You are on page 1of 14

ARTICLE IN PRESS

Hospitality Management 26 (2007) 10051018


www.elsevier.com/locate/ijhosman

Language representation of restaurants: Implications


for developing online recommender systems
Zheng Xianga,, Sang-Eun Kimb, Clark Huc, Daniel R. Fesenmaierd
a

School of Merchandising and Hospitality Management, University of North Texas,


P.O. Box 311100, Denton, Texas 76203, USA
b
National Laboratory for Tourism & eCommerce, School of Tourism & Hospitality Management,
Temple University, 1700 N. Broad Street, Suite 201F, Philadelphia, PA 19122-0843, USA
c
National Laboratory for Tourism & eCommerce, The Fox School of Business and Management, School of Tourism
& Hospitality Management, Temple University, 201, 1700 N. Broad Street, Philadelphia, PA 19122-0840, USA
d
Director, National Laboratory for Tourism & eCommerce, School of Tourism and Hospitality Management,
Temple University, 1700 N. Broad Street, Suite 201C, Philadelphia, PA 19122, USA

Abstract
As a marketing tool recommender systems have the potential to provide relevant and highly
personalized information to consumers. However, developing effective recommender systems
requires a substantive understanding of consumers preferences as well as meaningful ways to
represent hospitality and travel products. This paper argues that language holds the key to
understanding consumer preferences and therefore developing effective online recommender systems.
Specically, it explores the nature of the language used by consumers to describe their dining
experiences in contrast to the language used by restaurant websites. The ndings indicate that
consumers use substantially different vocabularies from restaurant websites to describe dining
experiences. This study provides implications for developing online recommender systems for
restaurants as well as general hospitality and travel products.
r 2007 Elsevier Ltd. All rights reserved.
Keywords: Restaurant; Consumer language; Recommender systems; Consumer mapping; Text analysis

Corresponding author. Tel.: +1 940 565 2436; fax: +1 940 565 4348.

E-mail addresses: philxz@gmail.com (Z. Xiang), sekim713@temple.edu (S.-E. Kim), clark.hu@temple.edu


(C. Hu), drfez@temple.edu (D.R. Fesenmaier).
0278-4319/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijhm.2006.12.007

ARTICLE IN PRESS
1006

Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

1. Introduction
As the Internet continues to become an essential part of everyday life, how to deliver
relevant and highly personalized information to both potential and existing customers has
become an important task for hospitality and tourism businesses. Numerous hospitality
and tourism-related businesses and marketing organizations have employed a variety of
approaches to provide travelers online information about their products and services with
the goal of facilitating trip planning. One common approach is to provide recommendations to travelers with respect to accommodation, activities, and even destinations (Burke,
2002; Fesenmaier et al., 2006; Riedl et al., 2002). In the hospitality sector, online businesses
have also extensively adopted this approach by offering consumers options to customize
their dining experiences. For example, online restaurant advisory websites such as
Zagat.com, Restaurants.com, and others not only allow customers to search for desired
restaurants, but also explicitly recommend certain restaurants to customers.
One of the limitations of these websites is that restaurants are often recommended
simply based upon a limited number of functional attributes such as price, cuisine, and
location and thus there is a lack of adequate representation of the holistic experience of
dining. Restaurants, and travel products in general, are arguably experience goods in
that full information on certain attributes cannot be known without the direct experience
(Klein, 1998). That is, when looking for information about travel products, a consumers
information search behavior is driven not only by functional and utilitarian needs but also
by many social and hedonic needs (Vogt and Fesenmaier, 1998). Considering dining out is
oftentimes a social behavior involving explicit or implicit identity construction and image
building (Finkelstein, 1989), a fundamental difculty in making online recommendations
lies in understanding what consumers really want in order to identify the products that can
possibly match their preferences. Thus, developing effective and meaningful solutions to
online recommendation of restaurants remains a challenging task.
This paper argues that understanding the language representation of restaurants can
provide a useful basis for developing effective online recommender systems. The success of
online recommendations hinges on a substantive understanding of, and the ability to
capture, consumer preferences (Schafer et al., 2001). Given the experiential nature of
dining, consumers language can be used as the key to understanding their perceptions of
restaurant attributes and making inferences about their preferences. Recent research on
online travel information search has revealed that the information provided by travel
service providers can be substantially different from what travelers are looking for (Pan
and Fesenmaier, 2006). This suggests that there exists a great potential to improve the
online representation of travel-related products. Thus, the goal of this paper is to explore
the nature of the language consumers use to describe their dining-out experience in
contrast to what is portrayed by restaurants online descriptions. Text analysis techniques
were used to examine the similarities and differences between the languages used by
consumers and restaurant owners. The ndings of this research provide useful implications
for the development of effective online restaurant recommender systems.
2. Literature review
A recommender system in the online environment represents a value-added process
because it helps solve problems related to information overload in consumers

