Content uploaded by Matthias Hauser
Author content
All content in this area was uploaded by Matthias Hauser on Jun 29, 2018
Content may be subject to copyright.
REDEFINING THE OFFLINE RETAIL EXPERIENCE:
DESIGNING PRODUCT RECOMMENDATION
SYSTEMS FOR FASHION STORES
Research in Progress
Hanke, Jannis, University of Würzburg, Germany, jannis.hanke@uni-wuerzburg.de
Hauser, Matthias, University of Würzburg, Germany, matthias.hauser@uni-wuerzburg.de
Dürr, Alexander, University of Würzburg, Germany, alexander.duerr@uni-wuerzburg.de
Thiesse, Frédéric, University of Würzburg, Germany, frederic.thiesse@uni-wuerzburg.de
Abstract
Retailers worldwide have started deploying smart service innovations in their stores to regain market
share lost to online competitors. Against this backdrop, this paper focuses on the design of product
recommendation systems for fashion stores. Our research particularly aims at answering the issues of
whether and to what extent (i) the sensing capabilities of smart fashion retail environments and (ii) the
integration of contextual information can improve the quality of such recommendations. To this end,
we consider smart fitting rooms with the ability to detect products and customers as a showcase; a
transaction dataset from a leading German fashion retailer; and contextual information about the time
of purchase, the store type, and the weather conditions. Our preliminary analyses indicate that sensor
information regarding garment and user identification, as well as further context data help to improve
product recommendations in fashion stores.
Keywords: Smart Service Systems, Recommendation Systems, Context Awareness, Internet of Things,
Retail Industry, Predictive Analytics, Cyberphysical Systems, Smart Fitting Rooms.
1 Introduction
Several scholars have called for more research in service-related areas over the last years (e.g., Böhmann
et al., 2014; Frost and Lyons, 2017; Ostrom et al., 2010). One important aspect of service research are
smart service systems (Frost and Lyons, 2017), which the National Science Foundation (2014) defines
as “system[s] capable of learning, dynamic adaptation, and decision making based upon data received,
transmitted, and/or processed to improve [their] response to a future situation.” One very promising
“playground for service systems innovation” are cyberphysical systems, which are based on technology
(e.g., sensing or communication capabilities) and allow for the integration of new sources of contextual
information (e.g., location or social contexts). The design of such systems is, however, challenging because
they have to bridge the boundaries between tangible and intangible resources (Böhmann et al., 2014) and
need to be woven around legacy systems (Hauser et al., 2017b; Weiser, 1999). In addition, such systems
should make use of the given contextual information to provide users with services that truly leverage the
business value that arises from the integration of physical and virtual worlds.
The present paper is concerned with service systems in smart retail environments. Recent developments in
the retail industry (e.g., the introduction of the “Amazon Go” store) point to a fundamental transformation
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 1
Hanke et al. /Fashion Store Product Recommendation Systems
of traditional brick and mortar (B&M) stores into smart stores. Such stores leverage sensor technology
and customer data to provide novel customer services (Gregory, 2015; Manyika et al., 2015). We consider
product recommendation systems that rely on recommendation algorithms and various data sources
(e.g., customer purchase histories or contextual information) to recommend products that suit individual
customer preferences (Liang et al., 2006). However, in contrast to most research on recommendation
systems, we do not focus on automated recommendations in e-commerce, but on systems for physical
store environments. Our showcase are smart fitting rooms that offer garment recommendations on screens
within individual cabins. Such systems help customers with ever-growing product ranges (Häubl and
Trifts, 2000) and can lead to additional sales and increased customer loyalty (Schafer et al., 1999).
We are particularly interested in the issues of whether and to what extent (i) the sensing capabilities of smart
fitting rooms (i.e., product or customer detection) and (ii) the integration of contextual information can
improve the quality of such recommendations. To this end, we first review literature on recommendation
systems for fashion stores focusing on (i) research that describes suitable recommendation algorithms
(Section 2.1) and (ii) studies that leverage contextual information in fashion stores (Section 2.2). The
review of suitable recommendation algorithms is crucial for our research, as the applicability of different
algorithms depends on the sensing capabilities of smart fitting rooms. In a second step, we propose
recommendation algorithms tailored to these sensing capabilities and describe a means by which to
integrate contextual information (Section 3). In this context, we also investigate to what extent the choice
of algorithm and the integration of selected context attributes affect recommendation quality.
2 Related Work
We conducted a comprehensive literature review on B&M recommendation systems following the sugges-
tions put forward by Webster and Watson (2002). Google Scholar and Scopus were used as databases and
searched by means of a search query consisting of two building blocks.
1
On the one hand, “brick and
mortar”, “offline”, “store”, “retail”, “stationary”, or “fitting room” had to appear in the titles of the papers.
On the other hand, “recommendation system” (or commonly employed synonyms) had to appear in the
papers’ full texts. We retrieved 397 papers on Google Scholar and 222 papers on Scopus, examined titles
and abstracts, and discarded papers not relevant for our research. Most of the non-relevant papers either
focus on the comparison of offline and online evaluations of e-commerce recommendation systems or
are concerned with environments outside of the retail world (e.g., a tourist attraction recommendation
system). We identified 37 relevant articles, which can roughly be categorized into
•
discussions of economic potentials (Kamei et al., 2011; Al-Kassab et al., 2009; B. Keller et al., 2015;
T. Keller and Raffelsieper, 2014; Kroon et al., 2007; Liaghat et al., 2013; Melià-Seguí et al., 2013;
Pfeiffer et al., 2015; Pous et al., 2013; Thiesse et al., 2009),
•
user acceptance studies (Daraghmi and Kadoori, 2016; Kowatsch and Maass, 2010a,b; Kowatsch et al.,
2009; Y. E. Lee and Benbasat, 2010; Resatsch et al., 2008), and
•
technology-focused papers (Chan and Capra, 2012; Cinicioglu and Shenoy, 2016; Giering, 2008;
Hansen and Loos, 2008; Hu et al., 2016; Kamei et al., 2010; Kronberger and Affenzeller, 2011; S.
-
L.
Lee, 2010; J. Li et al., 2014; Y.
-
M. Li et al., 2017; Liao et al., 2014; Luo et al., 2016; Nakahara and
Yada, 2011; Poulopoulos and Kyriazis, 2017; Sano et al., 2015; Sato et al., 2015; Skiada et al., 2016;
So and Yada, 2017; Walter et al., 2012; P. Wang et al., 2014; X. Zhang et al., 2016).
