. (1987). Understanding Computers and Cognition. Addison-Wesley PublishingCompany, Inc.61APPENDIXAppendix A Grading Scheme for Design QualityITEM POINTSASSIGNEDCorrect Entities 2Correct Attribute 1Correct Pnmary Key 1Extra Attribute 0Incorrect Entities 0Correct Relationships 2Correct Cardinality One 1Correct Cardinality Two 1Correct Attribute 1Incorrect Relationship 062APPENDIXAppendix B Standard DesignStandard E-R design for the advisory caseENTITIESCourseCore_CourseElective CourseSectionStudentProfessorPre_requisiteRELATIONSHIPSCore_Course (1,1) is_aElective_Course (1, 1) isaPrerequisite (1,1) is_aStudent(O, n) takesProfessor (1, n) teachesCourse(O,n) hasCourse(1,n) has[COURSE NUMBER, description, Section_per_year][waiver_allowed][enrollment limit)[SECTION_NUMBER, time, location, no_student][STUDENTNUMBER, name, address, phone, gpa, major][NAME,OFFICE NUMBER, area, rank][]Course(O, 1)Course(O,l)Course(O, 1)Course(l,n) [grade]Course (1, n) [Section number]Prerequisite(O, n) [must_take_before]Section(l,l)63APPENDIXStandard E-R design for the university selection caseENTITIESUniversity [NAME, address, phone, tuition]Program [NAME, credit_hours, year_started, gmat required]Alumni (NAME, ADDRESS, phone, year_graduation, major]Professor [NAME, ADDRESS, phone, area]RELATIONSHIPSUniversity(l,n) has Program(l,1)Program(1,n) has Professor(l,n)Program(1,n) has Alumni(l,1)64APPENDIXStandard E-R design for the tutorial caseENTITIEScollectionbookserialreferenceborrower[CALL NUMBER, title, status][author][issue number]RELATIONSHIPSbook (1,1)serial(1,1)reference (1,1)book(0,n)borrower(0, n)is ais ais ahasborrows[][CARD_NUMBER, name, phone, address]collection(0, 1)collection(0, 1)book (0,1)reference(0,n)book(0,1) [due_date]65APPENDIXAppendix C Statistical Results* * ANALYSIS OF VARIANCE -- DESIGN 1 * *EFFECT .. TASKMultivariate Tests of SignificanceTest Name Value Approx. F.12233 1.76546.13938 1.76546.87767 1.76546.12233PillaisHotellingsWil ksRoysUnivariate F—tests with (1,40) D. F.Variable Hypoth. SS Error SS Hypoth. MSTIME 408.09091 4017.27273 408.09091EOU 7.36364 1312.36364 7.36364QUALITY 79.11364 10809.2727 79.113641/2, N = 18Error DF Sig.38.0038.0038.00(S = 1, M =Hypoth. DF3.003.003.00of F.170.170.170Error MS F Sig. of F100.43182 4.06336 .05132.80909 .22444 .638270.23182 .29276 .59166APPENDIX* * ANALYSIS OF VARIANCE -- DESIGN 1 * *Univariate F—tests with (1,40) D. F.EFFECT .. COM1ONMultivariate Tests of SignificanceTest Name Value Approx. F2.931182.931182.93118PillaisHotellingsWi 1 ksRoys.18792.23141.81208.18792(S = 1, M = 1/2, N = 18Hypoth. DF Error DF Sig. of F3.00 38.00 .0463.00 38.00 .0463.00 38.00 .046Variable Hypoth. SS Error SS Hypoth. MS Error MS F Sig. of FTIME 567.36364 4017.27273 567.36364 100.43182 5.64924 .022EOU 192.36364 1312.36364 192.36364 32.80909 5.86312 .020QUALITY 560.20455 10809.2727 560.20455 270.23182 2.07305 .15867APPENDIX* * ANALYSIS OF VARIANCE -- DESIGN 1 * *EFFECT .. COMMON BY TASKMultivariate Tests of Significance (S = 1, M 1/2, N = 18Test Name Value Approx. F Hypoth. DF Error DF Sig. of FPillais .05656 .75944 3.00 38.00 .524Hotellings .05996 .75944 3.00 38.00 .524Wilks .94344 .75944 3.00 38.00 .524Roys .05656Univariate F—tests with (1,40) D. F.Variable Hypoth. SS Error SS Hypoth. MS Error MS F Sig. of FTIME 11.00000 4017.27273 11.00000 100.43182 .10953 .742EOU 48.09091 1312.36364 48.09091 32.80909 1.46578 .233QUALITY 141.84091 10809.2727 141.84091 270.23182 .52489 .47368APPENDIXAppendix D TutorialTHE UNiVERSITY OF BRITISH COLUMBIAThe Faculty of Commerce and Business AdministrationManagement Information Systems DivisionDatabase Design Expert System StudyTUTORIAL -- ONEDATABASE DESIGN BASICSDatabase design involves describing the structure of the portion of the real world thatyou want to have information about in your database. We design a database by defining theentities, relationships and attributes that are to be included. This is not very different fromdescribing something in ordinary English:Entities are like nouns they refer to persons, places, or things we want to haverepresented in the database. In deciding what entities to have in your database design, thinkabout what things you are going to want information about. For example, the entities for alibrary database would surely include book and borrower, and might also include branch,author, subject, etc., depending on exactly what purposes the database was intended to serve.Relationships correspond to verbs in ordinary language. They represent some kind ofconnection between entities. The most