ARTICLE IN PRESS
Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

1007

information search process (Furner, 2002; Schafer et al., 2001). However, the challenges
for online recommender systems lie in that consumer preferences are highly dependent
upon the nature of the product. An effective online recommender system must be based
upon an understanding of consumer preferences and successfully mapping potential
products onto the consumers preferences (Adomavicius and Tuzhilin, 2005). Following
from Gretzel et al. (2004) and Pan and Fesenmaier (2006), it is argued that this can be
achieved through the understanding of how consumers describe in their own language a
product, a place, and the experience when consuming the product or the place.
2.1. Modeling consumers restaurant preferences
Traditionally, choosing a restaurant has been considered as rational behavior whereby a
number of attributes contribute to the overall utility of a restaurant (Auty, 1992; Kivela
et al., 1999). For example, food type, food and service quality, image and atmosphere of the
restaurant, and availability of information about a restaurant, play an important role at
different stages in consumers choice-making (Auty, 1992; Kivela, 1997). While food quality
and food type have long been perceived as the most important variables for consumers
restaurant selection, soft attributes such as situational and contextual factors have been
found to be important determinants (Baker, 1986; Bell and Meiselman, 1995; Bruner, 1990;
Kivela, 1997). It is considered that the relative importance of traditional attributes of
restaurant product (i.e., price, food type, etc) changes with the restaurant situational and
contextual factors (Auty, 1992). Based upon these results, Kivela (1997) identied four
distinct types of restaurants, i.e., ne dining/gourmet, theme/atmosphere, family/popular,
and convenience/fast-food. Auty (1992) identied four types of dining-out occasions, namely
celebration, social occasion, convenience/quick meal, and business meal.
The difculty in modeling restaurant choice is caused by the fact that dining-out
experience often involves many socio-psychological aspects that go beyond the simple
functional characteristics of restaurants or food services. According to some researchers
(e.g., Caplan, 1997; Finkelstein, 1989), dining-out is essentially a social experience whereby
individuals seek to construct, reify, or reinforce their socio-cultural identities. The old
proverb we are what we eat underscores the importance of dining in the construction of
different cultures and self-image. Eating is no longer a simple biological activity for human
existence, and the consumption of food fundamentally structures social values and personal
identities. Indeed, recent research in restaurant choice modeling has indicated that a
high-quality meal and outstanding service are no longer considered the main factors evoking
pleasant and memorable dining experiences (Hanefors and Mossberg, 2003; Pine and
Gilmore, 1999). Caplans research (1997) clearly shows that social and personal identities
and perceptions of health risk can substantially inuence consumers restaurant choices.
2.2. Approaches to online recommendations
Online recommender systems can provide efcient and effective ways to match
consumers preferences over certain products. Generally, the main goal of a recommender
system can be described as the selection of an item from a the set of all possible items such
that it maximizes a specic users total utility (Adomavicius and Tuzhilin, 2005; Furner,
2002). Many approaches to making recommendations have been developed as e-commerce
applications, ranging from very simple retrieval or ltering applications (e.g., providing a