As per our research agenda, we are mainly interested in the technology-focused papers. To gain a more
comprehensive set of such papers, we additionally conducted a forward and backward search based on
the papers from all three research streams, which allowed us to identify an additional 19 relevant papers
1
We used the following Google Scholar search string: (intitle:“brick & mortar” OR intitle:“B&M” OR intitle:“brick and mortar”
OR intitle:“retail” OR intitle:“store” OR intitle:“offline” OR intitle:“stationary” OR intitle:“fitting room”) AND (“recommender”
OR “recommendation system” OR “recommendation agent”). For Scopus, “title” instead of “intitle” was used.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 2
Hanke et al. / Fashion Store Product Recommendation Systems
(Bodapati, 2008; Choi et al., 2015; Duc-Trong Le, 2017; Fang et al., 2012; Hao et al., 2016; Hauser et al.,
2017a; Hou and Chen, 2011; Hsu et al., 2004; Jie et al., 2012; Kawashima et al., 2006; Landmark and
Sjøbakk, 2017; Lawrence et al., 2001; E. H.
-
C. Lu et al., 2012; Ngai et al., 2008; Reischach et al., 2009;
Y.-F. Wang et al., 2004; Wong et al., 2012; W. Zhang et al., 2008; Z. Zheng et al., 2017).
The 40 technology-focused papers investigate (i) sensing capabilities of smart fitting rooms, (ii) algorithms
for B&M recommendation systems, and (iii) contextual attributes in B&M stores. We argue that for the
effective deployment of smart fitting rooms, these issues must be addressed in this exact order. In order
to leverage auto-ID technologies, the fitting rooms first must gather information about the environment.
While some authors mention the value of identifying individual customers (Hansen and Loos, 2008),
others put special emphasis on the identification of products (e.g., Hauser et al., 2017a). Secondly,
recommendation algorithms that can process this information are needed as the corresponding assumption
is that this helps generate better recommendations (Landmark and Sjøbakk, 2017). Finally, contextual
information should also be used because users’ preferences may change due to situational circumstances.
2.1 Algorithms for B&M Recommendation Systems
Recommendation algorithms can be categorized into (i) content-based methods where similarities between
item features are taken into account and (ii) collaborative-filtering approaches where product suggestions
are based on the previous behavior of users with similar preferences (Adomavicius and Tuzhilin, 2005).
Many papers in our article set rely on collaborative approaches (e.g., S.
-
L. Lee, 2010) and only a few apply
content-based approaches (e.g., Wong et al., 2012). This might be because content-based filtering requires,
by definition, additional information about the products (Pazzani and Billsus, 2007). This information is,
however, often unavailable or does not contain valuable data with which to distinguish between items users
like and dislike (Poulopoulos and Kyriazis, 2017). In the fast-changing fashion industry, the disadvantage
of being dependent on static knowledge (e.g., fashion experts) is particularly severe (Landmark and
Sjøbakk, 2017). In the following, we therefore focus on collaborative-filtering approaches.
In the course of the review, we identified (i) the user cold start problem and (ii) the absence of explicit
product ratings as the main challenges to the design of B&M recommendation algorithms.
User cold start problem.
This challenge describes the phenomenon that a system struggles to give good
recommendations to users for whom only some or no information is available (Bobadilla et al., 2012).
The latter case occurs if customers do not yet have a purchase history or cannot be identified by the
service system. In e-commerce, users can be identified as long as they are logged in with their account
while browsing. In addition, even if they are not logged in, information about their preferences can often
be gained from additional data sources (e.g., click-stream data) (Bobadilla et al., 2012). In contrast,
today’s B&M stores usually do not know the identity of the customers who are currently in their stores.
In most cases, customer identification takes place at the checkout (e.g., through a customer card or e-
payment). An earlier identification (e.g., in the fitting room) would require dedicated technical equipment.
However, if such technical equipment is not available, the service systems must still be able to provide
recommendations. To cope with this issue, Y.
-
M. Li et al. (2017) suggest using social media profiles to
deduce customers’ preferences. This, however, presupposes that users are willing to provide access to
their accounts. Another possibility for addressing the user cold start is the adoption of association rule
mining algorithms for product recommendations (Lawrence et al., 2001; S.
-
L. Lee, 2010; Skiada et al.,
2016; P. Wang et al., 2014). These algorithms do not require identified users, but rather attempt to derive
generic rules for products purchased together (Sarwar et al., 2000). These rules enable recommendations
based on the products with which customers are currently interacting (Shaw et al., 2010).
Absence of explicit product ratings.
In contrast to the context of e-commerce, numerical ratings (e.g., 1-5
star scale) are particularly difficult to obtain in B&M stores (Hansen and Loos, 2008). It is thus necessary
to deduce user preferences from data that only contains implicit feedback in the form of customers’
product purchases. Such data does not, however, contain negative product feedback, as not buying a
product does not indicate dislike (Sahoo et al., 2012). As a result, the majority of collaborative algorithms
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 3
Hanke et al. / Fashion Store Product Recommendation Systems
that depend on explicit ratings cannot be used in cases where only purchase data is available (Rendle
et al., 2009; Sahoo et al., 2012). To tackle this issue, several authors again fall back on association rule
algorithms as these depend on co-purchases and not on the availability of ratings (Cinicioglu and Shenoy,
2016; Fang et al., 2012; Hsu et al., 2004; Jie et al., 2012; Kronberger and Affenzeller, 2011). A major
drawback of this, however, is that the obtained recommendations usually lack personalization (P. Wang
et al., 2014). For example, if a rule is decisive for recommending a specific article to a certain customer,
but irrelevant for the rest of the data, it is desirable to assign this rule a high score for collaborative-filtering,
but typically not for association rules (Verstrepen et al., 2017). To make association rules less generic and
more customer-group-specific, Skiada et al. (2016) propose combining them with clustering approaches
whereas P. Wang et al. (2014) try to capture associations using probabilistic models. Scholars that do not
employ association analysis try to estimate user ratings by measuring physical distances between users
and objects (Kawashima et al., 2006), considering customer movement paths (So and Yada, 2017), or
using product prices and purchased quantities as additional information (Poulopoulos and Kyriazis, 2017).
Another promising possibility is investigated by Sato et al. (2015), who use a ranking-based collaborative
filtering approach designed for datasets that do not contain explicit product ratings (Rendle et al., 2009).
2.2 Contextual Information in B&M Stores
We identified several papers addressing the question of how contextual information can be acquired
and subsequently incorporated into B&M recommendation services. Context-aware systems are able
to adapt their operations to the current situation (e.g., location and time) and aim at increasing service
usability and effectiveness (Baldauf et al., 2007). They have considerable potential (Villegas and Müller,
2010)—particularly in retailing (Grewal et al., 2017)—as additional environmental information influences
the predictability of user preferences (Adomavicius and Tuzhilin, 2011). However, special attention must
be paid to the selection of contextual attributes, as it is difficult to determine their relevance a priori,
especially as the incorporation of irrelevant ones entails the risk of impairing recommendation quality
(Adomavicius and Tuzhilin, 2011; Odi´
c et al., 2013). While many papers investigate the integration of
contextual data into e-commerce recommendation systems (e.g., Jiang et al., 2015; J. Lu et al., 2015),
we found only a few that integrate contextual data into B&M recommendation systems. Five of them
explicitly use context data for the generation of product recommendations and leverage information about
trends and occasions (Wong et al., 2012), user locations (Nakahara and Yada, 2011; So and Yada, 2017),
store locations (Giering, 2008), and customer interactions with products (activities) (Kawashima et al.,
2006). In contrast, scholars who do not recommend products suggest points of interest (e.g., shops in a
mall) or recommend paths that guide customers through stores (e.g., Y.
