Professional Documents
Culture Documents
com
Abstract
In this paper, we describe the research and development of a fuzzy expert system for hotel selection. A prototype system,
called hotel advisory system (HAS), has been designed and developed to assist tourists in conducting hotel selection using fuzzy
logic. HAS is implemented on personal computers under a Microsoft WindowsTM environment. To evaluate the performance
of HAS, selected practitioners in the Hong Kong hotel industry and potential users from twelve nations were invited to
participate in testing the system. The potential users and hotel experts rated highly on the e5ectiveness and the usability of the
system. The results of the prototype evaluation were satisfactory and support the contention that HAS performs its functions
as expected. The viability of HAS as an e5ective procedure for hotel selection has been ascertained by the positive feedback
obtained from the survey questionnaires. Using HAS makes hotel selection simple because it can incorporate the linguistic
terms which are normally produced by tourists.
? 2003 Elsevier Science Ltd. All rights reserved.
0305-0483/03/$ - see front matter ? 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0305-0483(03)00050-1
276 E.W.T. Ngai, F.K.T. Wat / Omega 31 (2003) 275 – 286
2. Literature review vice quality provided. Ghalia and Wang [14] proposed an
intelligent system using fuzzy logic to estimate the future ho-
Many published studies focus on applications of AI/ES tel room demand. However, the applications of fuzzy logic
technology which support the hotel and tourism domain in hotel selection research are almost non-existent based on
in such areas as room rental, hospitality management, the results of a literature review conducted by the authors.
concierge service, and guided tour scheduling. McCool [3] This paper describes the development of a fuzzy expert
discussed some considerations necessary for developing system named HAS, that can be used e5ectively to as-
expert systems for the hospitality industry. Nissan [4] in- sist in hotel selection. Fuzzy expert systems have found
troduced three expert systems which were applied to the widespread use in engineering, particularly in control sys-
domains of real estate, room rental and hospitality man- tems. The advantages of these systems over conventional
agement. An expert system for forecasting menu items in production rule-based expert systems may be characterised
a foodservice operation was developed by Sanchez et al. as follows [15,16]: (a) fuzzy sets neatly symbolise natural
[5]. Cho et al. [6] argued that hotels could improve their language terms used by experts; (b) since the expert knowl-
concierge service, both human and electronic, by develop- edge captured in “IF... THEN” statements is often not nat-
ing an electronic system that makes use of expert system urally true or false, fuzzy sets a5ord representation of the
technology. Cho’s system itself engaged hotel guests in an knowledge in a smaller number of rules; and (c) smooth
on-screen dialogue to help them Ind information about ho- mapping can be obtained between input and output data.
tel services and other attractions in the area. The experience
gained in the development of an expert system called an ex-
pert system for tour advisory (ANESTA), which could act 3. Development of hotel advisory system (HAS)
as a tourist information station for generating self-guided
tour schedules as well as providing detailed transportation Fuzzy expert system is an expert system that uses fuzzy
information was reported by Low et al. [7]. Sterling et al. [8] logic instead of boolean logic. It can be seen as special
described lessons learned through the sequential construc- rule-based systems that use fuzzy logic in their knowledge
tion of four expert systems for menu planning. They have base and derive conclusions from user inputs and fuzzy in-
shown how to represent common sense knowledge about ference process [17] while fuzzy rules and the membership
food and menus in a form amenable to successful menu functions make up the knowledge base of the system. The
planning. The design and development of an expert system goal of a fuzzy expert system is to take in subjective, par-
for a tourist information center was outlined by Tsang et al. tially true facts that are randomly distributed over a sam-
[9]. The expert system was built to recommend a suitable ple space, and build a knowledge-based expert system that
travel schedule that satisIes user input constraints such as will apply to them certain reasoning and aggregation strate-
time period, budget and individual preferences. Yeung et gies to produce useful decisions [15]. The purpose of this
al. [10] discussed the implementation on the Internet of a research is to design and develop a fuzzy expert system
multi-agent based tourism industry. The system allowed which can achieve the goals of operational e5ectiveness and
the users to retrieve the most up-to-date information about ease-of-use in facilitating the selection of hotels. A proto-
Hong Kong through a web browser. The complete system type system, HAS, has been developed with a view to as-
consists of a set of software agents which handle various in- sisting tourists in selecting hotels to suit their needs.
