ArticlePDF Available

The effects of recommendations' presentation on persuasion and satisfaction in a movie recommender system

Authors:

Abstract and Figures

The explosive growth of Internet applications and content, during the last decade, has revealed an increasing need for information filtering and recommendation. Most research in the area of recommendation systems has focused on designing and implementing efficient algorithms that provide accurate recommendations. However, the selection of appropriate recommendation content and the presentation of information are equally important in creating successful recommender applications. This paper addresses issues related to the presentation of recommendations in the movies domain. The current work reviews previous research approaches and popular recommender systems, and focuses on user persuasion and satisfaction. In our experiments, we compare different presentation methods in terms of recommendations’ organization in a list (i.e. top N-items list and structured overview) and recommendation modality (i.e. simple text, combination of text and image, and combination of text and video). The most efficient presentation methods, regarding user persuasion and satisfaction, proved to be the “structured overview” and the “text and video” interfaces, while a strong positive correlation was also found between user satisfaction and persuasion in all experimental conditions.
Content may be subject to copyright.
REGULAR PAPER
The effects of recommendations’ presentation on persuasion
and satisfaction in a movie recommender system
Theodora Nanou George Lekakos
Konstantinos Fouskas
Published online: 19 May 2010
Springer-Verlag 2010
Abstract The explosive growth of Internet applications
and content, during the last decade, has revealed an
increasing need for information filtering and recommen-
dation. Most research in the area of recommendation sys-
tems has focused on designing and implementing efficient
algorithms that provide accurate recommendations. How-
ever, the selection of appropriate recommendation content
and the presentation of information are equally important
in creating successful recommender applications. This
paper addresses issues related to the presentation of rec-
ommendations in the movies domain. The current work
reviews previous research approaches and popular recom-
mender systems, and focuses on user persuasion and sat-
isfaction. In our experiments, we compare different
presentation methods in terms of recommendations’ orga-
nization in a list (i.e. top N-items list and structured over-
view) and recommendation modality (i.e. simple text,
combination of text and image, and combination of text
and video). The most efficient presentation methods,
regarding user persuasion and satisfaction, proved to be the
‘structured overview’’ and the ‘‘text and video’’ interfaces,
while a strong positive correlation was also found between
user satisfaction and persuasion in all experimental
conditions.
1 Introduction
During the last decade, recommender systems have seen an
impressive proliferation in various sectors (e.g. e-com-
merce, news, entertainment, and travel), especially in
Internet-based applications, mobile devices, and other
specialized appliances (e.g. personal video recorders).
Recent research [21,22] has shown the wide range of
recommender systems’ application domains, such as web-
sites, movies, TV shows, documents, news, music, and
restaurants. Among these, the most prominent are recom-
mendation systems developed for the information and
entertainment relevant domains (e.g. news, websites,
music, movies, books, and games). Recommender systems
have emerged, promising to help users cope with the
abundance of information items, by offering a personalized
service and proposing only those items the user is likely to
find interesting. The need for recommender systems is
especially evident in the movie domain, where conven-
tional information filtering technology cannot satisfy the
complex user selection process, which is often based on
emotional and esthetic grounds that are hard to define.
Recommender systems add value to existing applica-
tions, benefiting both the users (i.e. facilitate search and
retrieval, and increase quality of decisions) and the service
providers (i.e. attract and persuade customers, increase
sales, and create a trustful relationship with customers)
[7,22]. Therefore, key elements in the recommendation
process, with regard to the system goal, are persuasion (for
providers) and satisfaction (for users).
The extent to which these benefits are achieved depends
on the quality of the actual recommendations and their
presentation. The recommendation algorithm’s compe-
tence, in effectively combining user preferences and item
characteristics, is significant for generating potentially
T. Nanou
Hellenic Open University, Patras, Greece
e-mail: dnanou@gmail.com
G. Lekakos (&)
University of the Aegean, Samos, Greece
e-mail: glekakos@aegean.gr
K. Fouskas
University of Macedonia, Naoussa, Greece
e-mail: kfouskas@gmail.com
123
Multimedia Systems (2010) 16:219–230
DOI 10.1007/s00530-010-0190-0
interesting recommendations. The way that recommenda-
tion information is organized and presented to the user is
even more important for the recommendations’ perceived
quality and the user satisfaction [45]. The presentation
approach used often leads users to a subconscious evalu-
ation of the provided data quality, such as source’s credi-
bility [11] and relevance to their interests [36]. However,
previous research approaches have only partially studied
the ‘‘presentation level’’, especially issues related to the
recommendations’ modality and their organization in a list
of alternatives.
The study presented in this paper focuses on the pre-
sentation level of recommender systems. It researches the
impact of the recommendations’ presentation to the system
persuasiveness and the user satisfaction, while investigat-
ing any potential correlation between these two factors.
More specifically, the current work compares different
recommendation modalities (e.g. text, image, and video)
and organization methods (e.g. top-Nitem list and struc-
tured overview), with regard to persuasion and satisfaction,
in a movie recommender environment.
2 Literature review
2.1 The recommendation process
The recommendation process can be divided into three
phases: creation and maintenance of user profile, compu-
tation of recommendations, and visualization and presen-
tation of recommendations. During the last years,
researchers have started to focus on the presentation level
of recommender systems (e.g. recommendation lists and
explanations). Chen and Pu (2005) demonstrated how
variant presentation choices in recommendations and
explanations (e.g. graphs/text, short/long text explanations)
affect the process of creating a trustful relationship
between the system and the user. Herlocker et al. [13]
studied 21 different explanation interfaces in a movie
recommender system, in relation to the system’s persuasive
ability, and stressed the importance of good design of
explanation interfaces and the selection of appropriate
content for explanations. Additionally, they provided
experimental evidence that the use of explanations increase
user acceptance. Leino and Raiha [17] showed that the way
recommendations are presented affects the actual items
selected by the user. Finally, Van Barneveld and Van
Setten [39] studied variant rating scales in terms of sym-
bols, colors, scale symmetry, and range, as well as expla-
nation-related issues (e.g. level of detail), and they
identified several guidelines for the design of a TV rec-
ommender user interface.
In terms of organization, recommendations are usually
presented in lists, organized in a logical structure [37].
The most common ways of offering recommendations
include [31,37]: top item, top N-items list, similar to top
item list, predicted ratings for all items and structured
overview.
The above presentation methods of organizing and
presenting recommendations have been broadly used in
both academic and commercial recommender systems,
with ‘‘top N-items list’’ being a key option. However,
limited research [4] has focused on the impact of the
recommendations’ organization to a specific system goal
(e.g. inspire trust, persuade, or satisfy users). The appli-
cation domains in such studies mainly involve apartments
or electronic devices (e.g. digital cameras and computers).
The recommended items in these domains can be
explicitly described by objectively measured characteris-
tics of the items (e.g. price and capacity), in comparison
to other domains’ subjective characteristics (e.g. movie
genre and quality of production). These well-defined
characteristics can be used to form groups of items based
on their trade-offs. Previous research approaches com-
pared ‘‘top N-item lists’’ with explanation options (i.e. a
‘why’’ link leading to the recommendation’s explanation)
to a text ‘‘structured overview’’ of recommended items,
where recommendations that provide tradeoff alternatives
are grouped in one category. In the second case, group
titles that describe the items’ trade-offs are used as
explanations. In these studies Chen and Pu (2005) showed
that the recommendations’ organization affects the rec-
ommender system’s efficiency and the system’s ability to
build users’ trust. Their results demonstrate that a text
structured overview clearly increases users’ efficiency in
selecting a recommended item and facilitates the creation
of a trustful relationship between the users and the
system.
The literature review reveals a significant research gap
in the study of the recommender systems’ presentation
level, dealing with the organization of recommended items.
Furthermore, a significant research challenge exists in
studying the potential impact of the recommendations’
organization and presentation method to various system
goals (e.g. effectiveness, persuasiveness, and user satis-
faction), especially in domains where recommended items
are described by qualitative characteristics.
2.2 The information modality in recommendations
One important decision in designing the recommendations
list, after selecting the appropriate content for each item, is
to decide the modalities, which will be used to effectively
communicate information to users. The prevailing means
220 T. Nanou et al.
123
of offering recommendations is text. Many academic and
commercial recommendation systems [4,20](http://www.
Whattorent.com,http://www.Movielens.org) use only text
as a key information medium in recommendation lists.