ARTICLE IN PRESS
1008

Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

list of the top 10 restaurants) to extremely complex ones which rely on some form (indirect
or direct) of elicitation of users preferences (Burke, 2002; Furner, 2002; Riedl et al., 2002).
Collaborative ltering is a commonly used indirect approach whereby preferences are
constructed through modeling various online user behaviors. It focuses on identifying
individuals preferences and recommends items that people with similar preferences have
purchased in the past (Goldberg et al., 1992; Riedl et al., 2002). A well-known example of
systems based on collaborative ltering is Amazon.com, which suggests products to
customers by hinting that customers who bought this item also boughty (www.
amazon.com). Recommender systems can also use a content-based approach, which learns
a prole of the users interests based on the features present in items the user has rated
(Belkin and Croft, 1992). Basically, it relies on the customers past history to infer his/her
preferences and thus the customer will be recommended with items similar to those he/she
purchased in the past. Other approaches such as case-based reasoning are usually hybrids
that leverage both individual user preferences and collaborative similarities (c.f., Burke,
2002 for a comprehensive review). For restaurants, Burke (2000) developed a system called
Entree which uses a knowledge-based approach to make recommendations by nding
restaurants in a new city similar to the restaurants the user knows and prefers. It enables
the user to interact with the system by stating his/her preferences of a given restaurant and
rening search criteria by critiquing the systems suggestions. In addition to these modeling
approaches, Schafer et al. (2001) suggested that other factors such as the type of user input,
the existence and type of community, the presentation of the recommendation, and the
degree of personalization, are important in making effective online recommendations.
Making recommendations for travel and hospitality products in an online system is
considered a more complex task than general consumer goods (Fesenmaier et al., 2006;
Ricci, 2002). First, travel involves bundling of a large number of heterogeneous products
and services and thus, applying the content-based approach requires extensive domain
knowledge to be built for the particular application. Second, because the consumption of
travel products is individual-based and context-specic, it cannot be assumed that two
identical trips will result in the same experience, even if the two travelers go to the same
destination and visit the same attractions. In the case of restaurant services, it is difcult to
generalize an individuals dining-out experience to others because his/her experience has
been shaped by many contextual and situational factors. As such, the uniqueness of travel
and hospitality products poses a challenge for developing recommender systems. To
address this issue, a commonly used approach focuses on developing effective
communicative interactions between a user and a system. For example, Gretzel et al.
(2004) proposed an approach based upon the elicitation of individuals personality traits;
Ricci et al. (2005) proposed a method called recommendation by proposing aiming at
travelers by presenting images and simple descriptions of products to users.
Among these approaches, language representation of travel products has been
increasingly considered an important issue in developing travel recommender systems,
and online hospitality and tourism marketing in general. This recognition is rooted in the
understanding that travel and tourism has a discourse of its own which includes acts of
promotion as well as accounts of travelers (Dann, 1997). The language used to target
tourists and the language used by tourists form the basis of interpreting travel products
and tourist experiences (Gretzel et al., 2006). Recent research on online travel information
search (e.g., Pan and Fesenmaier, 2006) has found that online tourism information uses
signicantly different languages. That is, a large amount of marketing-oriented content is

ARTICLE IN PRESS
Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

1009

focused on the selling of travel products, while travelers use more subjective and
experiential language to describe their perceptions and expectations when searching for
travel products. The gap between the promotion of travel products and travel information
search indicates that marketers need to utilize different forms of communication to enable
travelers to express their need for information that is framed within their personal context.
From the system design point of view, Gretzel and Fesenmaier (2002) argued that the
existing approaches in tourism information systems, which generally rely on numeric data,
cannot capture the holistic experiential aspects of travel. They suggested that narrative
(story-telling) logic should be incorporated into online recommender systems to assist trip
planners to make sense of the world. According to them, narratives can better support the
generation of mental imageries and thus, provide guidance in terms of interpreting search
results and evoking imagination. More recently, Loh et al. (2003) proposed an approach to
mine customers textual messages when interacting with an online travel agent in order
to understand their preferences. From these studies, it is clear that the understanding of
how consumers perceive and describe travel products can help develop useful ways for the
representation of restaurant services in recommender systems.

3. Research questions
Based upon the literature reviewed, it is argued that online restaurant recommender
systems must be able to speak the same language as restaurant customers. Thus, the goal
of this study was to explore the nature of the language consumers use to represent their
preferences of restaurants as opposed to the language used by restaurant owners and
marketers on their websites. Specically, this study aimed to achieve two objectives: (1) to
identify and compare the vocabularies used by both consumers and restaurant websites in
describing the dining experience; and, (2) to understand the effect of situational factors
such as restaurant type and dining-out occasion on the language used by consumers to
describe their dining experiences. Hence, two research questions were formulated:
Q1. What is the nature of the languages used by consumers and restaurant websites in
describing a dining experience? How are they similar to and/or different from each
other?
Q2. To what extent does the nature of the restaurant type and the dining-out
occasion effect consumers language representation of their dining experience?