formation categories, such as hotels, shopping centres, and In this section, the development methodology of the sys-
cinemas. Law and Au [11] proposed using expert system tem is presented. The overview of the framework is shown
technology to assist tourists in locating the most suitable in Fig. 1.
hotel to meet their needs. These writers presented a revision Essentially, there are 11 fundamental phases in the de-
of the knowledge representation technique and expanded velopment of a fuzzy expert system that consist of a com-
the knowledge base of an expert system for hotel selections bination of the fuzzy inference process and the Ive-stage
in Hong Kong. Some other potential applications of expert development methodology [18]. In this study, the fuzzy
systems in tourism can be found in Moutinho et al. [12]. inference process proceeds in six steps that is a common
Fuzzy logic has proved useful for developing many prac- procedure for fuzzy inference which can be demonstrated in
tical applications, especially in the Ield of engineering, as it several past studies [13,19,20]. The choice of this approach
can handle inexact and vague information. Even though an to HAS development is based on our prior experience and
abundance of research in fuzzy logic has been conducted in lessons learnt from the development of several knowledge
the past, relatively little attention has been paid to applica- based systems such as [21,22]. It is easy to apply and will
tions of fuzzy logic technology in hotel/tourism-related in- provide valuable guidance for developing the proposed sys-
dustries. Petrovic-Lazarevic and Wong [13] underlined the tem. Eleven phases in the development are outlined in Table 1.
signiIcance of an application of fuzzy control in the hos- With reference to Table 1 above, phases 1– 6 (fuzzy in-
pitality industry in order to achieve or sustain competitive ference processing) are designed to reach a crisp solution
advantage. They applied general fuzzy control model in the to any problem involving a crisp-to-fuzzy transformation
hospitality industry to monitor and control the level of ser- (“fuzziIcation”), an inference mechanism that applies
E.W.T. Ngai, F.K.T. Wat / Omega 31 (2003) 275 – 286 277
Fuzzification
Fuzzy Inference
Process
Fuzzy Inference
Defuzzification
Construct a Conceptual
Framework
fuzzy rules, and a fuzzy-to-crisp transformation (“defuzzi- critical factor because Hong Kong is a compact city. With
Ication”). Phases 7–11 are used to construct HAS follow- the implementation of major mass transit and highway links
ing the Nunamaker et al. [18] Ive-stage methodology for (West Rail, MTR lines, Route 3, East Rail Ma On Shan
information system development. Extension, etc.) and the new Hong Kong International Air-
The detailed description of these phases is as follows: port at Chek Lap Kok nodal areas are created within new
networks, so that all hotels are easily accessible. However,
Phase 1: Identify the critical factors and de3ne member- HAS still reserve a location selection for tourist to select
ship functions and fuzzy sets. their accommodation place. Finally, we have identiIed three
The Irst phase involved the compilation of a list of critical factors which are critical in the selection of Hong Kong ho-
factors based on a literature review and indepth interviews tels: (1) price; (2) facilities; and (3) food type for fuzzy
with tourists and hotel practitioners. According to Chu and selection.
