=Paper=
{{Paper
|id=Vol-2855/main_short_1
|storemode=property
|title=Generating Multi-Day Round Trip Itineraries for Tourists
|pdfUrl=https://ceur-ws.org/Vol-2855/main_short_1.pdf
|volume=Vol-2855
|authors=Elif Erbil,Wolfgang Worndl
|dblpUrl=https://dblp.org/rec/conf/wsdm/ErbilW21
}}
==Generating Multi-Day Round Trip Itineraries for Tourists==
https://ceur-ws.org/Vol-2855/main_short_1.pdf
ACM WSDM WebTour 2021, March 12th, 2021 Jerusalem, Israel 1
Generating Multi-Day Round Trip Itineraries for Tourists
Elif Erbil Wolfgang Wörndl
elif.erbil@tum.de woerndl@in.tum.de
Technical University of Munich Technical University of Munich
Garching bei München, Germany Garching bei München, Germany
ABSTRACT for recommending routes with sequences of POIs for multi-day
Recommender systems facilitate the decision making process for trips. Our approach uses the hotel location of the traveller as a
users by selecting and ranking items according to users’ interests. start and end point for the generated routes, and groups the most
Travellers can highly benefit from recommender systems due to interesting POIs for each day of the trip to create balanced routes
vast amount of information available regarding places to visit and with optimized walking distances.
many different factors that affect the travel plans. Travel recom- The paper is structured as follows: Section 2 discusses related
mender systems tackle both the issues of recommending relevant work in travel recommender systems and route creation algorithms,
places and also sequencing them in a feasible order. However there Section 3 explains the methodology followed to create multi-day,
are many different constraints that affect the recommendations for round trip itineraries using recommender systems and different
travellers such as travel dates, weather and companions, which algorithms used for comparison, Section 4 discusses results from
makes the recommendation system more complex. In this paper, an offline study for the mentioned algorithms according to certain
we present a novel approach to create multi-day trips that start and criteria, whereas Section 5 extends these results with a user study
end at the accommodation of the user. We apply different clustering to evaluate the success of each algorithm. Section 6 concludes the
algorithms to tackle the issue of creating multi-day trips with bal- paper with the possible future work and final remarks.
anced itineraries and conduct a user study to understand how our
approach performed against baseline methods. Our results show
that our algorithm performs better than other selected methods
to recommend interesting points-of-interests to users and create
appealing itineraries.
2 RELATED WORK
CCS CONCEPTS Different approaches for TTDPs have been proposed to create feasi-
• Information systems → Users and interactive retrieval. ble sequences of POIs. Gavalas et al. [8] proposes different algorith-
mic approaches for different cases of TTDPs. Similar to Souffriau
KEYWORDS et al. [14], by modelling TTDPs as a variant of the Orienteering
Problem (OP), the authors extend these to different cases such as
recommender system, tourist trip design problem, points-of-interests,
single day tours using the Travelling Salesman Problem with Prof-
clustering, user study
its. The authors propose algorithmic approaches to solve further
constraints by using different variants of the OP. For example, the
1 INTRODUCTION system models travelling multiple days as Team OP (TOP) and for-
With the growing amount of data collected and presented to the user, mulates working hours of POIs as OP with Time Windows (OPTW).
finding relevant information gets harder every day. One approach to Another research by Garcia et al. [6] focuses on recommending
solve this problem and filter through the vast amount of information trips to users while bringing together different constraints as an
is using recommender systems. Recommender systems are tools instance of the Time Dependent Team Orienteering Problem with
and techniques to support users in finding products, services or Time Windows (TDTOPTW) and solves the problem in real time
information that are relevant according to the users’ query, profile using heuristics.
and context, and present them in an efficient way. One of the areas Kurata and Hara [11] use the Selective Travelling Salesman Prob-
that can highly benefit from recommender systems is the travel lem as a model to create single day route recommendations to users.
industry [13], e.g. in recommending travel plans or routes composed Research by Wörndl et al. [15] uses a variant of Dijkstra’s shortest
of multiple points-of-interests (POIs). This problem is formulated path problem to recommend single day walking tours for users
as the Tourist Trip Design Problem (TTDP) [8], which aims at with the given start and end points. The mobile single day route
combining interesting POIs along a route to maximize the value for recommender by Anacleto et al. [2] uses a lightweight solution
travellers. based on decision trees for a route creation process on mobile de-
TTDP addresses this issue in two parts: ranking and selection vices. Deitch et al. [3] models the problem of creating itineraries
of POIs that might interest the user, and creating a feasible route with attractive routes between POIs to be visited as bus-touring
using the selected POIs [16]. Recommender systems can use routing problem (BTP). The BTP is similar to the orienteering tour problem
algorithms to create routes that include highly rated POIs while and extends the OP to have the same start and end points, therefore
minimizing the distances in a suitable way for the user to follow. allowing round trips. Gavalas et al. [7] tackles the same issue by
The route creation process gets more complex in a scenario with modelling the problem as mixed team orienteering problem with
multiple travel days. In this paper, we investigate a novel approach time windows and solving it using iterated local search heuristics.
Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
ACM WSDM WebTour 2021, March 12th, 2021 Jerusalem, Israel 2
Elif Erbil and Wolfgang Wörndl
3 CREATING MULTI-DAY ITINERARIES
In this section we discuss our novel approach for creating multi-
day round trip itineraries. We propose to first cluster POIs that are
within the walking distance to the hotel for different days and then
run a round trip routing algorithm to create the itineraries for each
day. We compare our approach against a simple baseline algorithm.
3.1 Creating Round Trips
In order to create round trips starting and ending at the hotel, the
model is defined as a travelling salesmen problem with profits model
[4]. The traveller has a start location and each POI that is added
to the route brings a certain amount of profit, whereas walking
between POIs and visiting them has a cost. So the profit must be
maximized in order to have the most efficient route with high rated
POIs. Since TSP with Profits is NP-hard and it is computationally
inefficient to find an exact solution, we use the algorithm proposed
by [12] to approximate a solution for the given problem. [12] states
that the problem can be solved by selecting the most profitable POI Figure 1: Initial Route Created by the Greedy Algorithm
in a greedy fashion by selecting the POI with highest 𝑠𝑐𝑜𝑟𝑒/𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒
ratio towards the two ends of the route and adding them accordingly.
We have modified the following algorithm in three ways:
• changing the profit criteria to increase the weight of the
profit of the score to ensure that the algorithm will prefer
the higher rated POIs instead of lower rated but closer ones
• adding the visiting times of each POI to the cost of the POI
as well as the time needed to travel to reach to the POI
• re-adjusting the sequence of POIs using the 2-opt algorithm
for further optimization [5]
The 2-opt algorithm is used to optimize the solution after each
POI is selected so that it creates a shorter route that includes the
same POIs, if there is a better route sequence that can be created
by eliminating crossing routes (see Figures 1 and 2 for an example).
The idea behind 2-opt is to switch places of POIs if the total distance
of the new alignment is less than the original one. For each POI,
POIs i and j are swapped if:
𝑑𝑖𝑠𝑡 (𝑖, 𝑖 + 1) + 𝑑𝑖𝑠𝑡 ( 𝑗, 𝑗 + 1) > 𝑑𝑖𝑠𝑡 (𝑖, 𝑗) + 𝑑𝑖𝑠𝑡 (𝑖 + 1, 𝑗 + 1)
Figure 2: Routes in Figure 1 Optimized by 2-opt Algorithm
3.2 Creating Multi-Day Trips
In order to create multi-day trips, the POIs must be distributed
to get a baseline since this way the algorithm is able to take all
among each day so that the routes created will have unique POIs
POIs into account and therefore create itineraries that have closer
that are selected to create routes with most profit. Therefore while
POIs together and allows to create itineraries with multiple tours
creating POI lists for each day we need to take the following into
to the same location if the high rated POIs are clustered in the same
account:
location within the city. Although the itineraries are expected to
• each day must have enough POIs to fill the time limit that is include many POIs and high ratings, there are two disadvantages
set by the traveller, of the baseline approach that is expected:
• each day must have POIs that are rated highly by the trav-
• The baseline algorithm takes into account all of the POIs
eller,
rated by the user and if a POI with a lower score is much
• the POIs must be clustered in a way that closer POIs recom-
closer than a distant high rated POI, a selection of lower
mended must be in the same day.
rated POIs might be favored over higher rated ones.
3.2.1 Baseline algorithm. While creating multi-day trips, the base- • The baseline algorithm needs to go over all of the POIs to
line algorithm that is selected to create multiple round trips is decide the highest profit ones, therefore running the round
running the algorithm specified for creating round trips n times, n trip algorithm n times will have a longer runtime which is not
being the number of days. For each run, the POIs that are used in efficient for the mobile users with lower computation power
the previous days will be removed from the POI list. This allows and limited battery. Consequently, the baseline algorithm
Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
ACM WSDM WebTour 2021, March 12th, 2021 Jerusalem, Israel 3
Generating Multi-Day Round Trip Itineraries for Tourists
will be compared to certain clustering algorithms in terms
of both runtime and itinerary success criteria.