4. Method
This study was conducted in three stages. First, a pilot study was conducted to identify
the specic restaurants to be included in the study. Second, two types of language data
were collected: (1) consumers verbal representation of their dining experience at the
restaurants identied in the rst step; and, (2) descriptive text included on the websites
owned by the restaurants identied above. Third, text analysis as well as multivariate
analysis techniques were employed to compare the languages used by consumers and
restaurants and to assess the effect of dining-out occasion and restaurant type on consumer
languages (Hair et al., 1998; Woelfel and Woelfel, 1997).

ARTICLE IN PRESS
1010

Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

4.1. Pilot study


Twenty three college students in a northeastern metropolitan area in the United States
were asked to identify 10 restaurants where they had previous dining experience or about
which they had second-hand information. In order to obtain descriptions of the dining
experiences at these restaurants, respondents were asked to indicate the type of restaurant
for each restaurant they provided using one of the three types of labels: casual, family,
and ne dining. The results of the pilot study indicated that most respondents labeled the
restaurants as either casual or ne dining. The label family was dropped because only a
very small number of respondents actually applied it to the restaurants they identied. The
results also revealed that a large number of restaurants were labeled as both casual and
ne dining by different respondents, suggesting that there existed a fuzzy set among
these restaurants. Thus, a new label called overlapping was added as a third restaurant
type, which represents the blurring that may occur between casual dining and ne dining
categories (Goldman, 1993). Based on these preliminary results, nine restaurants
representing the three types were included in the subsequent data collection procedure
with each type containing three restaurants that were most frequently and distinctly
mentioned by respondents.
4.2. Collection of language data
Two types of language data describing the restaurants were collected. First, a
questionnaire was used to elicit consumers description of the nine restaurants identied
in the pilot study. One hundred and nine students from the public university located in the
same metropolitan area participated in the study in exchange for course credit in late 2004.
They were asked to describe their dining experience at these restaurants using up to ve
simple phrases or sentences for each restaurant. In addition, each student was randomly
assigned to one of three different dining-out occasions (namely entertainment, familydining, and romantic dinner), which were considered the most common occasions for
dining-out through the literature review earlier. Second, the textual contents (except the
navigational and functional elements) from the websites owned by the nine restaurants
were extracted. These textual contents were then used as the language representation of the
restaurants by the restaurant owners. Two text les were generated with one containing
consumers language data and the other the descriptive texts from the restaurant websites.
In the consumers data le, each case, i.e., the ve phrases or sentences used by one
respondent, was associated with two additional identiers, namely restaurant type
(i.e., casual, ne dining, and overlapping) and dining-out occasion (i.e., entertainment, family-dining, and romantic dinner).
4.3. Language data analysis
Before performing the text analysis, the two text les were pre-processed using the
following three steps: (1) removed stop words such as a, the, and, and we in order to
include only meaningful words in the analysis; (2) replaced plurals with singles and past
tense with present tense so that those words with the same linguistic roots (drinks,
drank, and drink/ eats, ate, and eat) can be treated as the same; (3) used a single

ARTICLE IN PRESS
Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

1011

token to represent all synonyms and proper names, e.g., waitstaff for synonyms such as
server, waiter, waitress, and staff, and Olive Garden to OliveGarden.
CATPAC (Woelfel and Woelfel, 1997) was then used to explore the nature of, and the
differences between, the languages used by consumers and restaurant websites in
describing a dining experience by extracting the unique words used in both text les.
Descriptive statistics were obtained to provide a comparison between consumers language
and restaurant websites language based upon word frequencies. In particular, the most
frequently used unique words in both textual les were extracted and presented side-byside to provide a comparative view of the commonalities and differences in the language
representations of restaurant attributes. Correspondence analysis was then employed to
examine the association between the unique words used and different dining-out scenarios.
Each scenario was a combination of a specic restaurant type and a dining-out occasion
and thus, nine scenarios were generated (i.e., 3 restaurant types  3 dining-out occasions).
The analysis used the frequency matrix for the top 160 most frequently used unique words
(rows) for each scenario (columns). A correspondence map was then constructed to
visualize the associative relationships in a two-dimensional space.
5. Results
Table 1 provides the basic descriptive statistics of the language data collected from both
consumers and restaurant websites. As can be seen, the original text le from the student
respondents consisted of a total of 9924 words and the nine restaurant websites consisted
of a total of 7219 words. After pre-processing (e.g., tokenization), there were 9893 words in
the consumer data le and 7211words in the restaurant website data le, respectively.
There were 1216 unique words in the consumer data le, representing 12% of the total
number of words; in contrast, there were 1904 unique words in the website data le,
representing 26% of the total number of words. The percentage of unique words was
obtained as the ratio between the number of unique words and the total number of words
in the original data le. This result indicates that, overall, restaurant websites used
substantially more diverse vocabularies than consumers in describing the restaurant
products and dining experience.
CATPAC was used to deconstruct the two text les by identifying the unique words
along with their frequencies in the two types of language data. The consumer language
data was further differentiated by different types of restaurants, i.e., casual, ne dining,
and overlapping. As shown in Table 2, consumers and restaurant websites share a total of
428 words, while consumers used 788 words that were not used by restaurant websites and
Table 1
Descriptive statistics of the two language datasets
Type of data

