=Paper=
{{Paper
|id=Vol-1685/paper7
|storemode=property
|title=Anything Fun Going On? A Simple Wizard to Avoid the Cold-Start Problem for Event Recommenders
|pdfUrl=https://ceur-ws.org/Vol-1685/paper7.pdf
|volume=Vol-1685
|authors=Stacey Donohue,Nevena Dragovic,Maria Soledad Pera
|dblpUrl=https://dblp.org/rec/conf/recsys/DonohueDP16
}}
==Anything Fun Going On? A Simple Wizard to Avoid the Cold-Start Problem for Event Recommenders==
https://ceur-ws.org/Vol-1685/paper7.pdf
Anything Fun Going On? A Simple Wizard to Avoid the
Cold-Start Problem for Event Recommenders
Stacey Donohue Nevena Dragovic & Maria Soledad Pera
relEVENTcity Dept. of Computer Science
stacey@releventcity.com Boise State University,Boise, ID, USA
{ndragovic,solepera}@boisestate.edu
ABSTRACT domain [3]. In fact, most existing works in this domain
In this demo, we showcase a set up wizard designed to bypass focus on suggesting specific places or events. Places are of-
the cold start problem that often affects recommendation ten associated with well-known geographical locations, i.e.,
systems in the event domain. We have developed a mobile Points-of-interest (PoI) [4], such as the Eiffel Tower or New
application for tourists, RelEVENT, which allows them to York Yankees Stadium. Events, on the other hand, are usu-
quickly and non-intrusively set up preferences and/or inter- ally short-lived and varied in nature. Within the tourism
ests related to events. This will directly affect the degree domain, events pose a special challenge for recommendation
to which they can receive personalized recommendations on- strategies given the lack of uniform event metadata and his-
the-fly and become aware of events happening around town torical information in the form of personal ratings. Events
that might be appealing to them. are varied in nature, ranging from concerts and sports games
to small gatherings or dinner parties and can occur in di-
verse locations that can often change and do not necessarily
CCS Concepts correspond to a PoI.
•Information systems → Decision support systems; Regardless of the domain, cold start is one of the most
Recommender systems; Personalization; “popular” challenges that hinders all recommendation sys-
tems. Cold start occur when the system is not able to create
Keywords recommendations due to unavailable historical data for new
users or items. This can be the reason why recommenders
Recommendation system; cold start; events; wizard; mobile cannot be more successful in creating personalized sugges-
application tions and linking items to users. The cold start challenge is
even harder to solve in the case of suggesting events. This is
1. INTRODUCTION due to the fact that events have short time span and cannot
Recommendation systems can help users in locating items be recommended after they end [2]. While this complicates
(e.g., products and services) of interest more quickly by filter- the issue from an event perspective, we can address this
ing and ranking them based on some criteria, i.e., location, problem by focusing on the users instead.
popularity, or preference, to name a few [1]. No matter if it In this demo, we present wizard used by RelEVENT, the
is related to shopping web-sites (Amazon, e-bay, cheapoair, mobile recommendation application we developed, for by-
etc.), news related web-sites (Yahoo, CNN, etc.), hotel search passing the cold start problem in suggesting events at specific
or restaurant search, recommendation systems have a huge in- cities that people may find useful or interesting during their
fluence on businesses success and users’ satisfaction. Thanks visit. The goal of this wizard is to collect enough data a-
to those systems, companies and products are able to get ad- priory to provide personalized suggestions without imposing
vertisement by being offered to potential buyers. At the same too much burden on the users. While the idea of a wizard
time, recommenders enhance users’ experience by assisting is not unique to RelEVENT, to the best of our knowledge,
them in finding information pertaining to their preferences. our strategy is the first one that offers a balance of initial
Recommenders focusing on common products or services, information to personalize suggestions and differs from strate-
such as books, movies, or restaurants, have been well-studied gies in the tourism domains, such as the one presented in
and developed. However, research efforts related to recom- [Bor15], which focus on type of traveler group, age, date,
mendations within the tourism domain are less prolific and and motivation. We are aware that some users may prefer
must address novel challenges pertaining specifically to this to bypass such a wizard, in which case the default options
will still aid RelEVENT in providing suggestions tailored
to proximity and popularity, i.e., provided suggestions will
relate to the most popular events at that time in a given city.