Choi [23] and a survey conducting by Hong Kong Tourism The data displayed in Fig. 2 was based on information ob-
Board [24], room rate, recreational facilities, and hotel food tained from the literature [2,25] and from indepth interviews
and beverage facilities are the importance factors for hotel with twenty potential users (tourists) and 10 hotel practition-
selection. The “Location” of the hotel is not included as a ers. These data serve as guidelines for selecting hotels. We
278 E.W.T. Ngai, F.K.T. Wat / Omega 31 (2003) 275 – 286
Table 1
Eleven phases in the development of HAS
Phases Objectives
1 Identify the critical factors and deIne membership functions and fuzzy sets
2 Construct the fuzzy rules
3 Perform fuzziIcation
4 Generate fuzzy inference
5 Perform defuzziIcation
6 Compare the overall rating for all potential hotels
7 Construct a conceptual framework
8 Develop a system architecture
9 Analyze and design the system
10 Build a prototype system
11 Observe and evaluate the system
assume that the decision-makers (the tourists) can assign rat- membership function of “Price Matching”. “Price Match-
ings to di5erent hotels under di5erent selection criteria using ing” indicates the degree of matching in price between “Ho-
common linguistic terms, for example, “cheap”, “moderate” tel Price” and the customer’s “Wanted Price”. It takes “low”,
and “expensive” as these are the linguistic terms used as cri- “medium” and “expensive” as its linguistic terms. The same
teria for “Hotel Price” and “few”, “some” and “many” are approach is used to deIne “Facilities Matching” and “Food
the criteria used to denote “Hotel Facilities”. Each linguis- Matching”. The deInition of fuzzy sets is based on the in-
tic term is deIned by a membership function which helps formation provided by the “OGcial Hotel Guide” published
to take the crisp input values and transform them into de- by the Hong Kong Tourist Association [25].
grees of membership. The most commonly used member- Phase 2: Construct the fuzzy rules.
ship function has three types: bell-shaped, triangle-shaped Fuzzy expert systems make decisions and generate output
and trapezoid-shaped. In the present study, we assume the values based on knowledge provided by the designer in the
input and output fuzzy numbers are triangular forms and form of IF condition THEN action rules. The rule base
these forms approximate human thought processes. Trian- speciIes qualitatively how the output parameter “Overall
gular membership functions have been used to deIne the Rating” of the hotel is determined for various instances of the
fuzzy sets for the linguistic values of “Hotel Price”, “Ho- input parameters of “Price”, “Facilities” and “Food Type”.
tel Facilities” and “Hotel Food Type”. The same triangu- Phase 3: Perform fuzzi3cation.
lar membership functions have been deIned for “Wanted FuzziIcation refers to the process of taking a crisp input
Price”, “Wanted Facilities” and “Wanted Food Type”. The value and transforming it into the degree required by the
E.W.T. Ngai, F.K.T. Wat / Omega 31 (2003) 275 – 286 279
(0.85665,0.7133)
(0.6434,0.2867)
the input value to the fuzzy set associated with each lin-
guistic label. For instance, an input value of “Hotel Price”
HK$2050 results in a degree of membership in the set
labelled “moderate” of 0.8726 and a degree of membership
in the set labelled “expensive” of 0.1274 (see Fig. 3).
1
high
Phase 4: Generate fuzzy inference.
Fuzzy inference is guided by the fuzzy rules. The stan-
Price Matching
dard max–min inference algorithm was used in the fuzzy
medium
inference process, as it is a commonly used fuzzy infer-
0.5
ence strategy. In the max–min composition fuzzy inference
(0.063,0.1274)
method, the min operation is used for the AND conjunc-
(0.1434,0.2867)
tion (set intersection) and the max operation is used for the
OR disjunction (set union) in order to evaluate the grade of
0
1
0
membership
equal to HK$2050. Suppose fuzziIcation for the variable
function
(2050,0.1274)
“Hotel Price” produces a 0.8726 degree of membership in
the set “moderate” and 0.1274 degree of membership in the
set “expensive”. Assume a tourist wants a price of $1400
and fuzziIcation for the variable “Wanted Price” produces
expensive
a 0.2867 of membership in the set “cheap” and a 0.7133
degree of membership in the set “moderate”, then:
3420
Rule 1: IF “Wanted Price” is cheap AND “Hotel Price”
