=Paper=
{{Paper
|id=Vol-1906/paper3
|storemode=property
|title=Context-Aware Tourist Trip Recommendations
|pdfUrl=https://ceur-ws.org/Vol-1906/paper3.pdf
|volume=Vol-1906
|authors=Christopher Laß,Daniel Herzog,Wolfgang Wörndl
|dblpUrl=https://dblp.org/rec/conf/recsys/LassHW17
}}
==Context-Aware Tourist Trip Recommendations==
https://ceur-ws.org/Vol-1906/paper3.pdf
Context-Aware Tourist Trip Recommendations
Christopher Laß Daniel Herzog Wolfgang Wörndl
Department of Informatics Department of Informatics Department of Informatics
Technical University of Munich Technical University of Munich Technical University of Munich
Boltzmannstr. 3 Boltzmannstr. 3 Boltzmannstr. 3
85748 Garching bei München, 85748 Garching bei München, 85748 Garching bei München,
Germany Germany Germany
christopher.lass@tum.de herzogd@in.tum.de woerndl@in.tum.de
ABSTRACT (TTDP) [13]. The TTDP defines the generic problem of personalized
Mobile and web-based services solving common tourist trip design tourist trip generation and is commonly seen as an extension of the
problems are available, but only few solutions consider context for Orienteering Problem (OP) [23]. The basic idea is to maximize an
the recommendation of point of interest (POI) sequences. In this objective score between an specific start and end point with several
paper, we present a novel approach to incorporating context into a POIs in between [11].
tourist trip recommendation algorithm. In addition to traditional In previous works, we aimed to solve the TTDP and introduced
context factors in tourism, such as location, weather or opening a mobile application and web service for tourist trip recommenda-
hours, we focus on two context factors that are highly relevant when tions around the world. It takes the user’s preferences, time and
recommending a sequence of POIs: time of the day and previously budget into account [16]. However, contextual information was not
visited point of interest. We conducted an online questionnaire to considered. In this work, we propose a novel, context-aware route
determine the influence of the context factors on the user’s decision recommendation algorithm that enhances our previous and related
of visiting a POI and the ratings of the POIs under these conditions. work. It incorporates various contextual information, including two
We integrated our approach into a web application recommending that are especially relevant for POI sequences.
context-aware tourist trips across the world. In a user study, we The rest of the paper is organized as follows. In Section 2 we
verified the results of our novel approach as well as the application’s present context factors that our CARS observes and explain how
usability. The study proves a high usability of our system and shows the respective contextual conditions ratings have been acquired.
that our context-aware approach outperforms a baseline algorithm. In Section 3, our novel context-aware route recommendation al-
gorithm is presented. Section 4 describes the application TourRec,
CCS CONCEPTS implementing our algorithm. In Section 5, the introduced CARS is
evaluated against the RS from our previous work. Section 6 and 7
• Information systems → Recommender systems;
list related work and conclude the paper.
KEYWORDS
2 EVALUATING CONTEXT FACTORS FOR
Context-Aware Recommender System, Tourist Trip Recommenda-
TOURIST TRIPS
tion, Point of Interest, Tourism
In this section we briefly discuss context and explain how context
is relevant for our RS. In order to integrate context-aware infor-
1 INTRODUCTION mation into a tourist trip RS, we first have to identify appropriate
context factors for the tourism domain and assess the influence
Recommender Systems (RSs) are commonly known as software sys-
of selected context factors. Also, the effects of each context factor
tems that suggest certain items to users in a predictive manner [1].
under several contextual conditions on the user’s route satisfaction
These systems facilitate the presentation of the available infor-
have to be investigated. Therefore, we conducted an online study to
mation typically by comparing user preferences to some reference
observe context factors specifically relevant for sequences of POIs
attributes. However, RSs can deliver more sophisticated suggestions
and derive information of similar pre-existing research for other
by adapting to the specific contextual situation of the recommenda-
context factors.
tion. Hence, context-aware recommender systems (CARSs) provide
different movie suggestions based on contextual factors like a user’s
mood or the time of the day. 2.1 Context in Tourism Recommender Systems
In the tourism domain, CARSs are also being developed and A common definition describes context as "any information that
researched. Several relevant contextual factors (e.g., weather) and can be used to characterize the situation of a [...] person, place,
their respective contextual conditions (e.g., raining) have already or object." [12]. Since we are only interested in information that
been identified. For example, a CARS reduces the relevance of is relevant in the tourism domain, we limit and categorize "any
outdoor activities while it is raining [3]. information" to physical context. Physical context can be described
Most tourism CARSs focus on suggesting single points of interest as the user’s immediate physical surroundings. This includes, but
(POIs). Only few solve route-planning problems for tourists who is not limited to, time of the day, light, weather, date, season, and
want to visit multiple interesting POIs consecutively. This prob- temperature [8]. This information could be retrieved by modern
lem statement is summarized as the Tourist Trip Design Problem smartphones with a GPS sensor or light sensor in conjunction with
RecTour 2017, August 27th, 2017, Como, Italy. 18 Copyright held by the author(s).
the current time. However, this does not work well for predictions.