-
M. Li et al., 2017; E. H.
-
C. Lu
et al., 2012; Z. Zheng et al., 2017) and therefore rely mainly on the locations of users.
For further investigation of relevant contextual attributes, we searched for papers examining contextual data
in e-commerce recommendation systems. We again followed the approach of Webster and Watson (2002).
2
The search was restricted to fashion retail, as context variables are often domain-specific (Adomavicius
and Tuzhilin, 2011). We retrieved 128 publications of which only 14 turned out to be relevant.
We analyzed the B&M and e-commerce papers and divided the context data used in them into categories
(see Figure 1). Frequently mentioned ones are time (e.g., season), location (e.g., geolocation of IP address
or in-store position), occasion, and weather. While most of the attributes can be applied in e-commerce
and B&M, attributes of the categories activity (e.g., trying on garments) (Hauser et al., 2017a) and
surroundings (e.g., shopping area is crowded) (Y.
-
M. Li et al., 2017) can only be collected in B&M stores.
According to Adomavicius and Tuzhilin (2011), context can be used in two ways for generating recom-
mendations: (i) as a search parameter to filter potentially recommendable items and (ii) for estimating
user preferences in certain contextual situations by observing users while they are interacting with the
2
We used the following Google Scholar search string: (intitle:fashion OR intitle:clothing) AND context AND (“recommender
system” OR “recommendation system” OR “recommendation agent”). On Scopus “title” instead of “intitle” has to be used.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 4
Hanke et al. / Fashion Store Product Recommendation Systems
service system. The predominant example relying on the first paradigm is that of ubiquitous systems
that use sensor information to recommend items in close proximity. All the papers from our B&M
recommendation systems article set that consider collaborative-filtering approaches are based on the
first paradigm. As we seek to employ context as auxiliary information to improve recommendations, we
can not use the algorithms applied in these papers. In contrast, we turn to the second paradigm to gear
our recommendations more towards situational user preferences. This paradigm proposes modeling the
context-sensitive preferences of users and generating recommendations by adopting existing collaborative
filtering methods to context-aware recommendation settings. Adomavicius and Tuzhilin (2011) identify
two applicable groups of algorithms. Contextual modeling, on the one hand, aims at integrating context
directly into the modeling technique and thus demands specialized algorithms. Contextual filtering, on the
other hand, uses contextual data to adapt the underlying database prior to actual training (i.e., pre-filtering)
or to adjust the outputted recommendations a posteriori (i.e., post-filtering) (Panniello et al., 2009).
Location
Time
Occasion
Trend
Weather
Mood
Activity
Surroundings
ECIS - Framework
ECIS - Framework
Figure 1. Contextual categories considered in B&M (left box) and e-commerce literature (right box)
3 System Design and Evaluation
In this section, we first (i) propose recommendation algorithms tailored to the sensing capabilities of smart
fitting rooms and (ii) describe a means by which to integrate contextual information. In this paper, we
start with analyzing the contextual categories time,location, and weather. We consider this initial step
crucial for assessing the importance of our project, as considering all contextual categories identified in the
literature review would be pointless if context does not help to improve recommendations. In a second step,
we investigate to what extent the choice of algorithm and the integration of the selected context attributes
affect recommendation quality. For this preliminary evaluation, we use a pre-compiled transaction dataset
from a leading German fashion retailer. Such offline assessments are common evaluation approaches in
recommendation system research (Gunawardana and Shani, 2009). In our offline evaluation we follow the
steps for building predictive models described by Shmueli and Koppius (2011):
Data collection.
The dataset contains 5,142,891 customer transactions, that is, information about when
customers bought particular products in specific stores. The dataset contains transactions from a period of
16 months and comprises a total of 660,472 different customers and 52,902 different products. In addition,
we collected historical weather data that includes the temperature and rainfall of the weather stations
closest to the individual stores (Deutscher Wetterdienst, 2017).
Data preparation.
We carried out several pre-processing steps to enrich the transaction data with con-
textual information. We first derived the attributes season,first week (of the month) (see, e.g., Hastings
and Washington, 2010), and day (weekday or Saturday) using the transaction data timestamps. We then
added the attributes temperature (cold, cool, warm, or hot) and rain (yes or no). To achieve this, we
first aggregated hourly measurements into daily averages and then subdivided these values into attribute
categories. Finally, we derived the attribute store format (city, mall, or standalone).
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 5
Hanke et al. / Fashion Store Product Recommendation Systems
Exploratory data analysis.
The sales in our transaction data follow a long tail distribution, a phenomenon
commonly observed in retailing (Brynjolfsson et al., 2011). 20% of all units sold are realized with only
3% of the available products, while the remaining sales are generated by the other 97% of the products.
However, the recommendation of rarely-purchased articles can lead to especially interesting suggestions
(Cremonesi et al., 2010). Figure 2 yields initial insights into the predictive power of the investigated
contextual attributes. The individual heatmaps show the proportionate sales distributions for the ten
best-selling product categories for the individual categories of the contextual attributes (e.g., yes and no for
the attribute rain). The heatmaps indicate that some situational attributes could influence the customers’
buying decisions (e.g., women’s knitwear seems to be particularly seasonal and temperature-dependent).
Figure 2. Purchase distribution per category subject to contextual situations
Choice of attributes.
Although exploratory data analysis can indicate the predictive power of attributes, it
is not suitable for excluding variables from prediction models a priori (Adomavicius and Tuzhilin, 2011;
Baltrunas et al., 2012). For this reason, we follow Baltrunas et al. (2012) and first train models for each
individual contextual attribute. In a second step, we consider only the contextual attributes that were used
in models that actually led to improved recommendation quality (all except store format) and use them to
build our final, context-aware recommendation system.
Choice of methods.
Our review showed that two main challenges have to be tackled when transferring
collaborative-filtering to cyberphysical contexts (i.e., user cold start and the absence of explicit product
ratings). As outlined, we argue that the applicability of particular recommendation algorithms depends on
the environment’s sensing capabilities and we aim to provide recommendations also in cases where such
information is not available (e.g., because fitting rooms are not equipped with card readers or customers
don’t have loyalty cards). Consequently, we distinguish between four different cases (see Figure 3):
•
If neither customers nor products are identifiable, we use the algorithm popular item, a naive approach
to recommend the most frequently-bought items in transaction datasets (Cremonesi et al., 2010).