However, some systems combine text with other modalities
in order to be more attractive or effective. Thus, text is
combined with image [18,35] or sound [16](http://
www.musicovery.com) in the recommendation lists, while
graphs are utilized mainly in recommendations’ explana-
tions [3,12]. Commercial movie recommenders, such as
Criticker (http://www.criticker.com) and Metacritic (http://
www.metacritic.com), provide movie trailers for some of
their movies, but not in their recommendation lists (only in
pages with detailed movie information).
The information modality has also been studied in other
similar domains, such as information retrieval environ-
ments, mainly in relation to a system’s effectiveness and
efficiency [15,40,41] but not in relation to system per-
suasion and user satisfaction (which is the aim of the
present study). The facts that video has not been previously
studied, as key information construct in movie recom-
mendation lists, and has insufficiently been exploited in
available commercial recommenders, renders it as a novel
and challenging research topic in recommender systems.
Video emerges as a new and promising information
modality, in the field of recommender systems, mainly due
to its increased popularity [1] and its intrinsic link to the
movies domain (i.e. a video trailer constitutes a sample of
the recommended item).
2.3 Studying recommendation quality in relation
to system goals
The accuracy of the employed recommendation algorithm
is closely related to the generation of recommendations that
match the user preferences. But still, computational accu-
racy is not enough to create useful or credible recom-
mendations [19]. The way information is selected,
organized, and presented is of critical importance to the
recommendations’ quality [4] and thus to the success or
failure of a recommendation system. However, a recom-
mendation’s quality and a recommender system’s success
can only be defined in terms of its goals. If a system aims at
supporting users’ decision-making process (i.e. select a
useful item), then the recommendations’ quality has to be
evaluated in terms of the system’s effectiveness. If a sys-
tem aims at increasing the provider’s sales, then the rec-
ommendations’ quality has to be evaluated in terms of the
system’s persuasive ability. Some recommenders may
target various goals at the same time (e.g. provide accurate
recommendations, create a trustful relation with customers,
and increase sales and satisfy users). Tintarev and Masthoff
[37] focused their research on the various dimensions of a
successful explanation facility, and their findings have been
broadly used to assess a recommendation’s quality in
general [4,6,17]. These research factors include trans-
parency (the visibility of a recommender’s system status or
operation), scrutability (a system’s property of being open
to inspection and able to recover from potential mistakes),
trust (confidence based on experience), effectiveness (sys-
tem’s potential to help the user make good decisions),
efficiency (system’s ability to help users avoid wasting
time and effort in finding a satisfactory item), persuasion,
and satisfaction.
Persuasion refers to a recommender system’s ability to
lead users to a change of attitude (e.g. purchase, use, and
evaluate). It is usually important for the service provider,
especially in e-commerce recommenders, where the
increase of sales is the main goal. Although, persuasion is
the direct or indirect goal of all recommender systems, it is
relatively rarely studied as a key construct (e.g. in com-
parison to effectiveness). Therefore, there is a great interest
in researching persuasion since it is affected by various
parameters (e.g. trust, transparency, and credibility) [4] and
is difficult to achieve as it requires a user’s emotional
change [30]. Herlocker [12] studied persuasion of 21 dif-
ferent explanation interfaces of a movie recommender
system. He concluded that the two best explanations were a
histogram of how similar users had rated the recommended
item and the percentage of the system’s previous successful
predictions for the specific user.
On the other hand, satisfaction is the contentment
obtained by the fulfillment of a need, desire, or expectation.
In recommender systems, satisfaction is usually studied
along with other parameters such as trust and acceptance of
recommendation technologies, but not in relation to per-
suasion [5,8,38,45].
What is apparent from the above review is the profound
deficiency in studying persuasion and satisfaction as key
constructs and their potential interdependence, in relation
to recommendations’ modality and organization.
3 Study hypotheses
The review of previous research approaches showed the
significant effect of recommendations’ presentation on
various system goals (e.g. persuasion, trust, transparency,
and effectiveness) [4,13,17]. The comparison of popular
movie recommender systems demonstrated the persistent
use of ‘‘top N-item’’ lists in organizing recommendations
and the total absence of the video modality in the recom-
mendation lists. However, it is still unclear how different
types of information modalities and their organization in a
list impact the system’s persuasive ability and the user
satisfaction.
The effects of recommendations’ presentation 221
123
Previous research from Chen and Pu [4] proved the
potential of a structured overview of recommendations to
increase users’ efficiency in selecting a recommended item
and create a trustful relationship between the users and the
system. Our study extends previous research by focusing
on persuasion and user satisfaction, in a different context
(i.e. movies domain), where recommended items are
described by qualitative characteristics.
H1 A movie recommender is more persuasive when
using a structured overview of recommendations, orga-
nized by the ‘‘movie genre’’ feature, in comparison to a top
N-items list.
H2 Users are more satisfied with a system that organizes
recommendations using a structured overview of recom-
mended movies, grouped by the ‘‘movie genre’’ feature, in
comparison to a system that uses a top N-items list.
Apart from the organization of recommendations in a
list, the information modality selected to communicate
information to users is another significant issue, when
designing a recommender’s presentation level. The present
study extends previous work [4,13,39] by moving the
research focus from the explanation level to the recom-
mendation level.
H3 A movie recommender is more persuasive when
using text and video modalities in the recommendation list,
in comparison to simple text information or a text and
image combination.
H4 Users are more satisfied with a system that uses text
and video modalities in the recommendation list, in com-
parison to a system that uses simple text information or a
text and image combination.
Persuasion is generally affected by various parameters,
such as perceived system credibility, recommendation
accuracy and quality of data, perceived system compe-
tence, and usability [4,9,44]. On the other hand, satis-
faction is also influenced by complex parameters such as
recommendation accuracy, system transparency, previous
trust relation with the system, and usability [14,29]. Since
both persuasion and satisfaction are depended on various
multi-dimensional parameters (some of which are common
in both), it is highly probable that persuasion and satis-
faction are somehow related. The investigation of the
nature of this potential relation between persuasion and
satisfaction is another research challenge in our work.
H5 Users’ satisfaction is positively correlated with a
system’s persuasive ability, regardless of the method of
presentation (i.e. the recommendations’ organization or the
information modalities used).
4 Methodological approach
4.1 Methodology overview
The current work investigates, both quantitatively and
qualitatively, the effects of the recommendations’ presen-
tation (in terms of information modality and organization
of recommended items) on the system persuasion and the
user satisfaction, in a movie recommender environment.
Therefore, different movie recommendation interfaces are
implemented, compared, and evaluated, following a social-
oriented evaluation method. A social-oriented evaluation
applies a user-centric approach and studies parameters that
affect user preference of a specific recommender system
and satisfaction [25]. In order to carry out this evaluation,
questionnaires were prepared and distributed to a pool of
potential movie recommender users.
Before the main evaluation, a preliminary survey took
place in order to select an appropriate users’ sample and
specifying their movie preferences. The criteria for
selecting a suitable user group mainly concerned a wide
range of demographic characteristics (e.g. men and women
of various ages, educational and occupational back-
grounds), medium to high familiarity with Internet appli-
cations, interest in watching movies and in seeking
information about them. The preliminary survey led to the
selection of 20 responders that matched most of the above
criteria. Furthermore, users rated 10–15 movies selected
from an indicative list of movies, created using the IMDB
list of popular movies by genre (http://www.imdb.com).
The movie ratings were utilized in the main research in
order to provide customized recommendations to the
selected user group. The preliminary survey led to the
selection of 20 responders that matched most of the above
criteria. Other supporting information was also collected,
concerning the users’ movie selection process. Users were
asked to prioritize three of the features they value the most
when selecting a movie to watch. The movie genre, the
current user mood, and the movie plot proved to be the
most important factors in the users’ selection process.
These features were taken into account during the design of
the movie recommender prototype.
After identifying the target user group, the main research
took place. All users participated in both experiments, which
were organized within a 10-day interval. In order to avoid
knowledge transfer within an experiment, the method of
random user interface presentation was selected. In each
experiment users were informed through e-mail. A link for
the tested interfaces was sent along with instructions, a
usage scenario, and an evaluation questionnaire. In all cases,
the users’ ultimate goal was to select a movie to watch.
222 T. Nanou et al.
123
Two basic presentation issues were studied in the main
research: the recommendation modality (e.g. text, image,
and video) and the organization of information (top N-
items vs. structured overview) in recommendation lists.