For a route RS this data mostly has to be retrieved by external
services such as openweathermap1 . Furthermore, the system itself
has to be aware of the users physical context at each segment of
the route recommendation. For example, a recommended route
should contain less outdoor activities during the night or while it
is raining.
Another relevant type of context which we do not yet consider
in this work is social context. Social context can be described as
the user’s social group composition at the time of taking the rec-
ommendation. The user’s standing and role in the group is also an
important factor [2]. For example, a recommended route should
contain no nightlife activities such as going to a club when children
are part of the group.
2.2 Acquiring Context Relevance
We designed an online questionnaire to acquire quantitative mea-
sures of how selected contextual factors influence a user’s decision
of going to a POI. The following approach assesses the context
relevance and is based on a methodology presented by Baltrunas et
al. [3].
For the preliminary questionnaire a set of possible context factors
should be selected by domain experts. The questionnaire partici-
pants are asked to imagine certain conditions and whether a specific
context factor (e.g., weather) has a positive or negative influence
on the rating of a particular item [1, 2].
With this methodology we observe the context factors time of
the day and previously visited POI. These are especially crucial
for sequences of POIs and have not yet been observed in related
work. For other context factors like day of the week, weather and
temperature, which are also relevant for single POIs, we can rely
on [3]. Also opening hours is considered, which is a context factor
Figure 1: Online questionnaire to acquire context relevance.
that does not require a preliminary user study. The mentioned
context factors are later incorporated within the context-aware
recommendation algorithm.
Preliminary, twelve POIs in Munich, Germany have been se-
In addition to the measured relevance (U ) of a context factor, our
lected and mapped into six predefined categories: Arts and Museum,
context-aware approach (cf. subsection 3.3) is also dependent on
Food, Music Event, Nightlife Spot, Outdoors and Recreation and Shop-
ratings for POIs under different contextual conditions. The dataset
ping. It is assumed that categories represent all their corresponding
resulting from the previous conducted questionnaire can also be
POIs. Figure 1 shows how participants are asked whether they
utilized to determine such a rating. To make the responses quan-
would visit a certain POI just after they have been to a different
tifiable Yes, I don’t know and No are mapped to the values 2, 1, 0.
POI. Additionally, we asked the participants at which times they
A simple approach would be to use the mathematical expectation
would go to certain POIs.
value as a rating of a POI category for each contextual condition.
The aim of this study was to evaluate the influence of the selected
However, this does not respect the variation of the rating for a POI
context factors on their decisions to visit a category represented by
when a contextual condition holds or not. Informally speaking, if
a selected POI as well as the change of POI popularity precipitated
a POI category is typically very popular, except during night, the
by contextual conditions. In total, we received 324 responses by 27
expectation value would not reflect the real value of the contextual
participants.
condition night. For example, the expectation value for the cate-
The measured relevance (U ) for each context factor for all POI
gory food is 1.3. However, if one only considers ratings for food
categories are computed and listed in Table 1. It is normalized to
under the contextual condition night, the expectation value is 0.749.
an interval [0, 1]; where U = 0 means that the context factor does
Hence, we must present a comparison between the average ratings
not have any influence for this POI category. U is also relevant for
of POI and ratings of the same items assuming a certain contextual
the actual context-aware route recommendation algorithm and is
condition holds. We achieve this by dividing the expected value
there being utilized as a weighting factor for the context assessment
for a specific contextual condition by the expected value over all
in Equation 5.
ratings for this POI category. For the category Food during night
1 https://openweathermap.org/ time, the calculation is therefore: 0.749/1.3 = 0.58. All computed
RecTour 2017, August 27th, 2017, Como, Italy. 19 Copyright held by the author(s).
Table 1: Measured relevance of the contextual factors by POI categories.
