=Paper=
{{Paper
|id=Vol-1278/paper13
|storemode=property
|title=Towards a Residential Micro-Location Based Product and Service Recommender System
|pdfUrl=https://ceur-ws.org/Vol-1278/paper13.pdf
|volume=Vol-1278
|dblpUrl=https://dblp.org/rec/conf/dmrs/KellerKT14
}}
==Towards a Residential Micro-Location Based Product and Service Recommender System==
https://ceur-ws.org/Vol-1278/paper13.pdf
Towards a Residential Micro-Location Based
Product and Service Recommender System
Thorben Keller1 , Marcus Koehler2 , and Stefanie Turber3
1
University of St.Gallen, Switzerland, thorben.keller@unisg.ch
2
University of St.Gallen, Switzerland, marcus.koehler@unisg.ch
3
University of St.Gallen, Switzerland, stefanie.turber@unisg.ch
Abstract. Mobile smartphone applications can be used to provide users
with suitable location-based product or service recommendations. Re-
cent developments of micro-location based recommender systems take
the customers’ in-store location into account and try to provoke imme-
diate purchases decisions in close proximity to the actual product. Fur-
thermore, an increasing number of mobile retailing apps allow to make
purchases online using the phone itself as the point-of-sale (POS) and
thus making the existence of a physical POS obsolete. However, such
online retailing applications do not leverage the customers’ micro loca-
tion. As people spend most of their free time at home we will implement
a micro-location based recommender system in the Smart Home con-
text. We suggest a novel approach of mapping rooms in a customers’
apartment to departments in a retail store, thereby transferring the idea
of in-store marketing to in-home marketing. As a result we construct
an improved recommender system for (1) micro-location based in-home
marketing and (2) interest-based marketing based on customers’ domes-
tic room preferences.
1 Introduction
Context aware product recommender systems are an ongoing field of research [1].
Contextual information in this case can for example be the customers’s mood, her
age, or the current date and time. The well-established pervasiveness of smart-
phones even allows to extend a customer’s context by location (e.g., [4]) which is
easily derived using GPS or WiFi. As soon as the location of a person is known it
is simple to recommend suitable products (e.g., There’s a Starbucks just around
the corner ) or services (e.g., The next gas station is only 100 meters away). In
the retail industry it is of particular interest to know where exactly in the store
a customer is located to send out personalized coupons or offers related to the
specific department. While in theory very interesting, it has been shown that
the technology available is either inconvenient (e.g., NFC tags placed on shelf)
or unreliable (e.g., WiFi triangulation). Lately Apple introduced iBeacon which
overcomes some weaknesses of GPS and WiFi and theoretically allows in-store
navigation. That way a recommender system would be able to recommend truly
context aware recommendations using micro-locating services, e.g., promoting
current wine discounts only to customers standing in the wine department.
It is reasonable that all those technologies try to provide recommendations
near the actual product to provoke immediate purchase decision. Various mo-
bile shopping apps exist, allowing customers to purchase products or services
wherever they are and independent of a physical point-of-sale in their proximity.
Thus, as smartphones are ubiquitous the point of sale became ubiquitous as well
(cf. Amazon or Ebay). Taking this into account we are currently developing a
micro-location based recommender system that is aware of the current location
of the customer in her own apartment, allowing to recommend suitable products
there similar to in-store promotions.
2 A residential micro-location
based recommender system
We extend the idea of offering micro-location based recommendations in-store
to an in-home scenario by arguing that customers are more accessible to rec-
ommendations if they factor in the current situation. Transferring the idea of
in-store navigation where we would recommend wine related offers if standing
in the wine area we recommend products and services based on the customers
position in the apartment. Figure 1 shows an architectural plan of an average
apartment. For the sake of simplicity we assume every apartment has a living
room where people probably spend most of their time, a bathroom for personal
hygiene, a kitchen for preparing (and maybe consuming) food, and a bedroom
to get rest.