•
The identification of the garments customers bring into cabins enables us to tailor the recommendations
to these items. Based on the findings from our first literature review, we suggest employing algorithms
based on association rule mining (ARM) (they are used in 13 of the papers) and following particularly
the approach of Sarwar et al. (2000). To this end, we first identify the products that were frequently
purchased together. In a second step, we select the most interesting rules considering the evaluation
metrics support and confidence. In the subsequent generation of product recommendations, these rules
are used to suggest products that match products customers bring into fitting rooms.
•
Targeting individual customers is only possible if they authenticate themselves (e.g., with loyalty
cards). Once they are identified, algorithms can take into account individual customer purchase
histories and must be able to cope with the lack of explicit product ratings. Algorithms that can
handle implicit feedback can be classified as (i) prediction-based or (ii) ranking-based algorithms
(Takács and Tikk, 2012). Similarly to Sato et al. (2015), we use Bayesian Probabilistic Ranking
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 6
Hanke et al. / Fashion Store Product Recommendation Systems
(BPR), an algorithm that falls into the second category and creates user-specific item rankings by
sampling positive (i.e., items previously purchased by the customer) and negative items (i.e., items not
previously purchased by the customer), as well as running pairwise comparisons (Rendle et al., 2009).
•
If products and customers are detected, it is possible to use algorithms designed to incorporate both
sets of information. Recently, researchers have been working on the development of (sequential)
algorithms that make this possible (e.g., factorized personalized Markov chains) (Rendle, 2010).
After selecting an algorithm tailored to the hardware, the recommendations can be enriched with con-
textual information. We rely on contextual pre-filtering, which allows for the subsequent use of any
recommendation algorithm (see Section 2.2). Specifically, we used item splitting (see, e.g., Baltrunas and
Ricci, 2014) for each of the implemented algorithms. Y. Zheng et al. (2014) describe it as one of the most
efficient pre-filtering techniques. Item splitting identifies relevant contextual situations for purchasing
specific products. If an item is purchased more often in a particular context, the algorithm considers that
item to be two different items. We used the chi-square test to split only items whose sales distributions
differed to a statistically significant degree depending on the context (Baltrunas and Ricci, 2014).
AB JETZT SCHRIFTGRÖSSE 8!!
ECIS - Framework
BPR FPMC
Popular item
Yes
No
Yes
Association rules
No
Product detection
Customer
detection
User cold start
No explicit product ratings
BPR FPMC
Popular item
Yes
No
Yes
ARM
No
Product detection
Customer detection
No explicit ratings
User cold start
Figure 3. B&M recommendation algorithms depending on environment’s sensing capabilities
Evaluation.
In this paper, we consider popular item,ARM, and BPR. We first split the transaction dataset
into two sets. 20% of the customers are randomly selected to be the test customers, while the remaining
80% of the customers are used for model training. To compare the performance of the proposed algorithms,
we use the frequently adopted (e.g., Herlocker et al., 1999) leave-k-out evaluation proposed by Breese
et al. (1998). Following this approach, the purchases of each test customer must be divided into two
different sets. The first set contains 20% of each customer’s purchases (i.e., the product purchases the
algorithms attempt to predict) and is withheld. The algorithms’ objectives is to provide a ranked list of
multiple (
N=1,2, ..., 10
) product recommendations for each customer. During this process, the remaining
80% of each customer’s purchases are used differently depending on the algorithm. For algorithms applied
in cases with identified users, the 80% (i.e., complete customer history) is used to deduce preferences. For
algorithms applied in cases with identified products, on the other hand, each product in the 80% is used
separately to derive item-based recommendations. In the simplest case (i.e., no identification of customers
or products) the 80% is not necessary for the evaluation. Using the withheld 20% allows for a comparison
of the algorithms’ suggestions with the products that customers actually bought. We rely on the common
metrics precision,recall, and F1 score (Cremonesi et al., 2010; Herlocker et al., 2004), to measure the
frequency with which a recommender system makes correct or incorrect predictions about whether an
item is relevant for a user or not. Precision measures the ratio of relevant items (i.e., hits = intersection of
recommended and bought) to the number (N) of recommended items for a certain customer and therefore
reflects the probability that a recommended product is relevant (
P@N=hits
N
). Recall measures the ratio
of products that were actually bought by and recommended to a customer (i.e., hits) in relation to this
customer’s total number (
|T|
) of products in the withheld 20% (
R@N=hits
|T|
). Therefore, recall represents
the probability that a relevant product will be recommended. Precision and recall are contradictory
metrics: with increasing N, the precision decreases while the recall increases (Sarwar et al., 2000). The
F1 score combines both measures into a single score by using the harmonic mean. These measures are
first calculated for each test customer and then averaged. Precision and recall depend heavily on the
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 7
Hanke et al. / Fashion Store Product Recommendation Systems
number of items purchased per user and should therefore not be interpreted as absolute measures, but
should only be used to compare algorithms (Cremonesi et al., 2008). Figure 4 presents the results for
the considered algorithms with and without incorporated contextual information (i.e., season,first week,
day,temperature, and rain). The results show that garment and user identification are valuable, as using
ARM and BPR increases recommendation quality. In addition, the integration of context leads to improved
results, but only for those algorithms that consider product or customer information. In the case of ARM,
F1 improves by 24% (averaged over
N=1,2, ..., 10
). In the case of BPR, context integration leads to an
even stronger F1 improvement of 34% (again averaged over N=1,2, ..., 10).
Figure 4. Preliminary evaluation for different algorithms with and without context information
4 Expected Contribution and Future Work
The present study is concerned with the design of product recommendation systems for fashion stores. We
found that (i) the applicability of different recommendation algorithms depends on the sensing capabilities
of smart retail environments and (ii) the implementation of recommendation systems in the physical world
allows for the integration of additional contextual information (i.e., consideration of customer activities
and information about the shopping area surroundings). Based on these findings, we propose a framework
of suitable recommendation algorithms and describe a means by which to integrate context. In our
preliminary evaluation, we consider smart fitting rooms (IT artifacts that offer product recommendations
on screens within individual cabins); contextual information about purchase times, store types, as well
as weather conditions; and a transaction dataset from a leading German fashion retailer. The evaluation
shows that the ability to identify garments and users in smart fitting room cabins enables the product
recommendation system to generate better recommendations. A further improvement of recommendation
quality can be achieved through the integration of the considered contextual information.
Going forward, we want to conduct a more comprehensive evaluation considering all contextual categories
(in particular activity and surroundings) and additional recommendation algorithms (in particular FPMC)
which promises to yield further recommendation improvements. The collection of such data might require
additional sensor systems (e.g., additional RFID systems or camera systems). In addition, the extraction
of necessary relevant information (e.g., activities of users) from low-level sensor data streams might
necessitate the application of complex data cleansing processes (see, e.g., H. Li et al., 2015). Secondly,
we consider evaluations based on pre-compiled datasets only as a first step towards a comprehensive
evaluation of product recommendation systems (Knijnenburg et al., 2012). To carry out a user-centric
evaluation in a real-world environment, the retailer we are collaborating with has already installed several
RFID-based smart fitting rooms. The data collected by this smart service infrastructure enables one to
determine which products customers bring into individual fitting room cabins and which they end up
buying. This allows for a detailed analysis of cause-and-effect relationships between particular product
recommendations and purchase decisions. Moreover, when aiming for the adoption of recommendation
services, there are a number of other important aspects that have to be considered. To this end, we are
planning on conducting evaluations similar to those proposed by Pu et al. (2011) and Weinhard et al.