Each issue was researched in a separate experiment. In
both experiments user satisfaction and recommender’s
persuasive ability were measured and evaluated. In order to
provide different presentation interfaces, the movieSTAR
prototype was implemented. The first experiment evaluated
two recommendation interfaces based on their presentation
approach: the first interface was a list of top N-movies and
the second one provided the same recommendations
organized in groups of different movie genres (structured
overview). The second experiment compared three movie
recommendation interfaces: one providing detailed text
information, a second one providing the same text with
additional image information, and a third one presenting
some basic text information (e.g. movie title, year, rating,
and genre) along with a video trailer option.
4.2 The movieSTAR prototype
In order to perform the evaluation of the different presen-
tation modes in both experiments, movieSTAR system was
designed and implemented. MovieSTAR is a prototype
movie recommender that presents personalized movie rec-
ommendations in various views for evaluation purposes.
Therefore, movieSTAR does not use a proprietary recom-
mendation mechanism (since it is out of the scope of this
research), but presents already stored recommendations for
each user. The personalized recommendations were gener-
ated through the recommendation mechanism of Amazon.
For each user participating in the main research, a user
account was created in Amazon, and each user’s movie
preferences (stated as movie ratings in the preliminary
survey) were stored. Then 15 of the generated movie rec-
ommendations were stored for each user. A total of 71
different movies were recorded, and information about
these movies was collected through IMDB and YouTube
websites. Both sites were selected due to their high popu-
larity and their vast pool of available information (text,
photos, and videos).
The movieSTAR interfaces used during the evaluation
process were carefully selected as the most significant and
representative information combinations in a movie rec-
ommender system. More specifically, structured overview
was utilized as a promising method of organizing recom-
mendations [4], novel in the domain of movie recom-
mendations, while the top N-items listing was used as the
most popular way of setting up recommendations in a list
[37]. In terms of content selection, the design of the
interfaces was based on previous research of academic and
commercial movie recommenders, as well as on the
preliminary survey conducted in the framework of the
current work. Tintarev and Masthoff [34,36] researched
the factors that subconsciously affect users when descri-
bing or evaluating a movie. They concluded that movie
genre, plot, actors, visuals, and atmosphere as well as the
users’ current mood played a vital role in their evaluations.
These results were also confirmed by the preliminary sur-
vey conducted in the framework of the current work. The
participants in our survey stated that the factors, which
affect their selection of a movie in general, are mainly the
movie genre, the plot, the actors, and the users’ current
mood. Finally, the review of seven popular movie recom-
menders (http://www.imdb.com,http://www.metacritic.
com,http://www.movielens.org,http://www.whattorent.com,
http://www.liveplasma.com,http://www.amazon.com, and
http://www.criticker.com), which was conducted in the
framework of our research, provided a useful insight into
the design of commercial recommenders and verified the
previous design choices. The review demonstrated that the
information features mostly exploited in movie recom-
mender interfaces include movie title, year of production,
and movie rating (users’, critics’ or predicted by the sys-
tem), while some systems also provide some sort of visual
information (e.g. movie poster, DVD cover, and actors’
photos). The review also confirmed that the top N-items
listing was the most popular way of setting up recommen-
dations in a list. All these factors, highlighted by our pre-
liminary survey and review of movie recommenders, verified
relevant literature review and were appropriately taken into
account during the design of the movieSTAR interfaces.
MovieSTAR was implemented using Drupal, a flexible,
open source, content management system, with many
international distinctions [23,28]. MovieSTAR was
designed to provide basic system functionality (user login
and user account management facilities), focusing on the
presentation of personalized recommendation content. It
provides five different views of the same customized rec-
ommendations per user. The first experiment evaluated the
structure of recommendations in a list of suggested items by
comparing two different views: one providing a top N-item
list of recommended movies (View1) and another view
offering the same recommendations, organized in groups
(structured overview) of the same movie genre (View4).
The second experiment evaluated three different rec-
ommendation interfaces based on the information modality
they used. The first interface provided recommendations
using only text information (View1) in a top N-items list,
and was common in both experiments. It was used as a
baseline interface to facilitate the comparison of the other
two presentations. The second interface offered the same
recommendations using text and image information
(View2), while the third one (View3) provided limited text
information supported by a video trailer capability.
The effects of recommendations’ presentation 223
123
4.3 Experimental conditions and measures
In order to investigate the effects of the recommendations’
presentation on system persuasion and user satisfaction,
two main experiments were conducted. The first one
investigated the potential influence of the recommenda-
tions’ organization on persuasion and satisfaction. Previous
research [4] proved the dominance of ‘‘top N-items list’
and ‘‘structured overview’’ as main organization methods
in other domains and contexts. This led to two between-
subject conditions to manipulate persuasion and
satisfaction:
An unstructured condition, where recommendations are
presented in a top N-items list, with no specific order or
conventional arrangement (View1 in Fig.1)
A structured condition where recommendations are
organized in lists of the same movie genre (View4 in Fig.1)
The second experiment investigated the potential influ-
ence of the recommendations’ modalities on persuasion
and satisfaction. The movieSTAR design decisions,
described in the previous section, led to a three between-
subject conditions to manipulate persuasion and
satisfaction:
A text condition where recommendations are presented
in a top N-items list using only text information (View1
in Fig.1)
A text and image condition where recommendations are
presented in a top N-items list using text and image
information (View2 in Fig.2)
Fig. 1 The movieSTAR top N-items list using text (View1) and the structured overview of the same recommendations organized by the movie
genre (View4)
Fig. 2 Two presentations using different modalities: text and image on the left (View2), text and video on the right (View3)
224 T. Nanou et al.
123
A text and video condition where recommendations are
presented in a top N-items list using limited text
information and a video trailer capability (View3 in
Fig.2)
In both experiments, user satisfaction and recommender’s
persuasive ability were evaluated. Well-established mea-
sures of persuasion and satisfaction were used for the eva-
luation process, and they are summarized in Table 1.
The main measure used to evaluate the recommender’s
system persuasive ability was the likelihood of selecting a
recommended movie to watch [37] in each experimental
condition. Another measure that is related to the potential
of a system to persuade users is the increase in users’ self-
efficacy [10], i.e. their belief of being capable of per-
forming a certain task or attaining a goal. The increase in
user’s self efficacy is used to measure the system’s
potential to persuade users.
A similar measure is the potential of the actual presented
content to lead users to a decision (i.e. to persuade them).
This compound measure evaluates the quality of the uti-
lized information as it was studied in previous research [9,
26,44]. It refers to the content structure, accessibility,
usability, timeliness, and sufficiency (e.g. amount of data)
to help users reach a decision.
Satisfaction was measured twofold: satisfaction with the
system and satisfaction with the actual recommendations.
Although the recommendations are generated using an
external recommender mechanism (Amazon), we evaluate
the users’ satisfaction with the actual recommendations
because it may impact or relate to other measured variables
(e.g. satisfaction with the system).
Satisfaction with the recommendation process (system)
refers to the user–system interaction, the selection of infor-
mation content, and modalities as well as their presentation
method. It is evaluated by directly asking users how satisfied
they feel about the interaction with the system [14].
All previous variables and measures were evaluated
using 5-point Likert-type questions for each experimental
condition and experiment. Supportive information to
evaluate satisfaction was also gathered using comparative
questions, related to the system’s ease of use [29,37] and
their intention to reuse it [27].
5 Empirical evaluation
The first experiment was conducted during the first half of
May 2009, and the second one during the second half of the
same month. The same user group consisted of 20 users
participated in both experiments. A 10-day interval between
the two experiments was set in order to minimize any
potential knowledge transfer during the evaluation process.
In the following sections, we describe the results of both
experiments. In each experiment, the dependent variables
were statistically evaluated for all compared conditions
using the Wilcoxon signed ranks test.
Finally, in order to investigate the potential relation
between persuasion and satisfaction, we utilized the
Spearman’s rank correlation coefficient (or Spearman’s q).
All these statistical analyses were performed using SPSS
16.0 software for Windows.
5.1 Results of the first experiment: organizing
recommendations
We examined whether a ‘‘top N-items list’’ presentation
(View1) is perceived by users as more or less persuasive
and satisfactory than a ‘‘structured overview’’ of the same
recommendations (View4). Table 2presents the mean
scores (average ratings) as well as the Wilcoxon signed
ranks test statistics for persuasion and satisfaction mea-
sured variables.