Original data
Pre-processed data
Unique keywords (percentage*)

Number of words
Consumer

Restaurant

9924
9893
1216 (12%*)

7219
7211
1904 (26%*)

*The number in the parenthesis indicates the percentage of unique keywords contained in the original data.

ARTICLE IN PRESS
Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

1012

Table 2
A comparative view of the frequencies of unique words in the language representations by both consumers and
restaurant websites
Restaurant type Number of unique words

Casual
Fine dining
Overlapping
Total

Consumer

Restaurant

604
588
582
1216*

385
1167
648
1904*

Number of common words Number of different words (%)

101
208
160
428*

Consumer

Restaurant

503
380
422
788*

284
959
488
1476*

(83.2%)
(64.6%)
(72.5%)
(64.8%*)

(73.8%)
(82.2%)
(75.3%)
(77.5%*)

*Note that the total number does not match the summated value of all three categories, since the number of
duplicated words among the categories were subtracted.

restaurant websites used 1476 words that were not used by consumers. This observation
supports the previous nding that restaurants use a much richer language than consumers.
Interestingly, the nature of the language used varied substantially among different types
of restaurants. As shown in Table 2, while consumers used roughly the same number of
unique words to describe restaurant attributes and dining experiences (604, 588, and 582
for casual, ne dining, and overlapping, respectively), there existed a substantial difference
in the language used by restaurant websites for difference types of restaurants. The number
of unique words used for casual and overlapping restaurants are 385 and 648, respectively.
However, the websites of ne dining restaurants used 1167 unique words to describe their
products. In addition, the numbers of words that were used only by consumers were very
close for the three types of restaurants (503, 380, and 422 for casual, ne dining, and
overlapping, respectively). However, websites of ne dining restaurants used a
substantially larger number of words (959) that were not used by consumers, as opposed
to the other two types of restaurants (284 for casual and 488 for overlapping). This nding
indicates that websites of ne dining restaurants use more elaborate approaches in
describing their products than the other types of restaurants.
To further analyze the commonalities and differences between the two types of language,
the unique words that were commonly shared by consumers and restaurant websites as
well as those used exclusively by either consumers or restaurant website, were extracted
and presented side-by-side to provide a comparative view (see Fig. 1). Due to the limitation
of space, only the top 50 unique words are listed: (1) those used by both the consumers and
restaurant websites (column 1); (2) those used only by consumers (column 2); and, (3)
those used only by the restaurant websites (column 3). The rows of the table represent the
three types of restaurants included in the study: casual, ne dining, and overlapping.
All of the common and different words are rank ordered by their frequencies in the
datasets. As shown in Fig. 1, the words most commonly shared by consumers and
restaurant owners were words such as food, atmosphere, service, Italian, and
restaurant, which represent the general attributes of restaurant products. Consumers
and restaurant websites used a large number of distinctively different keywords, most of
which were related to the restaurant attributes perceived as important by either party. For
example, regardless of restaurant type, consumers were concerned about attributes such as
the value for money and service quality of the restaurants (represented by words such as
price, expensive, inexpensive, excellent, and nice). However, restaurant websites

ARTICLE IN PRESS
Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

1013

Fig. 1. A comparative view of the top 50 words in the language representations of both consumers and restaurant
websites.

focused more on the types of products and the services available (represented by words
such as platter and offer) or the features that distinguish one restaurant from others
(represented by words such as collection, history, chef, and Havana). Interestingly,