(2017) to identify the determinants that motivate users to adopt such technologies.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 8
Hanke et al. / Fashion Store Product Recommendation Systems
References
Adomavicius, G. and A. Tuzhilin (2005). “Toward the next generation of recommender systems: A survey
of the state-of-the-art and possible extensions.” IEEE transactions on knowledge and data engineering
17 (6), 734–749.
Adomavicius, G. and A. Tuzhilin (2011). “Context-aware recommender systems.” In: Recommender
systems handbook. Springer, pp. 217–253.
Akabane, T., S. Kosugi, S. Kimura, and M. Arai (2011). “Method to consider familiarity in clothing
coordination recommender systems.” In: Proceedings of the 3rd International Conference on Computer
Research and Development (ICCRD). IEEE, pp. 22–26.
Baldauf, M., S. Dustdar, and F. Rosenberg (2007). “A survey on context-aware systems.” International
Journal of Ad Hoc and Ubiquitous Computing 2 (4), 263–277.
Baltrunas, L., B. Ludwig, S. Peer, and F. Ricci (2012). “Context relevance assessment and exploitation in
mobile recommender systems.” Personal and Ubiquitous Computing 16 (5), 507–526.
Baltrunas, L. and F. Ricci (2014). “Experimental evaluation of context-dependent collaborative filtering
using item splitting.” User Modeling and User-Adapted Interaction 24 (1-2), 7–34.
Bobadilla, J., F. Ortega, A. Hernando, and J. Bernal (2012). “A collaborative filtering approach to mitigate
the new user cold start problem.” Knowledge-Based Systems 26, 225–238.
Bodapati, A. V. (2008). “Recommendation systems with purchase data.” Journal of marketing research
45 (1), 77–93.
Böhmann, T., J. M. Leimeister, and K. Möslein (2014). “Service Systems Engineering: A Field for Future
Information Systems Research.” Business & Information Systems Engineering (BISE) 6 (2), 73–79.
Breese, J. S., D. Heckerman, and C. Kadie (1998). “Empirical analysis of predictive algorithms for
collaborative filtering.” In: Proceedings of the Fourteenth conference on Uncertainty in artificial
intelligence. Morgan Kaufmann Publishers Inc., pp. 43–52.
Brynjolfsson, E., Y. J. Hu, and D. Simester (2011). “Goodbye Pareto Principle, Hello Long Tail: The
Effect of Search Costs on the Concentration of Product Sales.” Management Science (8), 1373–1386.
Chan, S. and L. Capra (2012). “From online browsing to offline purchases: Analyzing contextual informa-
tion in the retail business.” In: CEUR Workshop Proceedings. Vol. 889.
Choi, S., Y. Yang, B. Yang, and H. Cheung (2015). “Item-level RFID for enhancement of customer
shopping experience in apparel retail.” Computers in Industry 71 (C), 10–23.
Yu-Chu, L., Y. Kawakita, E. Suzuki, and H. Ichikawa (2012). “Personalized clothing-recommendation
system based on a modified Bayesian network.” In: 2012 IEEE/IPSJ 12th International Symposium on
Applications and the Internet (SAINT). IEEE, pp. 414–417.
Cinicioglu, E. N. and P. P. Shenoy (2016). “A new heuristic for learning Bayesian networks from limited
datasets: a real-time recommendation system application with RFID systems in grocery stores.” Annals
of Operations Research 244 (2), 385–405.
Cremonesi, P., Y. Koren, and R. Turrin (2010). “Performance of Recommender Algorithms on Top-n
Recommendation Tasks.” In: Proceedings of the Fourth ACM Conference on Recommender Systems.
ACM, pp. 39–46.
Cremonesi, P., R. Turrin, E. Lentini, and M. Matteucci (2008). “An evaluation methodology for collabora-
tive recommender systems.” In: International Conference on Automated solutions for Cross Media
Content and Multi-channel Distribution. IEEE, pp. 224–231.
Daraghmi, E. Y. and T. Kadoori (2016). “Investigating Consumers’ Adoption of Interactive in-Store
Mobile Shopping Assistant.” In: Pacific Asia Conference on Information Systems, p. 282.
Deutscher Wetterdienst (2017). Klimadaten Deutschland.URL:
https://www.dwd.de/DE/leistungen/
klimadatendeutschland/klarchivstunden.html (visited on 04/07/2017).
Duc-Trong Le Hady W. Lauw, Y. F. (2017). “Basket-Sensitive Personalized Item Recommendation.” In:
Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI-17,
pp. 2060–2066.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 9
Hanke et al. / Fashion Store Product Recommendation Systems
Fang, B., S. Liao, K. Xu, H. Cheng, C. Zhu, and H. Chen (2012). “A novel mobile recommender system
for indoor shopping.” Expert Systems with Applications 39 (15), 11992–12000.
Frost, R. and K. Lyons (2017). “Service Systems Analysis Methods and Components: A Systematic
Literature Review.” Service Science 9 (3), 219–234.
Giering, M. (2008). “Retail sales prediction and item recommendations using customer demographics at
store level.” ACM SIGKDD Explorations Newsletter 10 (2), 84.
Gregory, J. (2015). “The Internet of Things, Revolutionizing the Retail Industry.” Accenture Strategy.
Grewal, D., A. L. Roggeveen, and J. Nordfält (2017). “The future of retailing.” Journal of Retailing 93 (1),
1–6.
Gunawardana, A. and G. Shani (2009). “A survey of accuracy evaluation metrics of recommendation
tasks.” Journal of Machine Learning Research 10, 2935–2962.
Hansen, T. and P. Loos (2008). “RFID-based Recommender Systems in Stationary Trade.” AMCIS 2008
Proceedings, 72.
Hao, T., J. Zhou, Y. Cheng, L. Huang, and H. Wu (2016). “User identification in cyber-physical space:
a case study on mobile query logs and trajectories.” In: Proceedings of the 24th ACM SIGSPATIAL
International Conference on Advances in Geographic Information Systems. ACM, p. 71.
Hastings, J. and E. Washington (2010). “The first of the month effect: consumer behavior and store
responses.” American Economic Journal: Economic Policy 2 (2), 142–162.
Häubl, G. and V. Trifts (2000). “Consumer decision making in online shopping environments: The effects
of interactive decision aids.” Marketing science 19 (1), 4–21.
Hauser, M., M. Griebel, J. Hanke, and F. Thiesse (2017a). “Empowering Smarter Fitting Rooms with
RFID Data Analytics.” In: Proceedings of the 13th International Conference on Wirtschaftsinformatik.