common kind of relationship connects two entities, asin “borrower borrows book”. Borrows is the verb in this phrase and it states a relationshipbetween the two entities, book and borrower.Although relationships such as the one just shown are the most common, it is possibleto have a relationship that involves only one kind of entity or more than two kinds. Forexample, “book references book” expresses the fact that one book can reference another. It isstill a relationship between two things, but in this case, they are both of the same kind.Similarly, “borrower borrows book from branch” is a relationship among three entities. This isfairly rare in practice.One of the characteristics of a relationship which must be specified is referred to as its“mapping ratios” or “cardinalities”. This describes the number of instances of the relationshipa single instance of that entity can participate in. It is not necessary to know the exact number-- just whether it is “at most one”, “at least one”, “exactly one” or “any number”. As an69APPENDIXexample, consider the relationship “borrower borrows book”. A single borrower canparticipate in any number of “borrow” relationships, but asingle book can participate in at most one since if it has been borrowed by one person,it can’t be borrowed by anyone else at the same time. Mapping ratios are sometimes expressedas a pair of numbers in parentheses. The first number is the minimum and the second, themaximum. In this notation, we usually use “n” to mean any number greater than 1. In thisnotation, the mapping ratios for borrower would be (0,n) and those for book would be (0,1).A particularly usefi,il type of relationship is “is_a”, as in “novel is_a book”. Themeaning of this relationship is that a novel is a kind of book. Therefore, anything that thedatabase system knows about books in general must also apply to novels,Attributes in a database design are used to describe entities and relationshipssomething like adjectives and adverbs in English. Attributes of an entity (noun) describeproperties of that kind of thing. For example, the Borrower entity might have attributes suchas card_number, name, address, no._of_overdue_books, etc. We should include in a databasedesign only those attributes that are necessary for the task. Every entity will have someattributes specified for it or there is no point in having it in the database at all!A key is a special kind of attribute. It refers to an attribute (or combination ofattributes) that can be used to uniquely identify a particular entity. For example,“card_number” is a key for the Borrower entity, since no two borrowers should ever have thesame card number. “Name” would not be a good key for Borrower because it is possible tohave two borrowers with exactly the same name. In such a case we could use a combinationof attributes as a key, such as “name, address”.Relationships can also have attributes. The relationship “borrower borrows book”might have the attribute “date_due”. Not all relationships have attributes.70APPENDIXTUTORIAL — TWOUSING THE VIEW CREATION SYSTEM (VCS)IMPORTANT: There is a prompt line at the bottom ofscreen at all time to showyou all the keys availableto you at that time. Please read them carefully. You can also access the system ‘s context sensitive help by pressingFl at any time. The experimenter will not answer questions during the experiment.Stage OneIn this stage, you will use the VCS to create an Entity-Relationship model for yourdatabase. Please read the Task Description carefully. Your design should include all and onlyinformation mentioned in the Task Description.1. Enter entities, relationships and attributes: Type on the screen directly. After each line, press [1]to go to the next line. Press [ENTER] on a line to enter attributes, keys and mapping ratios.• Entity and attribute names can be only one word. Relationship will be entered in theform ‘entity_one relationship entity_two’, as in ‘customer buysproduct’.• All names must be in lower case. All the entity names should be in singular form. i.e. use‘car’ instead ‘cars’ in all places.