Contextual Factor Arts and Museum Music Event Nightlife Spot Food Outdoors and Recreation Shopping
Previously visited POI 0.52 0.31 0.26 0.49 0.33 0.42
Time of the day 0.48 0.69 0.74 0.51 0.67 0.58
ratings for POI categories in different contextual conditions are 3.2 Baseline Algorithm
displayed in Table 2. We have improved a route recommendation algorithm in previous
work [16, 24] which is not context-aware.
3 A NOVEL APPROACH FOR The general idea is to combine as many single POIs as possible
to maximize the entertainment for the user while still respecting
CONTEXT-AWARE ROUTE
existing constraints like time. The process of generating a path
RECOMMENDATIONS from an origin to a destination while suggesting relevant POIs in
This section gives a detailed explanation of paradigms for incor- between can be generally divided into two subtasks:
porating context into RSs, the baseline path-finding algorithm and • The POI gathering and scoring, and
our approach for a context-aware path-finding algorithm. • executing a path-finding algorithm to find the optimal
route consisting of a subset of the gathered POIs.
3.1 Paradigms for Incorporating Context in For the POI gathering, we are using the Foursquare API2 to
Recommender Systems search for POIs in the general area between the source and the des-
This section describes how RS and CARS can be modeled and tination point. The gathered items are classified into six categories,
which paradigms exist to integrate context into a traditional, two- users can give preferences for these categories (see below). Our
dimensional (2D) RS. 2D RSs try to estimate the rating function R algorithm then computes a score for each item. The score is based
by considering only the User and Item dimensions [2]: on the total Foursquare rating and the number of votes, and also
the user preference for the corresponding category [24].
The algorithm to combine the POIs to a reasonable route is
R : U ser × Item → Ratinд (1) based on the well-known Dijkstra’s algorithm to find the shortest
path in a graph. Dijkstra’s algorithm is an iterative algorithm that
This rating function can be extended to model a three dimensional provides the shortest path from one particular starting node to all
(3D) recommendation [2]: other nodes in a graph with non-negative edge path costs. In our
scenario, the nodes are the places with the associated score and the
edges represent the distance between the places.
R : U ser × Item × Context → Ratinд (2) Prior to the graph spanning, each POI is assigned with a value
for the time to spend there. Then, a weighted graph is created
Adding context in the rating function increases the complexity using an feasible time value to walk the direct physical distance
of the recommendation algorithm. This is a non-trivial problem. between each vertex as edge weight. Prior to the comparison of
Adomavicius and Tuzhilin [2] identify three different paradigms a subpath with another path, it is checked whether the subpath
how to incorporate context in a traditional, 2D recommendation exceeds the specified timeframe. If the timeframe is exceeded, the
process: subpath will be rejected. If another valid subpath from the origin to
Contextual pre-filtering (or contextualization of recommendation the immediate vertex can be found prior to the current path, they
input): The context is utilized to construct a dataset only with the are compared against each other.
most relevant data. After that, a traditional RS can generate the To generate not the shortest, but the best path in the POI graph,
actual recommendations. we maximize the fraction entertainment/distance for each subpath.
Contextual post-filtering (or contextualization of recommendation The entertainment value is the accumulated sum of the scores of all
output): In contrast to contextual pre-filtering, the traditional RS items on the subpath. To adapt the number of items per category,
is executed on the entire data first and afterwards the context is the baseline algorithm uses the following formula Equation 3.
applied on the resulting set. This can be achieved by:
S = ppr e f ,poiCat eдor ies I nP at h × entertainment (3)
• Filtering out recommendations that are irrelevant (in a The idea is to maximize the product of entertainment and Pear-
given context), or son’s correlation coefficient between the user’s preferences and
• Adjusting the ranking of recommendations on the list (based the amount of POIs per category in the observed path. Pearson’s
on a given context). coefficient p gives values between -1 (indicating perfect negative
Contextual modeling (or contextualization of recommendation correlation) and +1 (perfect correlation), with 0 meaning no corre-
function): In this paradigm the 2D RS must be modified and directly lation exists between the datasets. Using the correlation coefficient
incorporate context into the recommendation algorithm. 2 https://developer.foursquare.com/start/search
RecTour 2017, August 27th, 2017, Como, Italy. 20 Copyright held by the author(s).
Table 2: Ratings for points of interest categories in different contextual conditions.