Hauser, M., S. Günther, C. Flath, and F. Thiesse (2017b). “Designing Pervasive Information Systems:
A Fashion Retail Case Study.” In: Proceedings of the Thirty Eighth International Conference on
Information Systems.
He, R. and J. McAuley (2016). “Ups and downs: Modeling the visual evolution of fashion trends with
one-class collaborative filtering.” In: Proceedings of the 25th International Conference on World Wide
Web. International World Wide Web Conferences Steering Committee, pp. 507–517.
Herlocker, J. L., J. A. Konstan, A. Borchers, and J. Riedl (1999). “An algorithmic framework for
performing collaborative filtering.” In: Proceedings of the 22nd annual international ACM SIGIR
conference on Research and development in information retrieval. ACM Press, pp. 230–237.
Herlocker, J. L., J. A. Konstan, L. G. Terveen, and J. T. Riedl (2004). “Evaluating collaborative filtering
recommender systems.” ACM Transactions on Information Systems (TOIS) 22 (1), 5–53.
Hou, J.
-
L. and T.
-
G. Chen (2011). “An RFID-based shopping service system for retailers.” Advanced
Engineering Informatics 25 (1), 103–115.
Hsu, C.
-
N., H.
-
H. Chung, and H.
-
S. Huang (2004). “Mining skewed and sparse transaction data for
personalized shopping recommendation.” Machine Learning 57 (1), 35–59.
Hu, X., S. Yu, M. Shobu, and U. Sumita (2016). “Development of Recommendation Engines for Enhancing
Sales of DIY (Do It Yourself) Stores.” In: 2016 5th IIAI International Congress on Advanced Applied
Informatics (IIAI-AAI). IEEE, pp. 468–473.
Jagadeesh, V., R. Piramuthu, A. Bhardwaj, W. Di, and N. Sundaresan (2014). “Large scale visual
recommendations from street fashion images.” In: Proceedings of the 20th ACM SIGKDD international
conference on Knowledge discovery and data mining. ACM, pp. 1925–1934.
Jiang, P., Y. Zhu, Y. Zhang, and Q. Yuan (2015). “Life-stage prediction for product recommendation in
e-commerce.” In: Proceedings of the 21th ACM SIGKDD International Conference on Knowledge
Discovery and Data Mining. ACM, pp. 1879–1888.
Jie, C., W. Dong, and L. Canquan (2012). “Recommendation system technologies of intelligent large-scale
shopping mall.” In: 2nd International Conference on Computer Science and Network Technology
(ICCSNT). IEEE, pp. 1058–1062.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 10
Hanke et al. / Fashion Store Product Recommendation Systems
Kamei, K., T. Ikeda, H. Kidokoro, M. Shiomi, A. Utsumi, K. Shinozawa, T. Miyashita, and N. Hagita
(2011). “Effectiveness of cooperative customer navigation from robots around a retail shop.” In:
Privacy, Security, Risk and Trust (PASSAT) and Third Inernational Conference on Social Computing
(SocialCom). IEEE, pp. 235–241.
Kamei, K., K. Shinozawa, T. Ikeda, A. Utsumi, T. Miyashita, and N. Hagita (2010). “Recommendation
from robots in a real-world retail shop.” In: International Conference on Multimodal Interfaces and
the Workshop on Machine Learning for Multimodal Interaction. ACM, p. 19.
Al-Kassab, J., N. Mahmoud, F. G. Thiesse, and E. Fleisch (2009). “A Cost Benefit Calculator for RFID
Implementations in the Apparel Retail Industry.” AMCIS 2009 Proceedings, 478.
Kawashima, H., T. Matsushita, S. Satake, M. Imai, Y. Shinagawa, and Y. Anzai (2006). “PORSCHE: A
physical objects recommender system for cell phone users.” In: Proceedings of the Second Interna-
tional Workshop on Personalized Context Modeling and Management for UbiComp Applications.
Keller, B., R. Schmidt, M. Möhring, R.
-
C. Härting, and A. Zimmermann (2015). “Social-data driven
sales processes in local clothing retail stores.” In: International Conference on Business Process
Management. Springer, pp. 305–315.
Keller, T. and M. Raffelsieper (2014). “Cosibon: an E-commerce like platform enabling bricks-and-mortar
stores to use sophisticated product recommender systems.” In: Proceedings of the 8th ACM Conference
on Recommender systems. ACM, pp. 367–368.
Knijnenburg, B. P., M. C. Willemsen, Z. Gantner, H. Soncu, and C. Newell (2012). “Explaining the
user experience of recommender systems.” User Modeling and User-Adapted Interaction 22 (4-5),
441–504.
Kobayashi, M., F. Minami, T. Ito, and S. Tojo (2008). “An implementation of goal-oriented fashion
recommendation system.” New Challenges in Applied Intelligence Technologies, 77–86.
Kowatsch, T. and W. Maass (2010a). “In-store consumer behavior: How mobile recommendation agents
influence usage intentions, product purchases, and store preferences.” Computers in Human Behavior
26 (4), 697–704.
Kowatsch, T. and W. Maass (2010b). “Online vs. In-Store Shopping: How Problem Solving Strategies
of Decision Support Systems Influence Confidence in Purchase Decisions.” In: Proceedings of 18th
European Conference on Information Systems (ECIS).
Kowatsch, T., W. Maass, and E. Fleisch (2009). “The Use of Free and Paid Digital Product Reviews on
Mobile Devices in In-Store Purchase Situations.” In: 4th Mediterranean Conference on Information
Systems (MCIS 09). Athens University of Economics & Business, pp. 114–124.
Kronberger, G. and M. Affenzeller (2011). “Market basket analysis of retail data: supervised learning
approach.” In: International Conference on Computer Aided Systems Theory. Springer, pp. 464–471.
Kroon, J., C. Vrijlandt, S. de Ridder, and F. van der Reep (2007). “RFID in Retail: New approaches, new
viewpoints.” European Retail Digest 55, 33.
Kumar, R. and K. Vaccaro (2017). “An experimentation engine for data-driven fashion systems.” In:
Proceedings of 2017 AAAI Spring Symposium, pp. 389–394.
Landmark, A. D. and B. Sjøbakk (2017). “Tracking customer behaviour in fashion retail using RFID.”
International Journal of Retail & Distribution Management 45 (7/8), 844–858.
Lawrence, R. D., G. S. Almasi, V. Kotlyar, M. Viveros, and S. S. Duri (2001). “Personalization of
supermarket product recommendations.” In: Applications of Data Mining to Electronic Commerce.
Springer, pp. 11–32.
Lee, S.
-
L. (2010). “Commodity recommendations of retail business based on decisiontree induction.”
Expert Systems with Applications 37 (5), 3685–3694.