Four out of the five measured variables present a sta-
tistically significant difference between the mean user
ratings for the two examined interfaces (View1 and
View4). This means that users clearly perceive one of the
two examined interfaces as more persuasive and satisfac-
tory (in terms of system interaction). The preferred pre-
sentation method in all cases is the ‘‘structured overview’
of recommendations (since the z-statistic is negative).
Therefore, H1 is confirmed for all measured variables,
while H2 is partially confirmed for the ‘‘satisfaction with
the system’’ measure.
5.2 Results of the second experiment: selecting
a recommendation modality
In the second experiment, we investigate the potential
influence of the recommendations’ modality on system’s
persuasive ability and the user satisfaction. We inquire
which modality combination is more persuasive and
Table 1 Dependent variables and measures
Dependent variables Measure
Persuasion The likelihood of selecting a
movie to watch
Increase in users’ self-efficacy
Information quality and
sufficiency to lead users to a
decision
Satisfaction Satisfaction with the
recommendation process
(system)
Satisfaction with the
recommendations
The effects of recommendations’ presentation 225
123
satisfactory in a list of recommended items. The examined
conditions are three, namely text only (in View1), text and
image (in View2), and finally, text and video (in View3).
Table 3presents the mean scores (average ratings) as well
as the results of the Wilcoxon signed ranks test for per-
suasion and satisfaction measured variables.
Four out of the five measured variables present a sta-
tistically significant difference between the mean user
ratings for the two examined interfaces (View1 and
View2). This means that users consider the ‘‘text and
image’’ condition (View2) as more persuasive and satis-
factory (in terms of system interaction) in comparison to
the ‘‘text only’’ condition (View1). The comparison of
‘text only’’ and ‘‘text and video’’ conditions present sim-
ilar results. All examined variables, except satisfaction
with recommendations, present a statistically significant
difference (in favor of ‘‘text and video’’ since z statistic is
negative) between the mean user ratings for the two
examined interfaces (View1 and View3).
The final comparison concerns two previously winning
experimental conditions (View2 and View3). Similarly, all
measured variables besides ‘‘satisfaction with recommen-
dations’’ variable present a statistically significant differ-
ence between the mean user ratings for the two examined
interfaces (View2 and View3). Summarizing the previous
findings, we conclude that a combination of ‘‘text and
video’’ is perceived as the best combination of information
modalities in a movie recommendations’ list, when trying
to persuade or satisfy the users. The worst choice is to use
simple text, while the combination of text and image is
considered as a moderate solution. Therefore, H3 is
confirmed for all measured variables, while H4 is partially
confirmed (only concerning the ‘‘satisfaction with the
system’’ measure). Unfortunately, our findings could not
lead us to a valid conclusion about the relation of the
recommendations’ modality and the satisfaction users’ gain
from the actual recommendations.
5.3 The correlation between persuasion
and satisfaction
In order to investigate a potential statistically significant
relation between persuasion and satisfaction, we tested the
main measures of each variable, namely the ‘‘Likelihood of
selecting a movie’’ and the ‘‘Satisfaction with the system’’,
using the Spearman’s rank correlation coefficient (or
Spearman’s q) in SPSS. Since users’ satisfaction is two-
fold, we included the measure of the ‘‘Satisfaction with
recommendations’’. This way, we can also investigate
whether the two forms of satisfaction affect one another.
The results of the Spearman’s correlation for all experi-
mental conditions and experiments are collectively pre-
sented in Table 4.
The Spearman’s qresults assume that regardless of the
recommendations’ organization in a list or the information
modalities used, a strong positive association exists
between a movie recommender’s persuasive ability and the
satisfaction users’ gain from the system interaction. This
correlation is assumed with absolute certainty at the 0.01
level (p=0), and therefore, H5 is confirmed. Additionally,
we observe no other significant correlation between the
three examined variables (i.e. the two forms of satisfaction
Table 2 Average ratings and Wilcoxon signed ranks test statistics for persuasion and satisfaction measures
Statistics Likelihood of
selecting a movie
Increase in users’
self-efficacy
Information quality
and sufficiency
Satisfaction
with the system
Satisfaction with
recommendations
View1 (top N-items list) 2.70 3.05 3.65 3.20 4.05
View4 (structured overview) 4.00 4.05 4.40 4.10 4.15
View1–View4 z-statistic (p)-2.940 (0.003) -2.748 (0.006) -2.236 (0.025) -2.858 (0.004) -1.000 (0.317)
Mean scores for persuasion and satisfaction measured variables (Experiment 1) (5-point Likert scale: 0 none, 5 a lot)
Table 3 Average ratings and Wilcoxon signed ranks test statistics for persuasion and satisfaction measures
Statistics Likelihood of
selecting a movie
Increase in users’
self-efficacy
Information quality
and sufficiency
Satisfaction
with the system
Satisfaction with
recommendations
View1 (text only) 2.55 3.10 3.55 2.85 4.10
View2 (text and image) 3.65 4.10 4.05 3.95 4.25
View3 (text and video) 4.70 4.65 4.55 4.60 4.25
View1–View2 z-statistic (p)-3.508 (0.000) -3.216 (0.001) -2.066 (0.039) -3.331 (0.001) -1.732 (0.083)
View1–View3 z-statistic (p)-3.614 (0.000) -3.199 (0.001) -2.623 (0.009) -3.774 (0.000) -1.732 (0.083)
View2–View3 z-statistic (p)-2.964 (0.003) -2.134 (0.033) -1.965 (0.049) -2.503 (0.012) 0.000 (1.000)
Mean scores for persuasion and satisfaction measured variables (Experiment 2) (5-point Likert scale: 0 none, 5 a lot)
226 T. Nanou et al.
123
or between the satisfaction with recommendations and the
likelihood of selecting a movie).
6 Discussion
6.1 The impact of the recommendations’ organization
on persuasion
Our results demonstrated that the recommendations’
organization clearly affects a system’s persuasive ability in
all three measured dimensions. The use of a structured
overview of movie recommendations was perceived by
users better in comparison to a simple top N-items list
presentation. This finding is mainly grounded on the
decision-making process users follow when selecting a
movie. Both previous research [34,38] and the preliminary
survey of the current work showed that the organization
criterion used in the structured overview of our recom-
mendations (i.e. the movie genre) is a significant factor in
describing or selecting a favorite movie. Additionally, a
user’s selection is based on ephemeral elements, such as
the user’s mood, which is closely related to the movie
genre. Therefore, it stands to reason that a well-organized
set of recommendations that facilitates users’ decision-
making process is more persuasive than a simple listing of
information.
The system’s potential to persuade users that was
measured through the increment of users’ self-efficacy
was also considered as higher in the case of structured
overview of recommendations. This increased self-effi-
cacy can be explained through the effective and easy to
use presentation of recommendations in the structured
overview interface. Human behavior theory [2] suggests
that two main ways of creating strong self-beliefs of
efficacy are through mastery experiences and reduction or
avoidance of people’s stress reactions. Thus, a system
that is effective, easy to use, and reduces users’ cognitive
effort is expected to increase users’ self-efficacy. Previ-
ous studies proved the effectiveness of the structured
overview presentation [4]. Additionally, participants in
our study perceive the structured overview presentation as
the easiest to use, with a high percentage of 75%, while
the incorporation of an organization criterion that mat-
ches the users’ decision-making process (i.e. movie
genre) reduces users’ cognitive effort in selecting a movie
to watch. These facts explain the increased system
potential to persuade users in the structured overview
presentation.
6.2 The impact of the recommendations’ organization
on satisfaction
A more structured organization of recommendations was
expected to increase users’ satisfaction. However, this
could only be partially confirmed in our study; users’ sat-
isfaction with the system was affected, but not satisfaction
with the actual recommendations.
The users’ satisfaction with the system is the main
variable that impacts overall users’ satisfaction, at least at a
presentation level. The results of our first experiment
exhibit a clear user preference in the structured overview
interface. A potential explanation of this finding is related
to the general user preference in brief and concise ways of
presenting information, especially in the recommendation
of low-risk items, such as movies [4]. The qualitative
comments of our study participants offered more insight
into this issue. Their comments focused on both the orga-
nization of recommendations in groups of the same
movie genre and the reduction of the directly available
information per recommended item, using the ‘‘more’’ link.