Contextual Condition\POI Category Arts and Museum Music Event Nightlife Spot Food Outdoors and Recreation Shopping
Previously visited POI (category)
Arts and Museum 1.36 1 1.16 1.43 1.25 0.72
Food 1.4 1.06 1.77 0.19 1.28 1.18
Music Event 0.04 1.32 1.69 1.1 0.6 0.11
Nightlife Spot 0 1.45 1.43 1.04 0.13 0
Outdoors and Recreation 1.63 1.42 0.86 1.37 0.76 1.52
Shopping 0.91 0.52 0.79 1.45 0.97 1.25
Time of the day
Morning 1.56 0.1 0.19 0.3 1.36 1.82
Midday 1.56 0.19 0.07 1.29 1.41 1.78
Afternoon 1.48 0.68 0.15 0.85 1.41 1.71
Evening 0.64 1.71 0.79 1.4 0.76 0.8
Night 0.42 1.55 2 0.58 1.07 0.11
aims to balance the amount of POIs in each category more ap- factor C. To calculate C, we use the values 0.49 and 0.51 for rele-
propriate in relation to the user’s preferences. The extension of vance of the context factors from Table 1 and the values 0.19 and
this adapted POI score with context-awareness is explained in the 0.85 for ratings for the category food in the current contextual
following subsection. condition from Table 2. After calculating C, the 2D score of 9.5 is
downscaled to 5.
3.3 Incorporating Context into the Baseline
Algorithm 0.19 × 0.49 + 0.85 × 0.51
C = 0.526 = (6)
The first challenge that arises is to determine how context-awareness 0.49 + 0.51
can be calculated for a route. Our collected dataset includes two
indications that can be utilizes for this task. First, ratings for cat- S = 5 = 9.5 × 0.526 (7)
egories in different contextual conditions as displayed in Table 2.
A rating rT C1...Ck indicates the evaluation for the POI category According to 3.1, one could assume that this algorithm adheres
T made in the context C1, ..., Ck and must be in the interval [0, 2]. to contextual post-filtering. However, the definition explicitly states,
Second, the relevance of contextual factors UC1...Ck of each con- that the traditional RS must be executed on the entire data first.
text C1, ..., Ck on a POI category T as displayed in Table 1. Like Since this is not the case, the paradigm contextual modeling was
illustrated in subsection 2.2 the measured relevance must be in an utilized to incorporate context into the baseline.
interval between [0, 1]. The full set of context factors considered in the current imple-
Given this data we can calculate a context-awareness factor C mentation and their values (contextual conditions) are:
with a simple weighted arithmetic mean:
• Previously visited POI (category): arts and museum, food,
Ík music event, nightlife spot, outdoors and recreation, shop-
i=1 UCi rT Ci ping
C= Ík (4)
i=1 UCi • Time of the day: morning (8am - 12pm), midday (12pm -
2pm), afternoon (2pm - 6pm), evening (6pm - 10pm), night
C can now be used to extend the 2D recommender baseline
(past 10pm)
algorithm by scaling the result of its comparison function:
• Day of the week: working day, weekend
• Weather: sunny, cloudy, clear sky, rainy, snowing
S = ppr ef ,poiCat eдor ies I nP at h × entertainment × C (5) • Temperature: hot, warm, cold
• Opening hours: open, closed
According to the given constraints, also C is in the interval [0, 2]
with 0 essentially nulling the score S while 2 would double its value. One benefit of the weighted arithmetic mean is the independence
To better explain the methodology we can illustrate it with an of the number of context factors. This list can easily be extended.
example comparison considering the two context factors time of Also the amount of context factors applied on POIs within a path
the day and previously visited POI : can vary. For example, designing context factors only known for a
It is 5 pm and the user has just been to a restaurant. The CARS specific POI category, e.g. nightlife spot, is not a concern. On the
should now calculate the score for another restaurant. In this sce- other hand, one disadvantage resulting from considering multiple
nario the 2D comparison algorithm would calculate a score of 9.5. context factors for C is that a supposedly drastic condition, e.g. the
The context-aware comparison algorithm (Equation 5) extends the POI is closed, can be balanced out by a different condition such as
2D comparison algorithm (Equation 3) by the context-awareness sunshine.
RecTour 2017, August 27th, 2017, Como, Italy. 21 Copyright held by the author(s).