2. OVERVIEW OF THE SYSTEM
RelEVENT includes the wizard made a specific set of
Copyright held by the author(s). questions that helps RelEVENT in filtering events for new
RecTour 2016 - Workshop on Recommenders in Tourism held in conjunc-
tion with the 10th ACM Conference on Recommender Systems (RecSys), users and avoiding the cold start problem and offer on-the-fly
September 15, 2016, Boston, MA, USA. suggestions.
(a) Categories (b) Location (c) Demographic (d) Context
Figure 1: Snapshots of RelEVENT setup wizard
As shown in Figure 1(a), initially, our application will ask match the physical abilities and expectations of each user, the
a user to select a number of well-known categories of interest. time, date, and budget will ensure that suggested activities
In addition, we included Facebook as a special category to are appealing to each users.
allow users to include in their list of possible events to be rec-
ommended events that are publically available on Facebook. 3. CONCLUSION
In doing so, our application can not only recommend the
In this demo we described the solution we implemented to
more “typical” events happening around a specific location,
deal with the cold start problem affecting event recommen-
from conferences to movies to sales, but can also focus on
dation systems. To provide the most relevant suggestions
more spontaneous and unique events, such a technical group
to each user, we created a short wizard that will allow new
meeting, e.g., ACM-W meeting at a university or âĂe,
users to off RelEVENT enough information about their in-
Tourists can be visiting a location for a short period of time
terests and preferences to instantaneously receive appealing
for an extended vacation. With that in mind (as illustrated
suggestions. Based on initial testing and feedback collected
in Figure 1(b)), by default, a user will receive recommen-
from users, we are encouraged with the performance and
dations occurring within seven days. However, if desired,
usability of wizard.
they will have the opportunity to decrease or increase the
range of dates from which recommendations will be gener-
ated. A key aspect of recommendations related to tourism is 4. REFERENCES
distance. Users may favour events within close proximity or [1] F. Gedikli, D. Jannach, and M. Ge. How should i
may be willing and able to move farther around town. Our explain? a comparison of different explanation types for
application uses by default a 20 miles radius to limit the recommender systems. International Journal of
locations where events to be suggested occur. This radius Human-Computer Studies, 72(4):367–382, 2014.
can be adjusted by each user based on their own preferences [2] H. Khrouf and R. Troncy. Hybrid event recommendation
and needs. using linked data and user diversity. In Proceedings of
Age (shown in Figure 1(c)) is another dimension considered the 7th ACM conference on Recommender systems,
by RelEVENT. While not novel, it is one of the simplest pages 185–192. ACM, 2013.
questions that will help the recommender engine eliminate [3] A. Moreno, L. Sebastiá, and P. Vansteenwegen. Tours’15:
from their set of candidate events to recommend those that do Workshop on tourism recommender systems. In
not target the demographic of the user.More importantly, it Proceedings of the 9th ACM Conference on
will help eliminate from the list of possible recommendations Recommender Systems, pages 355–356. ACM, 2015.
those pertaining to events that occur where minors cannot [4] Y.-T. Zheng, Z.-J. Zha, and T.-S. Chua. Mining travel
attend. patterns from geotagged photos. ACM Transactions on
As shown in Figure 1(d), the most interactive set of ques- Intelligent Systems and Technology (TIST), 3(3):56,
tions appear at the end of the wizard. RelEVENT is in- 2012.
terested in finding out, if possible, the context or type of
activities a visitor has in mind. In doing so, the recommender
the recommender engine will be able to further narrow down
the options available for each user and thus further personal-
ize the provided recommendations. While Level of activity
and Overall intention of events will lead to suggestions that