• Use a single word for name. If you want to use multiple words, join them with anunderline (not dash). For example, ‘insurance_num’ for ‘insurancenumber’.2. After you are satisfied with your design, please notify the experimenter.Stare TwoIn this stage, you will use the VCS to modify your original Entity-Relationship model andtranslate it into a Relational model. The VCS will ask you some questions regarding your design.You should always choose ‘Continue’ when presented with the choice ‘Continue/Go back to themain menu’. As in Stage One, you have access to the prompt line and context sensitive help. Readthem carefully before making your choice.1. Enter missing information: The most common case is missing key attribute. When the VCSfinds an entity without a key attribute, it will alert you and enter the entity screen and place thecursor at the first key attribute field. To add a key attribute, press [Insert]. Use the [Space Bar]to identify key attribute(s) in the pop-up window. Then press [F 10] to exit.2. Answer other questions from the VCS. For example, clarifying some words’ meaning,identifying multiple value, functional dependency. The terms will be explained in the ‘SystemMessage’ and Help.3. After the VCS solves all the problems and returns to the main menu, please notify theexperimenter.71APPENDIXAppendix E Tutorial TaskTUTORIAL — THREECASE DESCRIPTIONYou were asked by your community library’s director to develop a database for her library.After talking to her, you found out that the library has two categories of holdings: books andserials. Some information, like title, call number, status, should be kept for both kinds ofcollections. However, only books have author and only serials have issue number. Some books alsohave references which are also books.Besides book, you were told to store information on borrowers as well. This includesname, card number, phone and address.Finally, the director required that she be able to check who borrowed which books (onlybooks can be borrowed) and the due date, so the database has to store this information as well.You were told the database you design should be able to answer questions like thefollowing:I. What is the status ofa particular book?2. Which books are borrowed by a particular borrower and when is the due date?3. What is the reference book(s) for a particular book?72APPENDIXAppendix F Experiment TasksTUE UNiVERSITY OF BRITISh COLUMBIAThe Faculty of Commerce and Business AdministrationManagement Information Systems DivisionDatabase Design Expert System StudyTASK DESCRIPTIONAs a major in Entrepreneurship, you want to set up a one-person consultingpractice whereby you will provide information to students who are selecting businessprograms for their graduate studies. In order to do so you will need a database withinformation on universities and their business programs.First you would like to store some information for each university, forexample, name, address, phone, and tuition. Each university offers various types of degreesincluding MBA, M. Sc. and Ph.D. The number of credit hours required is different for eachprogram. The number of credit hours required can vary from one university to another, too.Some programs require GMAT scores for admission while others do not. It is important toidentify whether a GMAT score is required. Students would also be interested in the year auniversity started a program.Universities have a number of professors who may teach in more than oneprograms. Some potential students will request information on certain professors so that theycan contact them directly. Thus you will need to retain information on a professor thatconsists of name, address, phone, and area (for example, finance, marketing, etc.)73APPENDIXRecently, universities have started to engage alumni who are willing to talk toprospective students in the same program. Universities will provide you with a list of eachprogram’s alumni contacts which includes their name, address, phone, year of graduation andtheir major.Questions:The following is a list of questions that the database should be able to provideanswers for:1. What is the tuition at a particular university?2. What program(s) does a particular university offer (for example, MBA, M. Sc., etc.)?3. How long has a particular program at a particular university been established?4. Who is teaching at a particular university for a particular major?5. Who graduated from a particular program at a particular university?.74APPENDIXTHE UNIVERSITY OF BRITISH COLUMBIAThe Faculty of Commerce and Business AdministrationManagement Information Systems DivisionDatabase Design Expert System StudyTASK DESCRIPTIONYou have been assigned the task of developing a database for a studentadvisory function. The purpose of the database is to provide students with information on thecourses for their program, the pre-requisites for each of their courses; whether or not they areeligible for waiver exams; information on course offerings, etc.Courses