Hotel Price
is moderate
moderate
THEN “Price Matching” is medium
1850
EVALUATION: min (0.2867, 0.8726)= 0.2867 “Price
Matching” is medium
Rule 2: IF “Wanted Price” is cheap AND “Hotel Price”
cheap
280
is expensive
THEN “Price Matching” is low
membership
Matching” is low
Rule 3: IF “Wanted Price” is moderate AND “Hotel
Price” is moderate
THEN “Price Matching” is high
(1400,0.7133)
(1400,0.2867)
Matching” is high
Rule 4: IF “Wanted Price” is moderate AND “Hotel
3420
Price” is expensive
THEN “Price Matching” is medium
Wanted Price
moderate
Matching” is medium
Since Rules 1 and 4 have the same consequent label
medium, the max operation is used to resolve conQicts. As a
membership
280
function
Database
Hotel Information Data
Data Input/Update/
(Room rate,Facilities, Virtual Visit File Exchange Rate
Del
Food Type)
User Interface
Input
Parameters
Fuzzy Inference
Hotel Detailed
Information
Defuzzification
Hotel Virtual
Tour
Rule Base
are illustrated as follows: face design can be evaluated and the developers gain insights
into the application area and into the users’ work tasks and
(i) User interface: The interface which enables commu- the problems they face. The prototype system, HAS, was
nication between users and HAS is carried out mainly in developed according to the above conceptual framework. It
menus and graphics supplemented by natural language. The was written using Visual BASICTM 6.0 (VB) and ran on a
interface of the system allows the user to specify the search- personal computer under a Microsoft WindowsTM 98 envi-
ing criteria for the hotel and to weight the importance of ronment. Microsoft WindowsTM was selected for use as the
the criteria. During the consultation, the user can adjust the operating system because of its current popularity. A thor-
position of sliders in the menu (see Fig. 6). When a param- ough review of the available software was conducted prior
eter item is selected, the crisp input value is translated into to the Inal selection of VB as a programming tool for the
the fuzzy term. The user can deIne the relative importance prototype. VB for Windows was chosen because it is an
of each criterion directly from the menu by choosing the easy-to-learn and easy-to-use graphical user interface (GUI)
buttons. Through the window interface (Fig. 6), therefore, programming language which allows rapid prototype devel-
a combination of the following modiIcations can be per- opment. The software and hardware requirements for HAS
formed: (a) supply of a new parameter value and (b) modi- are as follows:
Ication of the importance (weight) associated with a certain
parameter. The hotels recommended are communicated as (i) Hardware
outputs to the user through the user interface.
(ii) Database: Microsoft AccessTM is used to support the • A Pentium PC, 16 Mbytes RAM and 500 Mbytes for hard
database subsystem which maintains the necessary informa- disk
tion on each hotel. The data are extracted from internal and • A SVGA monitor with 800 × 600 resolution
external data sources [2,25,27].
(iii) Fuzzy rule base: The fuzzy rule base which is a (ii) Software
mechanism for storing fuzzy rules as expert knowledge is
based on membership functions. • Microsoft WindowsTM 98
(iv) Fuzzy inference engine: This is a core part of the HAS • Microsoft Visual BASICTM 6.0
engine that executes the inference cycle of fuzzy matching, • Microsoft Internet Explorer TM 5.0 or above
fuzzy conQict co-ordinating and fuzzy rule-Iring according • Microsoft AccessTM or above
to given facts. • Apple QuickTimeTM 3.0 or above
(v) Membership function base: The membership function
base is a mechanism that presents the membership functions Phase 11: Observe and evaluate the system.
of di5erent linguistic terms. Once the prototype system is built, testing and evaluation
When the user inputs parameters through the user inter- of the prototype system can be performed. Researchers can
face, the fuzzy inference engine performs according to the capture information on what users like and dislike and what
fuzzy rules and membership functions, using data from the the system does and does not do to meet their needs. The
database, and sends fuzzy or crisp results through the user following subsections focus on the process adopted in the
interface to the user as outputs (see Fig. 5). evaluation of the proposed fuzzy expert system.