This reduction of detailed information in the list of
Table 4 Spearman’s rank
correlation results for
persuasion and satisfaction main
variables
** Correlation is significant at
the 0.01 level (two-tailed)
Spearman’s qLikelihood of
selecting a movie
Satisfaction with
the system
Satisfaction with
recommendations
Likelihood of selecting a movie
Correlation coefficient 1.000 0.753** 0.151
Sig. (two-tailed) 0.000 0.134
N100 100 100
Satisfaction with the system
Correlation coefficient 0.753** 1.000 0.137
Sig. (two-tailed) 0.000 0.175
N100 100 100
Satisfaction with recommendations
Correlation coefficient 0.151 0.137 1.000
Sig. (two-tailed) 0.134 0.175
N100 100 100
The effects of recommendations’ presentation 227
123
recommendations facilitated their movie search, at least in
the first step of their decision-making process (i.e. the
identification of a subset of interesting movies). A second
explanation is related to the perceived ease of use and other
usability dimensions that affect user satisfaction [27,29].
Since 75% of the users consider the structured overview
interface as the easiest to use, it is reasonable to gain more
satisfaction when interacting with it. Another relevant
usability parameter is the reduction of cognitive load in
selecting a movie [8,24]. As presented previously, the
utilization of a clustering criterion for recommendations
that is significant in the users’ decision-making process
definitely assisted users’ in selecting a movie by reducing
cognitive effort necessary to examine and evaluate all
possible options.
6.3 The impact of the recommendations’ modality
on persuasion
Our second experiment concluded that ‘‘text and video’’ is
perceived as the best combination of information modali-
ties in a movie recommendations’ list when trying to per-
suade users. This impact of the recommendations’ modality
on persuasion is confirmed in all three measured variables.
The increased likelihood of selecting a movie to watch
using the interface that combines text and video informa-
tion in recommendations (View3) is mainly attributed to
the particular nature of the video modality.
The users’ self-efficacy also seems to be severely
impacted by the use of different recommendation modali-
ties. The use of text and video (View3) presents the
strongest effect on the users’ self-efficacy increment, while
the text and image combination (View2) is ranked second.
The predominance of text and image (View2) over simple
text (View1) was expected as both interfaces present
exactly the same text information. The use of movie poster/
DVD cover provides additional information, which is sig-
nificant for users when evaluating a recommended movie
[32,38].
Concerning the potential of the recommendation content
and its presentation to lead users to a decision, the exam-
ined modalities followed the same ranking as previously
(first: View3, second: View2, and third: View1). It is
obvious that the simple text interface provides the least
available information, but it was still considered best in
terms of information sufficiency. This means that users
perceive the provided text information as adequate in order
to reach an initial decision (i.e. select a subset of favorite
movies) and the movie trailers as helpful in refining their
selection and making a final decision. Apart from the
recommendation content quantity, the quality played an
important role in the users’ persuasion process. The pre-
sentation of a video trailer definitely outweighs the other
two modalities since it provides objective data about the
recommended movie which users can evaluate themselves
(e.g. actors’ performance, and quality of production) and
increases its added value by facilitating the user selection
process [33].
6.4 The impact of the recommendations’ modality
on satisfaction
Our findings also suggest that the combination of ‘‘text and
video’’ (View3) is perceived as the best combination of
information modalities in terms of user satisfaction with
the system. Users perceive the ‘‘text and image’’ combi-
nation (View2) as a middle solution, while the simple text
interface (View1) was considered as inferior to the other
two.
A potential explanation is related to the perceived ease
of use and other usability dimensions that affect user sat-
isfaction [27,29]. The large majority of users in our study
(60%) consider the text and video interface as the easiest to
use; therefore, it is reasonable to gain more satisfaction
when interacting with it. Although users’ satisfaction with
the system was affected by the use of different information
modalities, we could not conclude the same about the
users’ satisfaction with the recommendations. As in the
case of the first experiment, these results were expected
since all examined interfaces (View1, View2, and View3)
presented to each user exactly the same recommendations.
These findings do not restrain our survey’s goal since we
were mainly interested in the satisfaction various interfaces
cause to users during their interaction. Possibly, a future
study with a larger user sample in the same experimental
conditions could lead to a more conclusive result con-
cerning the two forms of satisfaction examined and their
potential relation.
6.5 The interdependence of persuasion and satisfaction
When studying the results of both experiments conducted
and the generated statistics, we observe a significant
positive correlation between persuasion and satisfaction.
This means that when the users’ satisfaction with a movie
recommender is increased, so is the likelihood of selecting
a movie, using the specific system (and vice versa). This
strong positive association could be attributed to the mul-
tidimensional nature of both variables. The design and
esthetics of a recommender’s interface and the value-added
services it provides (e.g. delivery service) affect both users’
satisfaction [27] and system’s credibility and persuasive
ability [4]. If we carefully examine the results of our two
experiments, we will find out that both recommendations’
modalities and organization methods have a similar impact
on persuasion and satisfaction (i.e. the same interfaces are
228 T. Nanou et al.
123
perceived as persuasive and satisfactory in each experi-
ment). This fact justifies the positive direction of the cor-
relation between the two variables.
7 Conclusions and future work
This study builds upon previous work in the field of movie
recommendation interfaces and provides significant
empirical evidence on the influence of recommendations’
organization and modalities on system persuasion and
users’ satisfaction.
In our experiment, the use of a structured overview of
recommendations (grouped by the movie genre feature)
turned out to be the most persuasive and satisfactory pre-
sentation method in comparison to a top N-items listing of
the same recommendations. The study has also shown that
when describing proposed items in a recommendation list it
is important to select the appropriate amount and medium
of information to use. The use of text and video modalities
proved to be both persuasive and satisfactory in compari-
son to simple text descriptions or text and image combi-
nations. We must also clarify that satisfaction only refers to
the satisfaction users obtain when interacting with the
recommender system. Although, satisfaction with the
actual recommendations was also examined in relation to
recommendations’ organization and modalities, our results
were inconclusive. Further research is necessary to exam-
ine the impact of recommendations’ modalities or presen-
tation methods on the users’ satisfaction with
recommendations, if any exists. The findings of the current
study also suggest a strong positive association between a
system’s persuasive ability and the users’ satisfaction,
irrespective of the utilized recommendation modality or
organization method.
In order to fully understand the effects of recommen-
dations modalities and organization on persuasion and
satisfaction and to validate their correlation, it would be
useful to replicate this work in various product domains
(both for low- and high-risk products) and in larger eval-
uation groups in the future. Other future improvements
could also include more information modality combina-
tions and presentation schemes (e.g. top item, similar to top
item listings).
All study findings are particularly important both in the
research and the business arena. Research can benefit from
the current work by investigating other dimensions of the
same presentation challenges, such as
the impact of recommendations’ modalities and organi-
zation on the recommenders effectiveness and efficiency,
the potential of structured overview to counterbalance a
top N-items list with explanations, and
the study of structured overview in the movies domain
using multiple or user-defined clustering criteria (e.g.
movie genre and current user mood).
Recent advances in context-aware multimedia services
[42,43] present another potential research dimension for
our work. A future study could evaluate the tested pre-
sentation methods in smart phone environments.
The findings of the current survey can be exploited in the
design of effective movie recommender systems, especially
if we consider that few popular recommender systems fol-
low the proposed design choices (i.e. structured overview
and video modality in recommendation lists). Additionally,
the correlation between user satisfaction and system per-
suasion should also be taken into account when designing
movie recommenders (e.g. achieve user satisfaction through
usability, data organization, or appropriate selection of
content) since the strategic goal of any recommender pro-
vider (i.e. user persuasion and increase in sales) can be
achieved very easily with appropriate design decisions.
References
1. Alliance of Women Film Journalists (AWFJ): Opinion poll: all
about movie trailers. http://awfj.org/2008/05/07/awfj-opinion-
poll-all-about-movie-trailers/ (2008)
2. Bandura, A.: Self-efficacy. In: Ramachaudran, V. S., (ed.)
Encyclopedia of human behavior, vol. 4, pp. 71-81, Academic
Press (Reprinted in Friedman, H., [ed.] Encyclopedia of mental
health. San Diego: Academic Press, 1998) (1994)
3. Bilgic, M., Mooney, R.J.: Explaining recommendations: satis-
faction vs. promotion. In: Beyond Personalization: a Workshop
on the Next Stage of Recommender Systems Research at the
International Conference on Intelligent User Interfaces (2005)
4. Chen, L., Pu, P.: Trust building in recommender agents. In:
Proceedings of the Workshop on Web Personalization, Recom-
mender Systems and Intelligent User Interfaces at the 2nd
International Conference on E-Business and Telecommunication
Networks(ICETE’02) (2005)
5. Cosley, D., Lam, S.K., Albert, I., Konstan, J.A., Riedl, J.: Is
seeing believing? How recommender system interfaces affect
users’ opinions. In: Proceedings of the SIGCHI Conference on
Human Factors in Computing Systems (CHI ’03), pp. 585–592.