4 THE TOURREC WEB APPLICATION 4.1.3 Data Tier. User feedback is being stored and handled
This section firstly presents the multi-tiered and service-oriented within this tier. As it might not seem urgent to dedicate an own tier
system architecture we introduced in (removed for review process). for this data, the need increases when user accounts are introduced,
One of the main goals was to facilitate the integration of additional for example.
data sources, path-finding algorithms and clients. We then present
the user interfaces of our web application TourRec 3 . 4.2 Interfaces
The web application is built with help of the javascript framework
4.1 Architecture Vuejs6 and the CSS framework Bulma7 . It is mainly structured into
The system is distributed across multiple physical devices offload- three segments search, recommendation, feedback.
ing application logic, computation and storage onto multiple web In the search segment, as seen in Figure 2, users can enter their
services running in the cloud. The applications multi-tier archi- preferences for all six predefined categories, the origin and destina-
tecture can be partitioned into the presentation tier, application tion as well as the time frame for the route. The input is validated
logic tier and data tier, while the application logic tier itself is also both client side as well as server side. For example, the maximum
partitioned. This architectural style decomposes an application into distance between origin and destination is 7 kilometers and the
loosely coupled services, functionality can thereby be easily reused. maximum time frame is 12 hours.
For example, clients do not have to re-implement the whole applica- The recommendation segment, as seen in Figure 3, is structured
tion logic but rely on the application logic tier. Other advantages are as follows. A map with the suggested POIs and a rendered walking
an improved modularity and the fact that each segment is easier to path on it is located on the left-hand side. On the top left-hand side
understand, develop and test. It also simplifies further development some contextual information that has been acquired by the system
since components can be assigned more precisely to experts in their and is relevant for the user for this situation is displayed. Finally an
respective fields. An iOS developer gets to develop the iOS app, the ordered list of POIs and their respective estimated time of arrival
android developer the android application and the data scientists and departure can be found beneath the contextual information.
can improve the algorithm. Also multiple people can work parallel The feedback segment gives the user a short introduction into
and independently on different components. the feedback process. In essence, it is a simple table with multiple
Additionally, the whole application has been containerized with statements which can each be answered with radio buttons on a five-
Docker 4 containers. Containers consist of a complete and isolated point Likert scale ranging from strongly disagree to strongly agree. In
run time environment: the software including all its dependen- section 5 we describe the feedback setup in greater detail. Note, that
cies, libraries and other binaries, and configuration files needed only one route per request gets displayed. The application logic tier
to run it. By containerizing our application, the differences in OS randomly selects either the baseline or context-aware path-finding
distributions and underlying infrastructure are abstracted away. algorithm. The algorithm name is stored in conjunction with the
This makes it fairly easy for possible new contributors to run the feedback a user provides.
whole project on their local environment. In addition, it enables
continuous delivery and deployment. 5 USER STUDY
In the following, the three tiers are presented. The research question of this paper is whether a context-aware
algorithm distinguishing between several contextual conditions
4.1.1 Presentation Tier. The web application being demonstrated can improve the trip recommendations generated by a baseline.
as well as the mobile application we have developed in our previous We conducted a user study to evaluate the performances of both
work is representative for the presentation tier. It is end user facing, algorithms. We spread the link to the TourRec application via mail
aimed to provide high user satisfaction and is responsible to handle and added questionnaires to the interfaces which the users were
user input and display computed information. It utilizes services asked to complete. The participants could access the application on
from the underlying application logic tier and external services like any device with an internet connection since it is publicly available.
Google Maps.
4.1.2 Application Logic Tier. The two main functionalities of 5.1 System Usability Score
this tier are the gathering of POIs and executing path-finding al- The System Usability Scale (SUS) is a questionnaire for measuring
gorithms. First, POIs are gathered from multiple external service how people perceive the usability of a computer system [6]. The
providers such as Foursquare5 and normalized. Then, this data is questionnaire is composed of ten usability statements with five
passed over to the path-finding algorithms that are executed af- possible response options on a scale ranging from strongly disagree
terwards. Each path-finding algorithm is extracted into its own to strongly agree. SUS is technology independent and can be used
dedicated microservice and communicates with the application for hardware, software, websites, mobile applications and more.
logic tier via HTTP in order to return a ordered list of POIs to visit. The key benefits of SUS are reliability, validity, no need a baseline
The path-finding microservices can be implemented using arbi- and the fact that it is an industry standard [7].
trary programming languages, databases, hardware and software 19 participants completed the SUS questionnaire after using
environment. TourRec, the average score was 84,167. With the help of Sauro’s
graph, the score approximately converts to a percentile rank of
3 https://tourrec.arubacao.com/
4 https://www.docker.com/ 6 https://vuejs.org/
5 https://foursquare.com/ 7 http://bulma.io/
RecTour 2017, August 27th, 2017, Como, Italy. 22 Copyright held by the author(s).
