Expert Systems with Applications 39 (2012) 3257–3264

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa

Design and implementation of an intelligent recommendation system

for tourist attractions: The integration of EBM model, Bayesian network
and Google Maps
Hsu Fang-Ming a,1, Lin Yu-Tzeng a,b,⇑, Ho Tu-Kuang c,2
a
Department of Information Management, National Dong Hwa University, 1, Sec. 2, Dahsueh Rd., Shoufeng, Hualien 974, Taiwan, ROC
b
Department of Hospitality Management, Taiwan Hospitality & Tourism College, 268, Chung-Hsing St., Shoufeng, Hualien 974, Taiwan, ROC
c
Department of Tourism, Taiwan Hospitality & Tourism College, 268, Chung-Hsing St., Shoufeng, Hualien 974, Taiwan, ROC

a r t i c l e i n f o a b s t r a c t

Keywords: Selecting tourist attractions and collecting related site information is one of the most crucial activities for
Tourist attractions a tourist when making decisions for a trip. Although various recommendation systems have been dis-
EBM model cussed over the last decade, rarely do such systems take individual tourist preference information into
Bayesian network consideration. Based on the Engel–Blackwell–Miniard (EBM) model, this study used data published by
ROC curve
the Tourism Bureau of Taiwan to develop a decision support system for tourist attractions. The probabil-
Google Maps
ity of a tourist attraction appealing to a particular tourist is calculated utilizing a Bayesian network, and
the accuracy of the prediction is validated by a ROC curve test. Finally, recommended routes and tourist
attractions are presented through an interactive user interface using Google Maps. This study confirms
that by combining the EBM model with a Bayesian network to propose a decision support system called
the Intelligent Tourist Attractions System (ITAS). It has demonstrated good prediction of tourism
attractions and provides useful map information to tourists.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction The EBM model involves six phases to explain a customer’s

decision process: problem recognition, information search, alter-
Selecting tourist attractions is a critical problem in planning for natives evaluation, purchase decision, purchase and post-purchase
a trip (Huang & Bian, 2009). However, when it comes to visiting an evaluation (Blackwell, Miniard, & Engel, 2001). This process is sim-
unfamiliar place, tourists’ travel preferences are not explicitly ilar for most consumers, yet different needs, as well as other inter-
known and often hidden when they start to plan a trip (Loh, nal and external factors, can create indecisive results when
Lorenzi, Saldana, & Licthnow, 2003; Viappiani, Pu, & Faltings, studying trends in purchase decision making. A one-to-one person-
2002). Although various online travel recommendation systems alized recommendation system seeks to eliminate these indecisive
that provide useful internet resources for users to plan their trips results and to present purchasing information more relevant to the
have been developed during the last decade, few systems focus individual consumer, ultimately providing a recommendation sys-
on customized attraction recommendations for individual tourists. tem that most meets the needs of an individual (Good et al., 1999;
In order to recommend satisfactory tourist attractions to individual Resnick & Varian, 1997). In recent years, online recommendation
travelers, this paper provides an outline for an intelligent personal- systems for travel have been able to assist travelers in deciding
ized recommendation system design which can significantly on suitable travel plans and routes, and have been increasingly
reduce unnecessary additional costs for users in the information brought to the attention of scholars in the tourism field (Huang
search process, and best meet their needs and preferences when & Bian, 2009; Loh et al., 2003; Ricci, 2002; Wallace, Maglogiannis,
selecting tourist attractions. Karpouzis, Kormentzas, & Kollias, 2003). When a traveler selects a
travel plan, the process involves several phases, including the
selection of destinations and attractions, the choosing of accom-
⇑ Corresponding author at: Department of Information Management, National modations, plotting routes, etc. A tourist attraction is often the
Dong Hwa University, 1, Sec. 2, Dahsueh Rd., Shoufeng, Hualien 974, Taiwan, ROC. main motivation for a tourist to decide upon a destination (Jafari,
Tel.: +886 3 865 3906; fax: +886 3 865 3910.
2000; Richards, 2002). However, to build a personalized recom-
E-mail addresses: [email protected] (F.-M. Hsu), [email protected]
(Y.-T. Lin), [email protected] (T.-K. Ho).
mendation system, a wide array of real-time information on tourist
1
Tel.: +886 3 8633107; fax: +886 3 8633100. attractions must be included (Ardissono, Goy, Petrone, Signan, &
2
Tel.: +886 3 865 3906; fax: +886 3 865 3910. Torasso, 2003). This data is collected mostly from providers with