ACM Press, New York (2003)
6. Cramer, H., Evers, V., Ramlal, S., Someren, M., Rutledge, L.,
Stash, N., Aroyo, L., Wielinga, B.: The effects of transparency on
trust in and acceptance of a content-based art recommender. In:
User Modeling and User-Adapted Interaction, vol. 18, issue 5,
pp. 455–496. Kluwer, Dordrecht (2008)
7. Dias, M.B., Locher, D., Li, M., El-Deredy, W., Lisboa, P.J.: The
value of personalised recommender systems to e-business: a case
study. In: Proceedings of the 2008 ACM Conference on Rec-
ommender Systems (RecSys ’08), pp. 291–294. ACM Press,
New York (2008)
8. Felfernig, A., Gula B.: Consumer behavior in the interaction with
knowledge-based recommender applications. In: Proceedings of
the ECAI 2006 Workshop on Recommender Systems (2006)
9. Felfernig, A., Friedrich, G., Schmidt-Thieme, L.: Guest editors’
introduction: recommender systems. IEEE Intell. Syst. 22(3):18–
21 (2007)
The effects of recommendations’ presentation 229
123
10. Fogg, B.J.: Persuasive technologies. Commun. ACM 42(5), 26–
29 (1999)
11. Fogg, B.J., Soohoo, C., Danielson, D.R., Marable, L., Stanford,
J., Tauber, E.R.: How do users evaluate the credibility of web-
sites? A study with over 2,500 participants. In: Proceedings of the
2003 Conference on Designing for User Experiences (DUX ’03),
pp. 1–15. ACM Press, New York (2003)
12. Herlocker J.: Understanding and improving automated collabo-
rative filtering systems. Thesis, University of Minnesota (2000)
13. Herlocker, J.L., Konstan, J.A., Riedl, J.: Explaining collaborative
filtering recommendations. In: Proceedings of the 2000 ACM
Conference on Computer Supported Cooperative Work (CSCW
’00), pp. 241–250. ACM Press, New York (2000)
14. Herlocker, J.L., Konstan, J.A., Terveen, L.G., Riedl, J.T.: Eval-
uating collaborative filtering recommender systems. ACM Trans.
Inf. Syst. 22(1), 5–53 (2004)
15. Hughes, A., Wilkens, T., Wildemuth, B., Gary, M.: Text or
pictures? An eyetracking study of how people view digital video
surrogates. In: Bakker E.M. et al. (eds.) CIVR 2003, LNCS, vol.
2728, pp. 271–280. Springer, Berlin (2003)
16. Iskold A.: The art, science and business of recommendation engines.
Posted in ecommerce services of ReadWriteWeb technology blog.
http://www.readwriteweb.com/archives/recommendation_engines.
php (2007)
17. Leino, J., Ra
¨iha
¨, K.: Case amazon: ratings and reviews as part of
recommendations. In: Proceedings of the 2007 ACM Conference
on Recommender Systems (RecSys ’07), pp. 137–140. ACM
Press, New York (2007)
18. Linden, G., Smith, B., York, J.: Amazon.com recommendations:
item-to-item collaborative filtering. IEEE Internet Comput. 07(1),
76–80 (2003)
19. McNee, S.M., Riedl, J., Konstan, J.A.: Being accurate is not
enough: how accuracy metrics have hurt recommender systems.
In: CHI ’06 Extended Abstracts on Human Factors in Computing
Systems (CHI ’06), pp. 1097–1101. ACM Press, New York
(2006)
20. Miller, B., Albert, I., Lam, S.K., Konstan, J., Riedl, J.: Movie-
Lens unplugged: experiences with a recommender system on four
mobile devices. In: Proceedings of Human-Computer Interaction
Conference (HCI 2003) (2003)
21. Montaner, M., Lo
´pez, B., De La Rosa, J.L.: A taxonomy of
recommender agents on the Internet. Artif. Intell. Rev. 19(4),
285–330 (2003)
22. Nageswara Rao, K., Talwar, V.G.: Application domain and
functional classification of recommender systems—a survey.
DESIDOC J. Libr. Inf. Technol. 28(3), 17–35 (2008)
23. Needleman, R.: Complete list of 2009 Webware 100 winners,
CNET. http://news.cnet.com/8300-13546_109-29.html?tag=bc
(2009)
24. Nielsen, J.: Ten usability heuristics. http://www.useit.com/papers/
heuristic/heuristic_list.html (2005)
25. Perugini, S., Gonc¸alves, M.A., Fox, E.A.: Recommender systems
research: a connection-centric survey. J. Intell. Inf. Syst.
23(2):107–143 (2004)
26. Pipino, L., Lee, Y., Wang, R.: Data quality assessment. Commun.
ACM 45(4), 211–218 (2002)
27. Posselt, T., Gerstner, E.: Pre-sale vs. post-sale e-satisfaction:
impact on purchase intention and overall satisfaction. J. Interact.
Mark. 19(4), 35–47 (2005)
28. PRWeb: Drupal wins best overall 2008 open source CMS award
for second year in a row. http://www.prweb.com/releases/
2008/10/prweb1552274.htm (2008)
29. Pu, P.H., Kumar, P.: Evaluating example-based search tools. In:
Proceedings of the 5th ACM Conference on Electronic Com-
merce (EC ’04), pp. 208–217. ACM Press, New York (2004)
30. Rhoads, K.: Introduction to Influence. Retrieved from
http://www.workingpsychology.com/definit.html (2002)
31. Schafer, J.B., Konstan, J., Riedi, J.: Recommender systems in
e-commerce. In: Proceedings of the 1st ACM Conference on
Electronic Commerce, pp. 158–166. ACM Press, New York (1999)
32. Swearingen, K., Sinha, R.: Beyond algorithms: an HCI perspec-
tive on recommender systems. In: Proceedings of the SIGIR 2001
Workshop on Recommender Systems (2001)
33. Swearingen, J., Sinha, Q.: Interaction design for recommender
systems. In: Designing Interactive Systems. ACM Press,
New York (2002)
34. Tintarev, N.: Explaining recommendations. In: Conati C., McCoy
K., Paliouras G. (eds.) UM 2007. LNCS (LNAI), vol. 4511.
Springer, Heidelberg (2007)
35. Tintarev, N.: Explanations of recommendations. In: Proceedings
of the 2007 ACM Conference on Recommender Systems (RecSys
’07), pp. 203–206. ACM Press, New York (2007)
36. Tintarev, N., Masthoff, J.: Effective explanations of recommen-
dations: user-centered design. In: Proceedings of the 2007
ACM Conference on Recommender Systems (RecSys ’07),
pp. 153–156. ACM Press, New York (2007)
37. Tintarev, N., Masthoff, J.: Survey of explanations in recom-
mender systems. In: Proceedings of the Data Engineering
Workshop (ICDE), 2007 IEEE 23rd International Conference
(WPRSIUI), pp. 801–810. IEEE (2007)
38. Tintarev, N., Masthoff, J.: The effectiveness of personalized
movie explanations: an experiment using commercial meta-data.
In: Proceedings of the 5th International Conference on Adaptive
Hypermedia and Adaptive Web-Based Systems (AH ’08), vol.
5149, pp. 204–213. Springer, Berlin (2008)
39. Van Barneveld, J., Van Setten, J.: Designing usable interfaces for
TV recommender systems. In: Ardissono L., Kobsa A., Maybury
M. (eds.) Personalized Digital Television: Targeting Programs to
Individual Viewers. Human-Computer Interaction Series, vol. 6,
pp. 259–286. Kluwer, Dordrecht (2004)
40. Wildemuth, B., Marchionini, G., Wilkens, T., Yang, M., Geisler,
G., Fowler, B., Hughes, A., Mu, X.: Alternative surrogates for
video objects in a digital library: users’ perspectives on their
relative usability. In: Agosti M., Thanos C. (eds.) ECDL 2002,
LNCS, vol. 2458, pp. 493–507. Springer, Berlin (2002)
41. Wildemuth, B., Marchionini, G., Yang, M., Geisler, G., Wilkens,
T., Hughes, A., Gruss, R.: How fast is too fast? Evaluating fast
forward surrogates for digital video. In: Marshall C.C., Henry G.,
Delcambre L. (eds.) Proceedings of the 2003 Joint Conference on
Digital Libraries. IEEE Computer Society (2003)
42. Yu, Z., Zhou, X., Yu, Z., Zhang, D., Chin, C.: An OSGi-based
infrastructure for context-aware multimedia services. Commun.