Figure 2: TourRec Search Interface
Figure 3: TourRec Response Interface
94% [21]. This means TourRec performs better than about 94% of After every recommendation, a questionnaire for this part of the
systems tested in terms of perceived usability. Everything about evaluation is presented to the user. It is composed of the following
90% can be interpreted as an A in school grades. However, since six statements and also with five possible response options on a
TourRec is accessible from virtually any device, the actual systems scale ranging from strongly disagree (1) to strongly agree (5):
usability varies for different screen sizes, operating systems and
browser vendors. (1) Overall, I am satisfied with the recommended tour
(2) The number of places in my route is well chosen
(3) The selection of different categories in the trip is satisfying
5.2 Algorithm Performance (4) Places are suggested at the right times during the tour
In addition to the SUS, we conducted an A/B test to measure the (5) The tour is feasible for a walking tourist
effect of the novel approach on the user’s route satisfaction com- (6) I consider taking this route myself
pared to the baseline system that does not exploit context at all.
Hence, only one tourist trip recommendation is displayed after ev- In total, 15 forms were completed for the baseline algorithm and 9
ery request. Apart from the route, users are not able to distinguish for the context-aware approach. Figure 4 illustrates the performance
between them. The recommendation screen shown in Figure 3 dis- of both algorithms for each of the six questions in subsection 5.2.
plays contextual information (e.g. the weather) whether or not the Our novel approach for context-aware route recommendation per-
context is actually considered. forms somewhat better in the overall satisfaction ( : 3, 67, σ : 1, 41)
RecTour 2017, August 27th, 2017, Como, Italy. 23 Copyright held by the author(s).
Figure 4: Algorithm Performance Results
and number of places ( : 3, 78, σ : 1, 39). In terms of feasible walk- 7 CONCLUSION
ing route and consider taking the route the context-aware algorithm In this paper, we presented the web application TourRec that al-
is rated slightly lower than the baseline. However, for these four lows context-aware tour recommendations in arbitrary locations
mentioned questions the actual difference is almost neglectable. The across the world. The system takes a starting point, a destination, a
biggest difference can be seen for diversity of categories and espe- timeframe and user preferences for six predefined categories into
cially right times where the context-aware algorithm outperforms account. It solves a variant of the OP applied to the tourism domain.
the baseline. In a preliminary questionnaire, the influence of the context factors
A Mann-Whitney U test shows that the difference in right times time of the day and previously visited POI were measured as well as
is significant for α = 0.01 while the other results we obtained are ratings for POI categories in different contextual conditions. The
not yet significant. We conclude that our novel approach leads to results are utilized within the context-aware algorithm.
improved recommendations but we have to conduct a larger user Context-awareness is incorporated into the baseline algorithm
study in the future to verify our results. as a scaling factor altering a POI’s score depending on the imme-
diate contextual condition. The focus and innovation of our work
is on recommending sequences of items. Therefore the influence
6 RELATED WORK of an already visited POI on the score of additional items based on
their category is important. Users may not be interested in another
Some applications solving the TTDP have been developed [9, 23]
restaurant if they just had lunch or dinner, for example. In a user
and numerous publications on this topic ranging from the avoidance
study we evaluated that the incorporation of context-awareness
of trafficked walking paths [20] to randomizing these [19] exist. In
leads to a slightly improved user satisfaction and a significant im-
this section, we highlight some important related work based on a
provement of recommending POIs at the right time.
general overview of [18].
One disadvantage of our approach is the possible equalizing of
Gionis et al.[14] and Lim [17] consolidated POIs into categories
two or more extreme contextual conditions due to the weighted
to enforce a predefined visiting order. In [17] the visiting order
arithmetic mean. A modified version could solely consider the con-
was defined by user preferences, and time and budget constraints.
text factor that has the largest negative or positive effect on a POI.
Instead of enforcing a specific order, Vansteenwegen et al. [22]
Using one single factor could potentially characterize the POI’s
recommended tours comprising POI categories that best match
score in a more user satisfying way. Another improvement can be
user preferences while adhering to these trip constraints.
achieved by increasing the number of predefined categories or in-
Tour recommendation can also be formulated as a Generalized
troduce subcategories. The current limitation of only six categories
Maximum Coverage Problem [4]. The objective here is to find an
has led to issues in the route recommendation scenario.
optimal set of POIs considering its rating and the user’s preferences.