Lee, Y. E. and I. Benbasat (2010). “Interaction design for mobile product recommendation agents:
Supporting users’ decisions in retail stores.” ACM Transactions on Computer-Human Interaction
(TOCHI) 17 (4), 17.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 11
Hanke et al. / Fashion Store Product Recommendation Systems
Li, H., C. Ye, and A. P. Sample (2015). “IDSense: A Human Object Interaction Detection System Based
on Passive UHF RFID.” In: Proceedings of the 33rd Annual ACM Conference on Human Factors in
Computing Systems. ACM, pp. 2555–2564.
Li, J., L. Zhang, F. Meng, and F. Li (2014). “Recommendation algorithm based on link prediction and
domain knowledge in retail transactions.” Procedia Computer Science 31, 875–881.
Li, Y.
-
M., L.
-
F. Lin, and C.
-
C. Ho (2017). “A social route recommender mechanism for store shopping
support.” Decision Support Systems 94, 97–108.
Liaghat, Z., J. Melia-Segui, R. Pous, and R. De Porrata-Doria (2013). “Consumer experience modeling and
enrichment using RFID.” In: Proceedings of the 2013 ACM conference on Pervasive and ubiquitous
computing adjunct publication. ACM, pp. 247–250.
Liang, T.
-
P., H.
-
J. Lai, and Y.
-
C. Ku (2006). “Personalized content recommendation and user satisfaction:
Theoretical synthesis and empirical findings.” Journal of Management Information Systems 23 (3),
45–70.
Liao, S.
-
H., C.
-
H. Wen, P.
-
Y. Hsian, C.
-
W. Li, and C.
-
W. Hsu (2014). “Mining Customer Knowledge for
a Recommendation System in Convenience Stores.” International Journal of Data Warehousing and
Mining (IJDWM) 10 (2), 55–86.
Liu, S., L. M. Brown, Q. Chen, J. Huang, L. Liu, and S. Yan (2017). “Visual Attributes for Fashion
Analytics.” In: Visual Attributes. Springer, pp. 215–243.
Liu, S., L. Liu, and S. Yan (2013). “Magic mirror: An intelligent fashion recommendation system.” In:
2013 2nd IAPR Asian Conference on Pattern Recognition. IEEE, pp. 11–15.
Lu, E. H.
-
C., W.
-
C. Lee, and V. S.
-
M. Tseng (2012). “A framework for personal mobile commerce pattern
mining and prediction.” IEEE Transactions on Knowledge and Data Engineering 24 (5), 769–782.
Lu, J., D. Wu, M. Mao, W. Wang, and G. Zhang (2015). “Recommender system application developments:
a survey.” Decision Support Systems 74, 12–32.
Luo, P., S. Yan, Z. Liu, Z. Shen, S. Yang, and Q. He (2016). “From online behaviors to offline retailing.”
In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and
Data Mining. ACM, pp. 175–184.
Manyika, J., M. Chui, P. Bisson, J. Woetzel, R. Dobbs, J. Bughin, and D. Aharon (2015). The Internet
of Things: Mapping the Value Beyond the Hype. McKinsey Global Institute. URL:
http://www .
mckinsey .com /business - functions /digital - mckinsey/ our - insights/ the- internet -
of-things-the-value-of-digitizing-the-physical-world (visited on 09/01/2017).
Melià-Seguí, J., R. Pous, A. Carreras, M. Morenza-Cinos, R. Parada, Z. Liaghat, and R. De Porrata-Doria
(2013). “Enhancing the Shopping Experience through RFID in an Actual Retail Store.” In: Proceedings
of the 2013 ACM Conference on Pervasive and Ubiquitous Computing Adjunct Publication. ACM,
pp. 1029–1036.
Nakahara, T. and K. Yada (2011). “Extraction of customer potential value using unpurchased items
and in-store movements.” Knowledge-Based and Intelligent Information and Engineering Systems,
295–303.
National Science Foundation (2014). Partnerships for Innovation: Building Innovation Capacity (PFI:BIC).
Technical Report Program Solicitation NSF14-610. National Science Foundation.
Ngai, E., K. K.
-
L. Moon, J. N. Liu, K. Tsang, R. Law, F. Suk, and I. Wong (2008). “Extending CRM
in the retail industry: an RFID-based personal shopping assistant system.” Communications of the
Association for Information Systems 23 (1), 16.
Odi´
c, A., M. Tkalˇ
ciˇ
c, J. F. Tasiˇ
c, and A. Košir (2013). “Predicting and detecting the relevant contextual
information in a movie-recommender system.” Interacting with Computers 25 (1), 74–90.
Ostrom, A. L., M. J. Bitner, S. W. Brown, K. A. Burkhard, M. Goul, V. Smith-Daniels, H. Demirkan, and
E. Rabinovich (2010). “Moving forward and making a difference: research priorities for the science of
service.” Journal of Service Research 13 (1), 4–36.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 12
Hanke et al. / Fashion Store Product Recommendation Systems
Panniello, U., A. Tuzhilin, M. Gorgoglione, C. Palmisano, and A. Pedone (2009). “Experimental compari-
son of pre-vs. post-filtering approaches in context-aware recommender systems.” In: Proceedings of
the third ACM conference on Recommender systems. ACM, pp. 265–268.
Pazzani, M. J. and D. Billsus (2007). “Content-based recommendation systems.” In: The adaptive web.
Springer, pp. 325–341.
Pfeiffer, J., T. Pfeiffer, and M. Meißner (2015). “Towards attentive in-store recommender Systems.” In:
Reshaping society through analytics, collaboration, and decision support. Springer, pp. 161–173.
Piazza, A., P. Kröckel, and F. Bodendorf (2017). “Emotions and fashion recommendations: evaluating
the predictive power of affective information for the prediction of fashion product preferences in
cold-start scenarios.” In: Proceedings of the International Conference on Web Intelligence. ACM,
pp. 1234–1240.
Plumbaum, T. and B. Kille (2015). “Personalized fashion advice.” In: Smart Information Systems. Springer,
pp. 213–237.
Poulopoulos, D. and D. Kyriazis (2017). “Collaborative Filtering for Producing Recommendations in the
Retail Sector.” In: Proceedings of the European Mediterranean and Middle Eastern Conference on
Information Systems. Springer, pp. 662–669.
Pous, R., J. Melià-Seguı, A. Carreras, M. Morenza-Cinos, and Z. Rashid (2013). “Cricking: customer-
product interaction in retail using pervasive technologies.” In: Proceedings of the 2013 ACM conference
on Pervasive and ubiquitous computing adjunct publication. ACM, pp. 1023–1028.
Pu, P., L. Chen, and R. Hu (2011). “A user-centric evaluation framework for recommender systems.” In:
Proceedings of the fifth ACM conference on Recommender systems. ACM, pp. 157–164.
Reischach, F. von, D. Guinard, F. Michahelles, and E. Fleisch (2009). “A mobile product recommendation
system interacting with tagged products.” In: International Conference on Pervasive Computing and
Communications, 2009. IEEE, pp. 1–6.