ARTICLE IN PRESS
1014

Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

Table 3
Summary of the correspondence analysis
Dimension

1
2
3
4
5
6
7
8

Eigenvalue

0.208
0.088
0.066
0.056
0.037
0.029
0.020
0.019

Proportion of inertia
Accounted for (%)

Cumulative (%)

39.708
16.815
12.598
10.754
7.087
5.603
3.888
3.546

39.708
56.523
69.121
79.875
86.962
92.566
96.454
100.000

w2: 3211.753 (observed), 1397.379 (critical) with 1312 degrees of freedom; one-tail po0:0001.

consumers used more adjectives and nouns, while restaurant owners used more verbs such
as offer, continue, learn, try, and help, which represents an active tone of persuasion.
Table 3 provides the summary results of the correspondence analysis and indicates that
there were eight non-trivial dimensions which could be used to fully explain the variance in
the data. The eigenvalue (0.208) of the rst dimension explained only 39.7% of the total
variance, while the second dimension accounted for an additional 10.8% of the total
variance. The rst two signicant dimensions were used to construct the correspondence
map to facilitate further discussion.
Fig. 2 provides the correspondence map to represent the associations between the top
160 frequently used words and the nine dining-out scenarios in a two-dimensional space. It
shows that three distinct groups exist among the nine dining-out scenarios: Casual
restaurants (represented by labels CE, CR, and CM) are close to one another than to
other types of restaurants (i.e., ne dining, labeled as FE, FR, and FM; and,
overlapping, labeled as OE, OR, and OM), and vice versa. This suggests that the
dining-out occasions, namely entertainment, romantic, and family dining, did not
generate substantially different language representations of restaurant attributes. When
holistically examining the correspondence map, the displayed 160 unique words seemed to
form four clusters, including: (1) a set of words (enclosed in the central eclipse) were
commonly shared by consumers descriptions of the three types of restaurants, including
food, menu, nice, delicious, dinner, dining, cuisine, good, bar, wine, and so
forth; and, (2) other words were closely associated with each type of restaurant, indicating
a degree of uniqueness when consumers chose the words to describe the restaurant
products. For example, words such as casual, affordable, and variety are closely
associated with the casual restaurant type; words such as steak, wonderful, and
upscale appear to be associated with the ne dining restaurant type; last, words such as
tasty, international, portions, and entertainment are associated with the overlapping
restaurant type.
The correspondence map also shows there are words that fall in between two clusters of
restaurants. For instance, words such as italian, location, need, city, and scene (in
Quadrant I) are approximately equally distant from ne dining and overlapping
restaurants; words such as amazing, theme, family, beer, and asian (in Quadrant II)
are equally associated with overlapping and casual restaurants. These words are used by

ARTICLE IN PRESS
Z. Xiang et al. / Hospitality Management 26 (2007) 10051018
1.5

1015

II

Dimension 2 (16.8%)

0.5

-0.5

-1

III
-1.5
-1.5

IV
-1

-0.5

0.5

1.5

Dimension 1 (39.7%)

Fig. 2. Correspondence map for the top 160 key words represented with nine dining-out scenarios.

consumers to describe both types of restaurants and thus, represent the common space
in consumers perceptions of two kinds of dining experiences. Interestingly, ne dining
and overlapping restaurants share very few words in common, suggesting that these two
types of restaurants are perceived distinctively by consumers. In addition, there are words
that are distributed far from the center of the map, e.g., clean and crowded in Quadrant
II, spanish, grilled, and money in Quadrant III, and trendy, dressup, and ambience
in Quadrant IV. These words are less frequently used by consumers; however, they are
more closely associated with a specic restaurant type than others. These words appear to
represent the idiosyncrasies of restaurant type perceived by consumers. For instance,
words such as trendy, dress-up, and ambience reect consumers perceptions of ne
dining restaurants.
6. Discussion and conclusions
As an online marketing tool recommender systems guide behavior by suggesting
products or information relevant to consumers (Fesenmaier et al., 2006; Riedl et al., 2002).
This paper argues that understanding the language consumers use to describe restaurants
is essential to developing effective online recommendations. The ndings indicate that
consumers and restaurant websites, indeed, use very different languages to describe the
same restaurant products. In addition, the effect of different dining-out scenarios on the