We have designed our framework and architecture for easy ex-
Brilhante et al. then extended their algorithm later by incorporating
tension. We are working on a mobile solution [16] because context-
a variation of the Travelling Salesman Problem to find the shortest
aware route planning seems especially promising in a scenario
path within an optimal set of POIs [5]. To get from POI to POI
of mobile users with smartphones visiting a city. In this case, the
within a route, Kurashima et al. also consider different transport
recommended items should be adapted to the current position and
modes utilizing the Markov model depending on user preferences
other contextual conditions of the user. We also plan to imple-
and frequently traveled routes [15]. With patterns derived from taxi
ment more path-finding algorithms and compare them with the
GPS traces, Chen et al. developed a more context-aware solution
approach presented in this paper in larger user studies. Additional
considered traveling times based on different traffic conditions [10].
future work includes recommending for groups of visitors.
RecTour 2017, August 27th, 2017, Como, Italy. 24 Copyright held by the author(s).
REFERENCES Tourist Trip Design Problems. Journal of Heuristics 20, 3 (June 2014), 291–328.
[1] Gediminas Adomavicius and Alexander Tuzhilin. 2005. Toward the Next Gener- DOI:https://doi.org/10.1007/s10732-014-9242-5
ation of Recommender Systems: A Survey of the State-of-the-Art and Possible [14] Aristides Gionis, Theodoros Lappas, Konstantinos Pelechrinis, and Evimaria
Extensions. IEEE Trans. on Knowl. and Data Eng. 17, 6 (June 2005), 734–749. DOI: Terzi. 2014. Customized Tour Recommendations in Urban Areas. In Proceedings
https://doi.org/10.1109/TKDE.2005.99 of the 7th ACM International Conference on Web Search and Data Mining (WSDM
[2] Gediminas Adomavicius and Alexander Tuzhilin. 2008. Context-aware Rec- ’14). ACM, New York, NY, USA, 313–322. DOI:https://doi.org/10.1145/2556195.
ommender Systems. In Proceedings of the 2008 ACM Conference on Recom- 2559893
mender Systems (RecSys ’08). ACM, New York, NY, USA, 335–336. DOI:https: [15] Takeshi Kurashima, Tomoharu Iwata, Go Irie, and Ko Fujimura. 2010. Travel
//doi.org/10.1145/1454008.1454068 Route Recommendation Using Geotags in Photo Sharing Sites. In Proceedings of
[3] Linas Baltrunas, Bernd Ludwig, Stefan Peer, and Francesco Ricci. 2012. Con- the 19th ACM International Conference on Information and Knowledge Management
text Relevance Assessment and Exploitation in Mobile Recommender Sys- (CIKM ’10). ACM, New York, NY, USA, 579–588. DOI:https://doi.org/10.1145/
tems. Personal Ubiquitous Comput. 16, 5 (June 2012), 507–526. DOI:https: 1871437.1871513
//doi.org/10.1007/s00779-011-0417-x [16] Christopher Laß, Wolfgang Wörndl, and Daniel Herzog. 2016. A Multi-Tier
[4] Igo Brilhante, Jose Antonio Macedo, Franco Maria Nardini, Raffaele Perego, Web Service and Mobile Client for City Trip Recommendations. In The 8th
and Chiara Renso. 2013. Where Shall We Go Today?: Planning Touristic Tours EAI International Conference on Mobile Computing, Applications and Services
with Tripbuilder. In Proceedings of the 22Nd ACM International Conference on (MobiCASE). ACM. DOI:https://doi.org/10.4108/eai.30-11-2016.2267194
Information & Knowledge Management (CIKM ’13). ACM, New York, NY, USA, [17] Kwan Hui Lim. 2015. Recommending Tours and Places-of-Interest Based on
757–762. DOI:https://doi.org/10.1145/2505515.2505643 User Interests from Geo-tagged Photos. In Proceedings of the 2015 ACM SIGMOD
[5] Igo Ramalho Brilhante, Jose Antonio Macedo, Franco Maria Nardini, Raffaele on PhD Symposium (SIGMOD ’15 PhD Symposium). ACM, New York, NY, USA,
Perego, and Chiara Renso. 2015. On planning sightseeing tours with TripBuilder. 33–38. DOI:https://doi.org/10.1145/2744680.2744693
Information Processing and Management 51, 2 (2015), 1 – 15. DOI:https://doi.org/ [18] Kwan Hui Lim, Jeffrey Chan, Christopher Leckie, and Shanika Karunasekera.