may be core courses for some programs or elective courses.Obviously some information you will need will be the same for both types of courses. Forexample, each course should have a course number and a brief description and there should bean indication of the number of times it is offered throughout the year. The main differencebetween the two types of courses is that elective courses have enrollment limits and corecourses may allow waiver examinations.Each course has one or more Sections. For those courses that have only oneSection, the Section number will always be ‘001’. When a course has multiple Sections, eachSection need not be taught by the same professor (although this is possible of course).Courses can have one or more pre-requisites and it is necessary to retaininformation on what the pre-requisites are. A pre-requisite is usually taken before the coursethat requires it, although some may be taken concurrently with the higher level course.75APPENDIXFor each course offering the database should keep track of which professor isteaching which Section, plus the time and location (room number) where each Section meets.Assume that a student has just one major area (for example, accounting,marketing, etc.) which should be stored in the database along with the student’s name, studentnumber, address, phone number, and grade point average. For every course a student takes,his or her mark should be retained.For each professor, information is needed on the professor’s name, area ofexpertise (for example, accounting, marketing, etc.), office number and rank (assistant,associate, ftill).QuestionsThe following is a list of questions that the database should be able to provideanswers for:1. What is the description of a particular course?2. What are the pre-requisites for a particular course?3. What grades did a student receive in his or her courses?4. Who is teaching a particular Section of a given course?5. How many students were enrolled in a particular Section of a course?6. What is the seating limit for a particular elective course?76APPENDIXTHE UNiVERSITY OF BRITISH COLUMBIAThe Faculty of Commerce and Business AdministrationManagement Information Systems DivisionDatabase Design Expert System StudyQUESTIONNAIREPlease circle the appropriate wordsUsing the VCS enhances my effectiveness on the job.LIKELY I extremely quite I slightly I neither slightly I quite I extremely UNLIKELYUsing the VCS enables me to accomplish tasks more quickly.LIKELY I extremely I quite I slightly I neither I slightly I quite I extremely IUNLIKELYI believe that the VCS is cumbersome to use.LIKELY I extremely I quite I slightly I neither I slightly I quite I extremely IUNLIKELYUsing the VCS improves my job performance.LIKELY I extremely I quite I slightly I neither slightly I quite I extremely IUNLIKELYLearning to operate the VCS is easy for me.LIKELY lextremelyl quite I slightly I neither slightly I quite lextremelylUNLiKELYOverall, I find using the VCS to be advantageous in my job.LIKELY I extremely I quite I slightly I neither I slightly I quite I extremely IUNLIKELYUsing the VCS gives me Sreater control over my work.LIKELY I extremely quite I slightly I neither I slightly I quite I extremelylUNLIKELYMy interaction with the VCS is clear and understandable.LIKELYJ extremeiyj quite I slightly I neither I slightly I quite J extremelyJUNLIKELYUsing the VCS increases my productivity.LIKELY I extremely I quite I slightly neither I slightly I quite I extremely IUNLIKELYMy using the VCS requires a lot ofmental effort.LIKELY j extremely I quite I slightly I neither I slightly quite I extremely UNLIKELYIt is easy for me to remember how to perform tasks using the VCS.LIKELY I extremely I quite I slightly I neither I slightly I quite I extremely IUNLIKELY77APPENDIXUsing the VCS is often frustrating.LIKELY I extremely I quite I slightly I neither I slightly I quite extremelyIUNLIKELYI believe that it is easy to et the VCS to do what I want it to do.LIKELY I extremelyl quite slightly neither I slightly quite JextremelylUNLIKELYUsing the VCS improves the quality ofwork I do.LIKELY extremely I quite I slightly I neither I slightly I quite I extremely IUNLIKELYUsing the VCS makes it easier to do my job.LIKELY extremely I quite I slightly I neither I slightly I quite I extremely UNLIKELYOverall, I believe that the VCS is easy to use.LIKELY I extremely I quite I slightly I neither I slightly I quite I extremely IUNLIKELY78

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0087456/manifest

Comment

Related Items