ARTICLE IN PRESS
1016

Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

consumer language was also examined, revealing that, while there are commonalities
among dining-out occasions, substantial differences exist between restaurant types.
Through investigating the nature of consumer language, particularly the uniqueness and
idiosyncrasies in the ways consumers represent their dining experiences, the ndings offer
several important implications for developing online recommendations.
First, most of the existing online recommender systems have been built based upon the
information provided by producers. Indeed, the results of this study show online
restaurant websites, which are provided by restaurant owners or advertisers, largely ignore
the axiom in system design and content provision; that is, to speak the same language as
consumers (Dann, 1997). A direct negative consequence of this failure is that consumers
will nd it difcult to locate the restaurant they are looking for (Nielsen, 2006).
Second, variety seeking is especially pronounced in hospitality and travel. In the case of
restaurant service, consumers can be novelty or sensation seeking when looking for
restaurants. Consumers not only spend their time and money on eating, but also will use
the dining-out occasion to fulll heterogeneous needs. From a recommendation point of
view, it is important to provide messages relevant to the specic needs of the consumer in
search of an experience. Therefore, restaurant owners or online marketers should address
consumers experiential needs by delivering highly relevant and even personalized cues.
For example, using the vocabularies identied in consumers language, a restaurant website
could provide messages similar to the following conjectural narratives to mimic a
consumers way to express his/her preferences when searching for a restaurant (note: the
words in bold font type were identied as exclusively used by consumers in this study):
Id like to have good and nice food with my family in a casual restaurant that is
inexpensive. Also, I hope I can receive fast services from their friendly staff
for a casual restaurant; or,
This restaurant is upscale with high quality food and a great atmosphere. I bet people
dining there are trendy and classy and they are probably very similar to us.
for a ne dining restaurant.
The stories told by consumers, which span a variety of media and emerge in different
contexts of consumption, form the basis for interpreting travel experiences and
constructing meaning (Gretzel et al., 2006). Indeed, the evolution in communication on
the Internet offers profound implications for both service providers and consumers. With
the emergence of new forms of communication such as personal blogs, social networking
tools, and collaborative tagging, numerous opportunities exist for hospitality and travel
businesses to tap into the consumer knowledge readily available online in abundance in
textual format. As illustrated by this study, mining consumer knowledge through
analysis of language can reveal consumers preferences when selecting restaurants. As such,
it is important for hospitality and travel businesses to recognize these opportunities for
improving their understanding of consumers. It seems to be a promising sign that many
service providers (e.g., www.sheraton.com) are including their guests opinions on their
websites in different forms (e.g., stories and reviews).
There are two important limitations of this study. First, consumers language
representation of restaurants was collected from college students. Future research should
extend the current study to collect the language used by the general public to improve the
external validity as well as to examine the differences across heterogeneous social

ARTICLE IN PRESS
Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

1017

demographic segments. The sampling of consumer language data can be expanded to


include different sources, e.g., real-time search terms used in online search functions,
discussions in online communities, and personal blogs, which can be used to validate the
nature of consumer language as identied in this study. Second, while this study shows that
marketers and consumers use considerably different languages, the effect of exploiting
consumers language in the development of a recommendation system was not tested.
Therefore, future research should aim to empirically validate these ndings through the
development of language-sensitive recommender systems.