10.1016/j.ipm.2014.10.003 2015. Personalized Tour Recommendation Based on User Interests and Points
[6] John Brooke. 1996. SUS-A quick and dirty usability scale. Usability evaluation in of Interest Visit Durations. In Proceedings of the 24th International Conference
industry 189, 194 (1996), 4–7. on Artificial Intelligence (IJCAI’15). AAAI Press, 1778–1784. http://dl.acm.org/
[7] John Brooke. 2013. SUS: A Retrospective. J. Usability Studies 8, 2 (Feb. 2013), citation.cfm?id=2832415.2832496
29–40. http://dl.acm.org/citation.cfm?id=2817912.2817913 [19] Claudio Lucchese, Raffaele Perego, Fabrizio Silvestri, Hossein Vahabi, and
[8] P. J. Brown, J. D. Bovey, and Xian Chen. 1997. Context-aware applications: from Rossano Venturini. 2012. How Random Walks Can Help Tourism. In Pro-
the laboratory to the marketplace. IEEE Personal Communications 4, 5 (Oct 1997), ceedings of the 34th European Conference on Advances in Information Retrieval
58–64. DOI:https://doi.org/10.1109/98.626984 (ECIR’12). Springer-Verlag, Berlin, Heidelberg, 195–206. DOI:https://doi.org/10.
[9] Luis Castillo, Eva Armengol, Eva OnaindÃŋa, Laura SebastiÃą, JesÞs GonzÃąlez- 1007/978-3-642-28997-2_17
[20] Daniele Quercia, Rossano Schifanella, and Luca Maria Aiello. 2014. The Shortest
Boticario, Antonio RodrÃŋguez, Susana FernÃąndez, Juan D. Arias, and Daniel
Path to Happiness: Recommending Beautiful, Quiet, and Happy Routes in the
Borrajo. 2008. samap: An user-oriented adaptive system for planning tourist
City. In Proceedings of the 25th ACM Conference on Hypertext and Social Media (HT
visits. Expert Systems with Applications 34, 2 (2008), 1318 – 1332. DOI:https:
’14). ACM, New York, NY, USA, 116–125. DOI:https://doi.org/10.1145/2631775.
//doi.org/10.1016/j.eswa.2006.12.029
2631799
[10] C. Chen, D. Zhang, B. Guo, X. Ma, G. Pan, and Z. Wu. 2015. TripPlanner:
[21] J. Sauro. 2011. A Practical Guide to the System Usability Scale: Background,
Personalized Trip Planning Leveraging Heterogeneous Crowdsourced Digital
Benchmarks & Best Practices. CreateSpace Independent Publishing Platform.
Footprints. IEEE Transactions on Intelligent Transportation Systems 16, 3 (June
https://books.google.co.uk/books?id=BL0kKQEACAAJ
2015), 1259–1273. DOI:https://doi.org/10.1109/TITS.2014.2357835
[22] Pieter Vansteenwegen, Wouter Souffriau, Greet Vanden Berghe, and Dirk Van
[11] Munmun De Choudhury, Moran Feldman, Sihem Amer-Yahia, Nadav Golbandi,
Oudheusden. 2011. The City Trip Planner. Expert Syst. Appl. 38, 6 (June 2011),
Ronny Lempel, and Cong Yu. 2010. Automatic Construction of Travel Itineraries
6540–6546. DOI:https://doi.org/10.1016/j.eswa.2010.11.085
Using Social Breadcrumbs. In Proceedings of the 21st ACM Conference on Hypertext
[23] Pieter Vansteenwegen and Dirk Van Oudheusden. 2007. The Mobile Tourist
and Hypermedia (HT ’10). ACM, New York, NY, USA, 35–44. DOI:https://doi.
Guide: An OR Opportunity. OR Insight 20, 3 (2007), 21–27. DOI:https://doi.org/
org/10.1145/1810617.1810626
10.1057/ori.2007.17
[12] Anind K. Dey. 2001. Understanding and Using Context. Personal Ubiquitous
[24] Wolfgang Wörndl, Alexander Hefele, and Daniel Herzog. 2017. Recommending a
Comput. 5, 1 (Jan. 2001), 4–7. DOI:https://doi.org/10.1007/s007790170019
sequence of interesting places for tourist trips. Information Technology & Tourism
[13] Damianos Gavalas, Charalampos Konstantopoulos, Konstantinos Mastakas, and
17, 1 (2017), 31–54. DOI:https://doi.org/10.1007/s40558-017-0076-5
Grammati Pantziou. 2014. A Survey on Algorithmic Approaches for Solving
RecTour 2017, August 27th, 2017, Como, Italy. 25 Copyright held by the author(s).