Mag. 44(10), 136–142 (2006). IEEE
43. Yu, Z., Zhou, X., Zhang, D., Chin, C., Wang, X., Men, J.: Sup-
porting context-aware media recommendations for smart phones.
Pervasive Comput. IEEE. 5(3), 68–75 (2006)
44. Zanker, M., Bricman, M., Gordea, S., Jannach, D., Jessenitschnig,
M.: Persuasive online selling in quality and taste domains. In:
Bauknecht K., Pro
¨ll B., Werthner H. (eds.) EC-Web 2006, LNCS,
vol. 4082, pp. 51–60. Springer, Berlin (2006)
45. Ziegler, C.N., McNee, S.M., Konstan, J.A., Lausen, G.:
Improving recommendation lists through topic diversification. In:
Proceedings of the 14th International World Wide Web Confer-
ence (WWW2005), pp. 22–32. ACM Press, New York (2005)
230 T. Nanou et al.
123
... Festinger, 1954;Surendren and Bhuvaneswari, 2014;Schwind et al., 2011;Kuan et al., 2007;Schwind and Buder, 2014;Nguyen et al., 2007Persuasion 4.1.2 Fogg, 2002Perloff, 2020;Meske and Potthoff, 2017;Yoo et al., 2012;Gretzel and Fesenmaier, 2006;Jugovac et al., 2018;Yoo and Gretzel, 2011;Nanou et al., 2010;Cremonesi et al., 2012;Felfernig et al., 2008a;Herlocker et al., 2000;Tintarev and Masthoff, 2012;Berdichevsky and Neuenschwander, 1999;Smids, 2012Interactions & Interfaces 4.1.3 Knijnenburg et al., 2011Knijnenburg and Willemsen, 2015;Bollen et al., 2010;Chen and Pu, 2010b;Chen and Pu, 2010a;Hu and Pu, 2011;Ekstrand et al., 2014;Jannach, 2017 Attitudes &Beliefs 4.1.4 ...
... Related work finds the credibility of recommender systems (Yoo and Gretzel, 2011) is a decisive factor in a recommender system's persuasiveness. Nanou et al. (2010) observe that the presentation of recommendation lists in the context of movie recommendations influences persuasiveness. They compare top-N recommendation lists with a structured overview of recommendations, in which recommendations are organized by movie genre and are presented either as purely textual recommendation lists or as a multimodal representation of recommendations (text, images, video). ...
Book
Full-text available
Personalized recommender systems have become indispensable in today’s online world. Most of today’s recommendation algorithms are data-driven and based on behavioral data. While such systems can produce useful recommendations, they are often uninterpretable, black-box models that do not incorporate the underlying cognitive reasons for user behavior in the algorithms’ design. This survey presents a thorough review of the state of the art of recommender systems that leverage psychological constructs and theories to model and predict user behavior and improve the recommendation process – so-called psychology-informed recommender systems. The survey identifies three categories of psychology-informed recommender systems: cognition-inspired, personality-aware, and affectaware recommender systems. For each category, the authors highlight domains in which psychological theory plays a key role. Further, they discuss selected decision-psychological phenomena that impact the interaction between a user and a recommender. They also focus on related work that investigates the evaluation of recommender systems from the user perspective and highlight user-centric evaluation frameworks, and potential research tasks for future work at the end of this survey.
... Festinger, 1954;Surendren and Bhuvaneswari, 2014;Schwind et al., 2011;Kuan et al., 2007;Schwind and Buder, 2014;Nguyen et al., 2007Persuasion 4.1.2 Fogg, 2002Perloff, 2020;Meske and Potthoff, 2017;Yoo et al., 2012;Gretzel and Fesenmaier, 2006;Jugovac et al., 2018;Yoo and Gretzel, 2011;Nanou et al., 2010;Cremonesi et al., 2012;Felfernig et al., 2008a;Herlocker et al., 2000;Tintarev and Masthoff, 2012;Berdichevsky and Neuenschwander, 1999;Smids, 2012Interactions & Interfaces 4.1.3 Knijnenburg et al., 2011Knijnenburg and Willemsen, 2015;Bollen et al., 2010;Chen and Pu, 2010b;Chen and Pu, 2010a;Hu and Pu, 2011;Ekstrand et al., 2014;Jannach, 2017 Attitudes &Beliefs 4.1.4 ...
... Related work finds the credibility of recommender systems (Yoo and Gretzel, 2011) is a decisive factor in a recommender system's persuasiveness. Nanou et al. (2010) observe that the presentation of recommendation lists in the context of movie recommendations influences persuasiveness. They compare top-N recommendation lists with a structured overview of recommendations, in which recommendations are organized by movie genre and are presented either as purely textual recommendation lists or as a multimodal representation of recommendations (text, images, video). ...
... This was done for the following 3 main reasons. First, movie genres are crucial features that best represent the characteristics of movies and are considered a key criterion impacting users' satisfaction and experiences when using a movie recommendation system [25]. The second reason is that the properties of other predicates within DBpedia such as ''dbo:director'' and ''dbo:writer'' are often literal nodes (represented as string values) that are not useful. ...
Article
Full-text available
Leveraging knowledge graphs for post-hoc recommendation explanations has been investigated in recent years. Existing approaches rely mainly on the overlap properties (encoded by knowledge graphs) that characterize both user liked items and the recommended ones. These approaches, however, do not fully leverage the property hierarchy of knowledge graphs which may lead to flawed explanations. In this paper we introduce an approach that takes the whole property hierarchy into account. This is done with a limited computation time overhead thanks to efficient algorithmic optimizations relying on sub-ontology extraction. The hierarchical relationships among properties are also considered to avoid redundant properties for explanation. We carried out a user study of 155 participants in the movie recommendation domain and used both offline and online metrics to assess the proposed approach. Significant improvements, in terms of informativeness (by 39%), persuasiveness (by 22%), engagement (by 29%) and user trust (by 26%), are suggested by the obtained results, as compared to the state-of-the-art property-based explanation model. Our findings indicate the superiority of accounting for the whole property hierarchy when dealing with post-hoc recommendation explanations.
... Other factors might have an impact on users' experience with recommender systems, and can therefore influence users' perception of a recommended item (Jugovac and Jannach, 2017). The way the recommendations are presented (Nanou et al., 2010) as well as the system's embodiment (Herse et al., 2018), latency , and sentence length might all affect users' perception. ...
Article
Full-text available
Unhealthy eating behavior is a major public health issue with serious repercussions on an individual’s health. One potential solution to overcome this problem, and help people change their eating behavior, is to develop conversational systems able to recommend healthy recipes. One challenge for such systems is to deliver personalized recommendations matching users’ needs and preferences. Beyond the intrinsic quality of the recommendation itself, various factors might also influence users’ perception of a recommendation. In this paper, we present Cora, a conversational system that recommends recipes aligned with its users’ eating habits and current preferences. Users can interact with Cora in two different ways. They can select pre-defined answers by clicking on buttons to talk to Cora or write text in natural language. Additionally, Cora can engage users through a social dialogue, or go straight to the point. Cora is also able to propose different alternatives and to justify its recipes recommendation by explaining the trade-off between them. We conduct two experiments. In the first one, we evaluate the impact of Cora’s conversational skills and users’ interaction mode on users’ perception and intention to cook the recommended recipes. Our results show that a conversational recommendation system that engages its users through a rapport-building dialogue improves users’ perception of the interaction as well as their perception of the system. In the second evaluation, we evaluate the influence of Cora’s explanations and recommendation comparisons on users’ perception. Our results show that explanations positively influence users’ perception of a recommender system. However, comparing healthy recipes with a decoy is a double-edged sword. Although such comparison is perceived as significantly more useful compared to one single healthy recommendation, explaining the difference between the decoy and the healthy recipe would actually make people less likely to use the system.
... The use of recommender systems, on the other hand, makes them consume content they would not have consumed otherwise. Another study examined the persuasive nature of recommender systems and found that users tend to rely on the system's recommendation when preceding recommendations were satisfactory [23]. Although this persuasive power may also lead to potential negative psychological effects such as excessive use and addictive behavior (e.g. ...