References
Adomavicius, G., Tuzhilin, A., 2005. Toward the next generation of recommender systems: a survey of the stateof-the-art and possible extensions. IEEE Transactions on Knowledge and Data Engineering 17 (6), 734749.
Auty, S.G., 1992. Consumer choice and segmentation in the restaurant industry. Service Industries Journal 12 (3),
324339.
Baker, J., 1986. The role of the environment in marketing services: the consumer perspective. In: Czepiel, J.,
Congram, C., Shanahan, J. (Eds.), The Service Challenge: Integrating for Competitive. Advantage American
Marketing Association, Chicago, IL, pp. 7984.
Belkin, N.J., Croft, W.B., 1992. Information ltering and information retrieval: Two sides of the same coin?
Communications of the ACM 35 (12), 2938.
Bell, R., Meiselman, H. L., 1995. The role of eating environments in determining food choice. In: Marshall, D. W.
(Ed.), Food Choice and the Consumer Blackie. London, pp. 292310.
Bruner, G.C., 1990. Music, mood and marketing. Journal of Marketing 54 (4), 94104.
Burke, R., 2000. Knowledge-based recommender systems. In: Kent, A. (Ed.), Encyclopedia of Library and
Information Systems, vol. 69. Marcel Dekker, New York.
Burke, R., 2002. Hybrid recommender systems: surveys and experiments. User Modeling and User-Adapted
Interaction 12 (4), 331370.
Caplan, P., 1997. Food, Health and Identity. Routledge, New York.
Dann, G.M.S., 1997. The Language of Tourism. CAB International, Wallingford, UK.
Fesenmaier, D.R., Wober, K., Werthner, H., 2006. Introduction: recommendation systems in tourism.
In: Fesenmaier, D.R., Wober, K., Werthner, H. (Eds.), Destination Recommendation Systems: Behavioral
Foundatins and Applications. CAB International, London.
Finkelstein, J., 1989. Dining Out: A Sociology Of Modern Manners. Polity Press, Cambridge, MA.
Furner, J., 2002. On recommending. Journal of the American Society for Information Science and Technology 53
(9), 747763.
Goldberg, D., Nichols, D., Oki, B.M., Terry, 1992. Using collaborative ltering to weave an information tapestry.
Communications of the ACM 35 (12), 6170.
Goldman, K., 1993. Concept selection for independent restaurants. Cornell Hotel and Restaurant Administration
Quarterly 34 (6), 5972.
Gretzel, U., Fesenmaier, D.R., 2002. Building narrative logic into tourism information systems. IEEE Intelligent
Systems 17 (6), 5961.
Gretzel, U., Fesenmaier, D. R., OLeary, J. T., 2006. The transformation of consumer behaviour. In: Buhalis, D.,
Costa, C. (Eds.), Tourism Business Frontiers Butterworth-Heinemann, Oxford, pp. 918.
Gretzel, U., Mitsche, N., Hwang, Y.-H., Fesenmaier, D.R., 2004. Tell me who you are and I will tell you where to
go: use of travel personalities in destination recommendation systems. Information Technology & Tourism 7
(1), 312.
Hair, J., Anderson, R., Tatham, R., Black, W., 1998. Multivariate Data Analysis. Prentice-Hall, Upper Saddle
River, NJ.
Hanefors, M., Mossberg, L., 2003. Searching for the extraordinary meal experience. Journal of Business and
Management 9 (3), 249270.
Kivela, J., 1997. Restaurant marketing: selection and segmentation in Hong Kong. International Journal of
Contemporary Hospitality Management 9 (3), 116123.

ARTICLE IN PRESS
1018

Z. Xiang et al. / Hospitality Management 26 (2007) 10051018

Kivela, J., Reece, J., Inbakaran, R., 1999. Consumer research in the restaurant environment. Part 2: Research
design and analytical methods. International Journal of Contemporary Hospitality Management 11 (6),
269286.
Klein, L.R., 1998. Evaluating the potential of interactive media through a new lens: search versus experience
goods. Journal of Business Research 41 (3), 195203.
Loh, S., Lorenzi, F., Saldana, R., Licthnow, D., 2003. A tourism recommender system based on collaboration and
text analysis. Information Technology & Tourism 6 (3), 157165.
Nielsen, J., 2006. Use old words when writing for ndability. Available /http://www.useit.com/alertbox/
search-keywords.htmlS.
Pan, B., Fesenmaier, D.R., 2006. Online information search: vacation planning process. Annals of Tourism
Research 33 (3), 809832.
Pine, B.J., Gilmore, J.H., 1999. The Experience Economy. Harvard Business School Press, Boston.
Ricci, F., 2002. Travel recommender systems. IEEE Intelligent Systems 17 (6), 5557.
Ricci, F., Wober, K., Zins, A., 2005. Recommendations by collaborative browsing. In: Proceedings of the the
International Conference on Information and Communication Technologies in Tourism (ENTER2005),
Innsbruck, Austria. pp. 172182.
Riedl, J., Konstan, J., Vrooman, E., 2002. Word of Mouse: The Marketing Power Of Collaborative Filtering.
Warner Business Books, New York, NY.
Schafer, J.B., Konstan, J., Riedl, J., 2001. E-commerce recommendation applications. Data Mining and
Knowledge Discovery 5 (12), 115153.
Vogt, C., Fesenmaier, D.R., 1998. Expanding the functional information search model. Annals of Tourism
Research 25 (3), 551578.
Woelfel, J. K., Woelfel, J. D., 1997. CATPAC for Windows (Version 2.0).

You might also like