Fig. 1. Mapping the customers’ apartment to retail store departments
We furthermore assume that there is some kind of entrance where the event
of somebody entering or leaving the apartment can be tracked. Similar to in-
store locating it is not trivial to locate a customer in a specific room of her
apartment. In our experimental setup we will use the Comfy 4 security solution
which is capable of determining the presence of a user in a specific room. This
4
Comfy is a startup developing a smart security solution which can detect presence
of individuals in specific rooms.
presence information is then used by Cosibon’s5 mobile product recommender
system to determine relevant products in the consumer’s current context.
We regard the apartment of a user as a retail store and the room she is
currently residing in as the respective department. Consequently we will only
recommend products and services related to the respective department. That
might be consumer electronics or Blu-Ray movies in the living room, kitchen
equipment or food in the kitchen, sanitary products in the bathroom and so forth.
We might also argue that customers using their smartphones in the bathroom
are bored and have nothing else to do, thus actually being accessible to any
kind of product or service. A sample mapping of a customer’s apartment to the
respective departments is shown in Figure 1. We also included some examples
of suitable product categories.
3 A micro-location based recommender system at home
In this section we suggest two approaches of how to leverage the knowledge
derived from Comfy’s security light solution (or any other Smart Home technol-
ogy capable of capturing presence data on a room level) and use it as input to
Cosibon’s mobile recommender system (or any other recommender system).
Let A = LivingRoom, Bathroom, Kitchen, Bedroom be an apartment, P be
the set of available products and services and p ∈ P be an item. Every product
has an associated set pd ⊆ A indicating the relevant rooms (i.e., departments).
The set S of products to be recommended for a given room r ∈ A is then simply
calculated as
S(r) = {p ∈ P | r ∈ pd }.
While this first approach is limited by the fact that the customer actually needs
to be in her apartment, the second approach uses a weighted measure to deter-
mine how relevant a product is based on the presence history. Therefore, the
second approach is applicable everywhere, since it relates the presence with the
product. By evaluating the presence data we can derive the importance of a
room to an individuum, mapping this against the importance of a product to a
room using some similarity function. Furthermore, for every room r we assign a
weight ωr indicating the usefulness of that product for the room.
P presencecustomer (r)
r∈A wr (p) · presenceavg (r)
sim(p, A) = 1S(A) (p) P
r∈A r (p)
w
That way for example, if a customer spends a lot of time in the kitchen
although we know she has also has a living room we might assume she is inter-
ested in cooking and thus will more likely recommend kitchen related products
or services.
5
Cosibon is a startup developing mobile apps for retailers which include advanced
product recommendation algorithms amongst other features. If integrated with a re-
tailer the system has access to a customers’ purchase history together with extensive
tracking data to evaluate her purchase intentions and feedback.
In any case both approaches are planned to be part of a two-step recom-
mender system. We will use the presented approaches to pre-sort results based
on the respective room (e.g., living-room vs. kitchen) and then use Cosibon’s
existing recommendation algorithm to return a more detailed recommendation
based on further information like purchase history or personal interests.
4 Conclusion
In this paper we have suggested a novel approach of how to leverage smart home
devices to map a customers’ apartment to respective retail store departments.
This can be used to enhance existing product and service recommendation al-
gorithms. As to our knowledge there is no ongoing research in this field, we
would be happy to present out approach at RecSys’14 in order to gain valuable
feedback from the recommender systems community.
References
1. G. Adomavicius and A. Tuzhilin. Recommender Systems Handbook, chapter
Context-Aware Recommender Systems, pages 217–253. Springer, 2011.
2. Comfy Startup Project. http://www.comfyhome.eu.
3. Cosibon AG. http://www.cosibon.com.
4. M.-H. Park, J.-H. Hong, and S.-B. Cho. Location-based recommendation system
using bayesian user’s preference model in mobile devices. UIC’07, pages 1130–1139,
2007.