Rendle, S. (2010). “Factorization machines.” In: 10th International Conference on Data Mining (ICDM).
IEEE, pp. 995–1000.
Rendle, S., C. Freudenthaler, Z. Gantner, and L. Schmidt-Thieme (2009). “BPR: Bayesian personalized
ranking from implicit feedback.” In: Proceedings of the twenty-fifth conference on uncertainty in
artificial intelligence. AUAI Press, pp. 452–461.
Resatsch, F., U. Sandner, J. M. Leimeister, and H. Krcmar (2008). “Do Point of Sale RFID-Based
Information Services Make a Difference? Analyzing Consumer Perceptions for Designing Smart
Product Information Services in Retail Business.” Electronic Markets 18 (3), 216–231.
Sahoo, N., P. V. Singh, and T. Mukhopadhyay (2012). “A hidden Markov model for collaborative filtering.”
MIS Quarterly 36 (4), 1329–1356.
Sano, N., N. Machino, K. Yada, and T. Suzuki (2015). “Recommendation system for grocery store
considering data sparsity.” Procedia Computer Science 60, 1406–1413.
Sarwar, B., G. Karypis, J. Konstan, and J. Riedl (2000). “Analysis of recommendation algorithms for
e-commerce.” In: ACM Press, pp. 158–167.
Sato, M., H. Izumo, and T. Sonoda (2015). “Discount Sensitive Recommender System for Retail Business.”
In: Proceedings of the 3rd Workshop on Emotions and Personality in Personalized Systems 2015.
ACM, pp. 33–40.
Schafer, J. B., J. Konstan, and J. Riedl (1999). “Recommender systems in e-commerce.” In: Proceedings
of the 1st ACM conference on Electronic commerce. ACM, pp. 158–166.
Shaw, G., Y. Xu, and S. Geva (2010). “Using association rules to solve the cold-start problem in recom-
mender systems.” Advances in Knowledge Discovery and Data Mining, 340–347.
Shmueli, G. and O. R. Koppius (2011). “Predictive analytics in information systems research.” Mis
Quarterly, 553–572.
Skiada, M., G. Lekakos, S. Gkika, and C. Bardaki (2016). “Garment Recommendations for Online
and Offline Consumers.” In: Proceedings of the Mediterranean Conference on Information Systems
(MCIS), p. 62.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 13
Hanke et al. / Fashion Store Product Recommendation Systems
So, W. T. and K. Yada (2017). “A Framework of Recommendation System Based on In-store Behavior.”
In: Proceedings of the 4th Multidisciplinary International Social Networks Conference. ACM, p. 33.
Takács, G. and D. Tikk (2012). “Alternating least squares for personalized ranking.” In: Proceedings of
the sixth ACM conference on Recommender systems. ACM, pp. 83–90.
Thiesse, F., J. Al-Kassab, and E. Fleisch (2009). “Understanding the Value of Integrated RFID Systems:
A Case Study from Apparel Retail.” European Journal of Information Systems 18 (6), S. 592-614.
Vaccaro, K., S. Shivakumar, Z. Ding, K. Karahalios, and R. Kumar (2016). “The Elements of Fashion
Style.” In: Proceedings of the 29th Annual Symposium on User Interface Software and Technology.
ACM, pp. 777–785.
Verstrepen, K., K. Bhaduriy, B. Cule, and B. Goethals (2017). “Collaborative Filtering for Binary,
Positiveonly Data.” ACM SIGKDD Explorations Newsletter 19 (1), 1–21.
Villegas, N. M. and H. A. Müller (2010). “Managing Dynamic Context to Optimize Smart Interactions
and Services.” The smart internet 6400, 289–318.
Walter, F. E., S. Battiston, M. Yildirim, and F. Schweitzer (2012). “Moving recommender systems from
on-line commerce to retail stores.” Information Systems and e-Business Management 10 (3), 367–393.
Wang, Y.
-
F., Y.
-
L. Chuang, M.
-
H. Hsu, and H.
-
C. Keh (2004). “A personalized recommender system for
the cosmetic business.” Expert Systems with Applications 26 (3), 427–434.
Wang, L., X. Zeng, L. Koehl, and Y. Chen (2015). “Intelligent fashion recommender system: Fuzzy logic
in personalized garment design.” IEEE Transactions on Human-Machine Systems 45 (1), 95–109.
Wang, P., J. Guo, and Y. Lan (2014). “Modeling retail transaction data for personalized shopping recom-
mendation.” In: Proceedings of the 23rd ACM International Conference on Conference on Information
and Knowledge Management. ACM, pp. 1979–1982.
Webster, J. and R. T. Watson (2002). “Analyzing the past to prepare for the future: Writing a literature
review.” MIS quarterly 26 (2), xiii–xxiii.
Weinhard, A., M. Hauser, and F. Thiesse (2017). “Explaining Adoption of Pervasive Retail Systems with
a Model based on UTAUT2 and the Extended Privacy Calculus.” In: Proceedings of the 21st Pacific
Asia Conference on Information Systems.
Weiser, M. (1999). “The computer for the 21st century.” Mobile Computing and Communications Review
3 (3), 3–11.
Wong, W.
-
K., S. Leung, Z. Guo, X. Zeng, and P. Mok (2012). “Intelligent Product Cross-Selling System
with Radio Frequency Identification Technology for Retailing.” International Journal of Production
Economics 135 (1), 308–319.
Zeng, X., L. Koehl, L. Wang, and Y. Chen (2013). “An intelligent recommender system for personalized
fashion design.” In: IFSA World Congress and NAFIPS Annual Meeting (IFSA/NAFIPS), 2013 Joint.
IEEE, pp. 760–765.
Zhang, W., T. Matsumoto, J. Liu, M. Chu, and B. Begole (2008). “An intelligent fitting room using
multi-camera perception.” In: Proceedings of the 13th international conference on Intelligent user
interfaces. ACM, pp. 60–69.
Zhang, X., J. Pei, and X. Ye (2016). “Demographic transformation and clustering of transactional data for
sales prediction of convenience stores.” In: International Conference on Cloud Computing and Big
Data Analysis (ICCCBDA). IEEE, pp. 102–108.
Zhang, X., J. Jia, K. Gao, Y. Zhang, D. Zhang, J. Li, and Q. Tian (2017). “Trip Outfits Advisor: Location-
Oriented Clothing Recommendation.” IEEE Transactions on Multimedia 19 (11).
Zheng, Y., R. Burke, and B. Mobasher (2014). “Splitting approaches for context-aware recommendation:
An empirical study.” In: Proceedings of the 29th Annual ACM Symposium on Applied Computing.
ACM, pp. 274–279.
Zheng, Z., Y. Chen, S. Chen, L. Sun, and D. Chen (2017). “Location-Aware POI Recommendation for
Indoor Space by Exploiting WiFi Logs.” Mobile Information Systems 2017.
Twenty-Sixth European Conference on Information Systems (ECIS2018), Portsmouth, UK, 2018 14