Chapter
Streaming Video-on-demand (SVOD) services are getting increasingly popular. Current research, however, lacks knowledge about consumers’ content decision processes and their respective influencing factors. Thus, the work reported on in this paper explores socio-technical interrelations of factors impacting content choices in SVOD, examining the social factors WOM, eWOM and peer mediation, as well as the technological influence of recommender systems. A research model based on the Theory of Reasoned Action and the Technology Acceptance Model was created and tested by an n = 186 study sample. Results show that the quality of a recommender system and not the social mapping functionality is the strongest influencing factor on its perceived usefulness. The influence of the recommender system and the influence of the social factors on the behavioral intention to watch certain content is nearly the same. The strongest social influencing factor was found to be peer mediation.
... Yet, self-reinforcing loops, constraining the user to certain interaction mechanisms, must be avoided [46]. For this reason, among others, it is finally important to explore the possibilities for the 2d) presentation: Earlier works on RS have shown, e.g., significant effects of presenting items or the entire interface in different ways [4,19,40,44]. Whereas only behavioral data were considered in these cases, studying factors such as personality has a long tradition in user interface design [3]. This might turn out useful for an adaptive presentation of DA, especially for raising awareness of the mechanisms the system has predicted to be of relevance before, in a persuasive but unobtrusive manner: Explanations, currently used in RS mainly to explain item recommendations [55], but also perceived differently depending on user characteristics [22], could be used, e.g., to highlight the benefits of continuing the interaction with a specific DA. ...
Conference Paper
Full-text available
On the one hand, users' decision making in today's web is supported in numerous ways, with mechanisms ranging from manual search over automated recommendation to intelligent advisors. The focus on algorithmic accuracy, however, is questioned more and more. On the other hand, although the boundaries between the mechanisms are blurred increasingly, research on user-related aspects is still conducted separately in each area. In this position paper, we present a research agenda for providing a more holistic solution, in which users are supported with the right decision aid at the right time depending on personal characteristics and situational needs. [ Full text available at: https://bit.ly/3Cful4l ]
Chapter
Due to the fact that individuals each have a different perception on privacy concerns, it becomes complicated to find the ultimate privacy policy that fulfils all the users’ needs. Previous research frameworks for privacy recommendations mostly focused on a unidimensional privacy profile representation and, in rare cases, contextualised frameworks are used where other factors including the trustworthiness of the information recipients, information sensitivity or other privacy attitudes are considered in the model. Presented in the following sections privacy recommender systems are described for the following reasons: the privacy wizard is the one of the earliest works on developing an automated privacy management support tool; the personalised privacy assistant system is the latest practical solution that combines results of several years of research; YourPrivacyProtector is an example of an interesting approach but with weak technical rationale.
Thesis
Dans le contexte actuel de surcharge causée par l'important volume de données numériques accessibles, les systèmes de recommandation permettent de guider l'utilisateur dans ses activités d'apprentissages, d'achats, de loisir, d'écoute musicale, de lectures..., en lui suggérant des items personnalisés. Pour cela, ils prédisent ses préférences relativement aux items qu'il n'a pas encore évalués. Des approches classiques de recommandation, comme le filtrage collaboratif par exemple, reposent sur les données collectées par le biais de retours d'utilisateurs, généralement sous la forme d'une matrice de notes, et tentent d'y découvrir les informations pertinentes pour caractériser et prédire les goûts des utilisateurs. En complément de ces données, les connaissances liées aux items eux-mêmes représentent également un atout majeur pour l'amélioration des performances des systèmes de recommandation. L'ingénierie des connaissances, plus spécifiquement le Web sémantique et les graphes de connaissances, peuvent y jouer un rôle central. Tel est le cadre de notre travail de recherche qui propose différentes voies d'amélioration des systèmes de recommandation, adoptant une vision transversale « des données aux connaissances », et ce sur trois aspects différents : la pertinence (accuracy), la diversification et l'explicabilité des recommandations.Notre première contribution est principalement axée sur les données. Elle concerne l'accuracy des recommandations en termes de prédiction des goûts des utilisateurs. Nous proposons EBCR (Empirical Bayes Concordance Ratio), une méthode simple et générique inspirée de l'inférence bayésienne, qui permet d'ajuster les calculs de similarité entre utilisateurs (ou entre items) mis en œuvre dans le filtrage collaboratif. Cet ajustement est réalisé en fonction du nombre d'items co-notés (ou du nombre d'utilisateurs ayant noté un même item). Les expériences menées sur des jeux de données de référence ont confirmé que cette méthode améliore systématiquement l'accuracy du filtrage collaboratif pour toutes les mesures de similarité considérées.Notre deuxième contribution concerne la diversification des recommandations. Nous avons mené une étude approfondie visant à comparer et analyser la performance de sept modèles de recommandation incluant des modèles classiques comme le filtrage collaboratif et le facteur latent ainsi que ceux, plus récents, se basant sur les réseaux de neurones profonds ou les plongements (embeddings) de graphe de connaissances. Nous avons évalué leur capacité à fournir des items diversifiés et proposé une approche qui permet d'ajuster la diversité aux besoins spécifiques des utilisateurs. Afin d'estimer la diversité des recommandations, nous avons considéré des mesures de similarité sémantique en tirant parti des connaissances liées aux items à l'aide du Web sémantique.Enfin, notre troisième contribution concerne l'explicabilité des recommandations. Ici, nous exploitons plus en profondeur les connaissances du domaine, en proposant une approche de l'explication post-hoc des recommandations qui considère efficacement la hiérarchie des concepts au sein du graphe de connaissances de DBpedia. Les résultats de l'évaluation de notre approche basée sur une étude comprenant 155 participants suggèrent des améliorations significatives en termes d'engagement, de confiance et de persuasion.
Conference Paper
p>Food recommender systems optimize towards a user's current preferences. However, appetites may vary, in the sense that users might seek healthy recipes today and look for unhealthy meals tomorrow. In this paper, we propose a novel approach in the food domain to diversify recommendations across different lists to 'serve' different users goals, compiled in a multi-list food recommender interface. We evaluated our interface in a 2 (single list vs multiple lists) x 2 (without or with explanations) between-subject user study (N = 366), linking choice behavior and evaluation aspects through the user experience framework. Our multi-list interface was evaluated more favorably than a single-list interface, in terms of diversity and choice satisfaction. Moreover, it triggered changes in food choices, even though these choices were less healthy than those made in the single-list interface. </p
Article
Full-text available
Knowledge-based recommender technologies pro- vide a couple of mechanisms for improving the accessibility of product assortments for customers, e.g., in situations where no solution can be found for a given set of customer require- ments, the recommender application calculates a set of repair actions which can guarantee the identification of a solution. Further examples for such mechanisms are explanations or product comparisons. All these mechanisms have a certain effect on the behavior of customers interacting with a rec- ommender application. In this paper we present results from a user study, which focused on the analysis of effects of dif- ferent recommendation mechanisms on the overall customer acceptance of recommender technologies.
Article
The amount of scientific and technical information is growing exponentially. As a result, the scientific community has been overwhelmed by the information published in number of new books, journal articles, and conference proceedings. In addition to increasing number of publications, advances in information technology have dramatically reduced the barriers in electronic publishing and distribution of information over networks virtually anywhere in the world. As a result, the scientific community is facing the problem of locating relevant or interesting information. To address the problem of information overload and to sift all available information sources for useful information, recommender systems or filtering systems have emerged. Generally, recommender systems are used online to suggest items that users find interesting, thereby, benefiting both the user and merchant. Recommender systems benefit the user by making him suggestions on items that he is likely to purchase and the business by increase of sales. Filtering information or generation of recommendations by the recommender systems mimic the process of information retrieval systems by incorporating advanced profile building techniques, item/user representation techniques, filtering and recommendation techniques, and profile adaptation techniques. This paper addresses the application domain analysis, functional classification, advantages and disadvantages of various filtering and recommender systems.
Article
In this study 2,684 people evaluated the credibility of two live Web sites on a similar topic (such as health sites). We gathered the comments people wrote about each siteís credibility and analyzed the comments to find out what features of a Web site get noticed when people evaluate credibility. We found that the ìdesign lookî of the site was mentioned most frequently, being present in 46.1% of the comments. Next most common were comments about information structure and information focus. In this paper we share sample participant comments in the top 18 areas that people noticed when evaluating Web site credibility. We discuss reasons for the prominence of design look, point out how future studies can build on what we have learned in this new line of research, and outline six design implications for human-computer interaction professionals.
Article
An abstract is not available.