Phase 9: Analyze and design the system. System validation. This concerns system evaluation by
Analysis and design are important parts of a system de- implementers and domain experts. All the modules of the
velopment process. Design involves an understanding of the HAS were tested for accuracy and completeness by the de-
domain being studied, the application of various alterna- velopment team during system development and debugged
tives, and the synthesis and evaluation of proposed alterna- by hotel practitioners, who had not been involved in system
tive solutions. Design speciIcations are used as a blueprint development, as they were written. We believe that HAS
for the implementation of the system [18]. Almost no work should be able to meet the real needs of the users by helping
has been done researching the design of a fuzzy expert sys- them in Inding a suitable hotel. The algorithms were tested,
tem for hotel selection. In order to begin the process and and these tests produced answers that were operationally re-
to determine user needs, we interviewed 20 tourists and 10 alistic or meaningful [28].
hotel practitioners in our study. The Irst step in the system Prototype testing. This concerns an evaluation of the
design process was to decide how the functions of the HAS system by domain experts and non-implementers. It allows
will be performed. This initial phase considered the design the user to test and evaluate the prototype and to provide
of data structures, database, user interface, and Inal output. feedback to the development team. The evaluation focuses
Phase 10: Build a prototype system. on the impressions and attitudes toward the design fea-
The procedure of building a prototype system has been tures and capabilities of the proposed system. The Irst pro-
widely used in software engineering research [18] because totype of HAS was produced in January 1999. Initially,
basic inherent problems emerge at an early stage and can be two academicians were invited to evaluate it. Later, 35
addressed promptly. In addition new concepts of user inter- students from the part-time degree in Hotel and Catering
E.W.T. Ngai, F.K.T. Wat / Omega 31 (2003) 275 – 286 283
Table 2
Mean responses to the system evaluation by hotel practitioners/experts and potential users
Management and students from the degree of Tourism Man- (2) Hotel experts’ evaluation
agement at The Hong Kong Polytechnic University, were Evaluations by domain experts help to determine the
invited to participate in the evaluation. At the evaluation ses- accuracy of the embedded knowledge [30], and the con-
sion, the system prototype was demonstrated and feedback sistency and completeness of responses. The system was
was solicited from the students through discussion and an evaluated by two academics and 23 part-time bachelor de-
evaluation form. In addition interviews were conducted to gree students, in Hotel and Catering Management at The
determine the exact impressions and attitudes toward spe- Hong Kong Polytechnic University. These students are ho-
ciIc features of the proposed system. The students rated the tel practitioners with an average of 4 years of experience in
design features of the system as moderate and their assess- the hotel industry.
ments of HAS during the discussion were generally positive. Consolidate and compare the outcome evaluation. Con-
These comment were noted and contributed to the develop- solidation and comparison of outcome evaluation can be
ment of a reIned prototype system. achieved through a questionnaire survey. An open-ended
Outcome evaluation. The evaluation objective is to as- questionnaire is analogous to an interview in that it gives
sess the overall value of a fuzzy expert system [29]. Out- respondents an opportunity to say what they want. It is de-
come evaluation consists of two phases; the Irst phase is signed to obtain verbal comments from the subject with a
potential user evaluation. The second evaluation is domain request for examples to support each item discussed. We
expert evaluation. The outcome evaluation of the prototype particularly wanted the potential users to tell us what they
is described below: considered to be the strengths and weaknesses of the pro-
(1) Potential users’ (Tourists’) evaluation totype system, and how it should be improved. A formal
Evaluations by users help to determine the utility of the questionnaire containing both closed and open-ended ques-
system according to the following criteria: (a) its ease of in- tions was designed and consisted of three sections: (a)
teraction, (b) the extent of its capabilities, (c) its eGciency demographic data, (b) the e5ectiveness of the prototype
and speed, (d) its reliability and whether it produces useful system, and (c) the usability of the prototype system.
results [30]. We randomly selected a total of 85 potential The potential users were asked to use Ive-point scales
users, coming from 12 nations, who had just arrived at the (1 = strongly disagree, 3 = undecided; 5 = strongly agree)
Hong Kong International Airport during the period of 15 –20 to rate the following two main aspects of the prototype
May 1999, and they agreed to participate in the evaluation. system: (i) its e5ectiveness and (ii) its usability.