0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.09.013
3258 F.-M. Hsu et al. / Expert Systems with Applications 39 (2012) 3257–3264

vastly diverse backgrounds, and therefore cannot be easily inte- 2. Attribute-based recommender system: This type of recom-
grated. Furthermore, when a traveler begins to consider a trip, he mender system provides to the user information or products
or she has less than evident inclinations, especially when planning that matches the user’s preferences.
to visit a foreign location (Loh et al., 2003). Therefore, the purpose 3. Item-to-item correlation recommender system: This type of
of this study is to eliminate for travelers the uncertainties involved recommender system analyzes historical data to find connec-
in the information search stage of a buyer’s decision process, while tions between frequently purchased items and displays recom-
also avoiding unnecessary costs. To ensure the integrity, accuracy mendations for correlated items. This study utilizes this type of
and practicality of the ITAS, this study follows the following proce- recommender system.
dure: (1) extract measures from EBM model for tourist attractions; 4. People-to-people recommender system: This type of recom-
(2) collect data from ‘‘2007 Annual Survey Report on Visitors mender system classifies users into groups based on preference
Expenditure and Trends in Taiwan’’ published by the Tourism Bu- of products, interests, or needs and recommends appropriate
reau of Taiwan; (3) calculate probability of a tourist attraction items to users within the group according to peer review of
appealing to a particular tourist by utilizing Bayesian network; products or information.
(4) verify accuracy of the prediction by ROC curve test; and (5)
present recommended routes and tourist attractions through The above classification shows that the main purpose of a rec-
system with Google Maps. ommendation system is generally to analyze and compare a user’s

Table 1
2. Literature review Inbound tourists to Taiwan in 2007.

Demographic variables No. %

2.1. EBM model as a decision model
Gender Male 1176 48.4
Female 1253 51.6
The whole buying process is a continuous course of actions
Age Less than 29 721 29.7
rather than numerous isolated events from the viewpoint of EBM 30–49 1057 43.5
model to buyer behavior. A buying process is composed of six 50 or above 651 26.8
stages: problem recognition, information search, alternatives eval- Education Elementary 53 2.2
uation, purchase decision, purchase and post-purchase evaluation, Middle 590 24.5
and the buying decision is finally formed by factoring in internal College or higher 1769 73.3
and external influences that affect the consumer (Blackwell et al., Annual income Less than US$29,999 844 34.7
2001). Many scholars argue that consumer’s behavior is a process US$30,000 to US$69,999 606 24.9
of information processing to make a decision (Blackwell et al., More than US$70,000 343 14.1
Unsteady Income 636 26.2
2001; Peter & Olson, 2008; Solomon, 2007; Walters & Blaise,
1989), and various decision making models have been developed Vocation Manager 287 11.8
Technician 756 31.1
on this basis. Three of the most popular and practical models are
Clerk 553 22.8
the Nicosia model (Nicosia, 1968), the Howard–Sheth model Service marketer 136 5.6
(Howard & Sheth, 1989), and the EBM model (Blackwell et al., Housekeeping 209 8.6
2001). Of the three, the EBM model is currently the most compre- Student 272 11.2
Other 216 8.9
hensive behavioral model, and has the following strengths: it has a
Nationality Japan 779 32.1
mature framework, it can be sequenced and factorized, and it is China 721 29.7
able to easily analyze core problems (Blackwell et al., 2001; Korea 224 9.2
Howard & Sheth, 1989; Teo & Yeong, 2003; Traci & Trevor, 2006). Europe or U.S. 347 14.3
Besides, tourist’s behavior is also a process of decision making, Other Asian countries 358 14.7

and the selection of travel destinations or recreational activities

is the purchase of an ‘‘intangible product’’ (Chen, Yung, & Chen,
2004; Lin, Sung, & Chen, 2006). Therefore, this study uses the Table 2
EBM model as a basis to develop a decision support system for Traveling purpose, information source and traveling type.

tourist attraction selection. Items Content No. %

Traveling Sightseeing 1772 73.0
2.2. Recommendation system purpose Business 268 11.0
Visiting family or friend 204 8.4
A recommendation system is an information filtering mecha- Attending international conference or 49 2.0
exposition
nism that utilizes machine learning to analyze an item or compare
Study 31 1.3
it with the past experiences of others, and suggests or recommends Other 105 4.3
to the user potentially needed information, services, or products.
Information Television 1906 78.5
Such a system matches an item to users based on preference, inter- source
est, action, or need, and can significantly reduce unneeded addi- News paper and magazine 1861 76.6
tional costs for users in the information search process (Rashid Internet 1827 75.2
et al., 2002). According to their characteristics, decision support Light box advertisement on train and bus 1027 42.3
International travel fair 996 41.0
systems with recommendation functions can be generalized into Outdoor billboard 934 38.5
one of the following four categories (Schafer, Konstan, & Riedl,
Tour type Group inclusive tour 1167 48.0
1999): Independent tour 749 30.8
Flight and accommodation tour booked by 313 12.9
1. Non-personalized recommender system: This type of recom- agency
mender system does not take into account user-provided infor- Personally planned tour arranged by local 178 7.3
agency
mation or products. It displays all available information to the
Tour arranged by local agency after arrival 22 0.9
user and the exact same results are displayed for any given user.