At the evaluation session, the system prototype was demon- The results of the analysis of the questionnaire are shown
strated and feedback was solicited through discussion and a in Table 2. The potential users and hotel experts rated the
formal questionnaire. system highly on the above two aspects with a mean score
284 E.W.T. Ngai, F.K.T. Wat / Omega 31 (2003) 275 – 286
of at least 3.7 on a Ive-point scale with ratings of 5 being e5ectiveness and usability of the prototype at a 0.05 level
‘strongly agree’, 3 being ‘undecided’ and 1 being ‘strongly of signiIcance.
disagree’. The prototype is seen to be a promising system
for supporting the selection of hotels based on the positive 3.1. Illustrative example of using the HAS
results of its evaluation. In addition, the viability of HAS
as an e5ective procedure for hotel selection has been In the following section, some examples of a dialogue be-
ascertained by the positive feedback obtained from the tween a user (a tourist) and the prototype, HAS are shown.
questionnaires. Annotations are added to give a deeper insight into the op-
Further analysis was conducted to investigate whether eration of HAS.
there is a di5erence between hotel practitioners and potential Fig. 6 shows the main menu of HAS where the items
users in the mean ratings of e5ectiveness and usability of the “Searching Criteria”, and “Importance of Criteria”, which
prototype in the prototype evaluation. The non-parametric allow the user to specify choices, can be seen. The “Help”
Wilcoxon signed-rank tests were used to examine the di5er- in the main menu provides information on relevant topics
ence between hotel practitioners and potential users in the and o5ers the user rapid assistance and easy explanations
mean rating of e5ectiveness and usability of the prototype; using hypertext as a medium.
as the survey data are in an ordinal scale, non-parametric To start the session of “fuzzy hotel search”, the user can
tests are more appropriate for testing the hypotheses. The specify the searching criteria, for example “Price”, “Facili-
di5erence between hotel practitioners and potential users ties” and “Food Type” of the hotel by adjusting the slider
in the mean ratings of e5ectiveness and usability of the accordingly. The user can proceed to deIne the importance
prototype was found to be not signiIcant (for e5ective- of the selected criteria by stating the weighting either as
ness, Z-score = −1:6036, P-value = 0:1088; for usability, “most important”, “important” or “less important”. One
Z-score = −1:8593, P-value = 0:0630). We concluded example of a selection is for “Price” “expensive”, for Facil-
that there were no signiIcant di5erences between hotel ities “some” and for “Food Type” “less”. The importance
practitioners and potential users in the mean ratings of of criteria as “most important” for “Price”, “important” for
E.W.T. Ngai, F.K.T. Wat / Omega 31 (2003) 275 – 286 285
[20] Machacha LL, Bhattacharya P. A fuzzy-logic-based approach [26] Cox E. The Fuzzy systems handbook: a practitioners’ guide
to project selection. IEEE Transactions on Engineering to building, using and maintaining fuzzy systems. San Diego,
Management 2000;47(1):65–73. CA: AP Professional, 1999.
[21] Tsang EF, Ngai EWT. EXSGACM: an expert system for gas [27] Hong Kong Tourist Association. URL: http://www.hkta.org/
crisis management. Expert Systems with Applications: An hotels/index.html
International Journal 1996;10(1):75–90. [28] Metersky ML. A decision-oriented approach to system design
[22] Ngai EWT, Cheng TCE. MSS4TQM: a management support and development. IEEE Transactions on Systems, Man and
system for TQM. IEEE Transactions on System, Man and Cybernetics 1993;23(4):1024–37.
Cybernetics 1999;29(3):371–86. [29] O’Keefe RM, Balci O, Smith EP. Validating expert system
[23] Chu RKS, Choi T. An importance-performance analysis of performance. IEEE Expert 1987;2(4):81–90.
hotel selection factors in the Hong Kong hotel industry: [30] Gasching J, Klahr P, Pople H, Shortli5e E, Terry A. Evaluation
a comparison of business and leisure travellers. Tourism of expert systems: issues and case studies. In: Hayes-Roth F,
Management 2000;21:363–77. Waterman DA, Lenat DB, editors. Building expert systems.
[24] Hong Kong Tourist Board, Hong Kong Tourism Board Hotel Reading, MA: Addison-Wesley, 1983. p. 241–80.
ClassiIcation System 2001. Hong Kong, 2001.
[25] The OGcial Hotel Guide. Hong Kong Tourist Association,
1998.