Figure 4: Sample 4 Day Trip with Random Data Points using
k-means Approach
Figure 3: Sample 4 Day Trip with Random Data Points using
cluster until the time limit is reached at that cluster or the number
Baseline Approach
of clusters created is equal to the number of days (which might be
the case if there are not enough POIs or the number of days are
3.2.2 k-means Clustering. Secondly, we used the k-means algo- too high). Since the user is expected to spend a certain time at each
rithm to cluster these items. k-means allows to cluster POIs that are POI, it can be assumed that the POIs in the cluster will be sufficient
closer to each other by starting with k random clusters (where k is for a day trip, as there is also the travelling time between these
the number of days for the trip) and then iteratively reassigning POIs that must be accounted for. By this way, each cluster will have
each POI to the cluster that is the closest one [10]. By this approach, POIs that are close to each other and each day will have roughly
the round trip algorithm will be run on each of the clusters cre- the same number of POIs assigned. This idea is followed by the
ated which is expected to decrease the runtime. However, k-means assumption that in most of the cities, POIs are clustered densely
algorithm might not behave optimally in the following cases: at the city centers and it might not be feasible to visit all the POIs
• k-means clusters POIs according to their distances, therefore in the center in one day. Therefore, this approach allows to divide
for the cities where most of the POIs are gathered in a close these POIs in multiple days so that the traveller would not miss
perimeter and a few are outside this perimeter, the clusters them if they are preferred over POIs that are outside the city, which
might not have equal distribution of POIs for each day. would be in a different cluster due to their distance.
• Higher rated POIs that might be preferred over lower rated The second aim is to have higher rated POIs added to each cluster.
ones might be bundled into one cluster, therefore the ratings In order to create balanced itineraries for each day, we perform
of itineraries might be lower. pre-processing on the POI data to select the higher rated POIs
over the ones that are closer to the clusters. In order to do the
3.2.3 Time-Limit Approach. In order to eliminate the problems pre-processing, the recommender systems first predict ratings for
that might be encountered with the above approaches, we have each POI by matching the user profile and information collected
developed a novel approach called time-limit approach. Our novel about POIs. One way the recommender system can predict ratings
approach uses agglomerative clustering as a starting point since it for the user is proposed by [15], which assigns venues to different
similarly follows a bottom-up fashion to bring closer POIs together categories and predicts a score by using the ratings on Foursquare.
in the same clusters [1]. Agglomerative clustering yields a similar The recommender system then eliminates POIs that have a low
solution as k-means since it creates clusters with a single POI and score according to other users’ ratings as well as traveller’s category
merges these clusters that are closest to each other together until preferences. After the ratings are obtained, each POI is assigned
there is expected number of clusters [9]. However, these clustering to three groups according to their ratings: must-visit, can-visit
algorithms create a fixed number of clusters rather than dividing or don’t-visit. Since each user might have a different rating scale,
elements in a balanced way, therefore the number of items in each instead of assigning fixed ratings to separate POIs into these groups
cluster may not be equal. In our system, we aim to have balanced a different approach is followed. Must-visits are POIs that have
clusters so that the users can visit more POIs each day. Therefore, ratings one standard deviation above the mean of the predicted
instead of fixing the number of clusters, our approach has a con- ratings of the user for all POIs. These POIs are added to the clusters
straint on each cluster where the sum of the visiting durations of first. Don’t-visits are POIs that have ratings one standard deviation
the POIs within that cluster does not exceed the visiting duration below the mean rating of the user for all POIs and these POIs
assigned to a day. So the POIs closer to each other are added to a are discarded by the system before the clustering algorithm. This
Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
ACM WSDM WebTour 2021, March 12th, 2021 Jerusalem, Israel 4
Elif Erbil and Wolfgang Wörndl
ensures lower rated POIs that are close to others are not preferred that the ratings of the POIs of each proposed route are as high as
over higher rated ones. The rest of the POIs are marked as can-visits, possible.
which are added to the itinerary if there is time left after adding
all the must-visits. Therefore, the lowest rated POIs are eliminated
and there is sufficient amount of must-visits that can be added to
the itineraries.
Figure 6: Mean of Average Rating of Routes for Different
Number of Days per Route
Figure 5: Sample 4 Day Trip with Random Data Points using
Time Limit Approach The results presented in Figure 6 show that the time-limit ap-
proach has routes with the highest number of average ratings per
day compared to the baseline and k-means approaches. This is due
After grouping the POIs according to their relevance to the to the fact that the POIs are first grouped to create clusters that
user, the clustering algorithm is first run on must-visit POIs to maximize user satisfaction by pre-filtering for highly rated POIs,
create clusters. Then the algorithm is re-run using the clusters which is absent in the other approaches. Therefore by the time-limit
created in first step and the can-visit POIs to ensure that each approach, POIs that have lower scores but which were closer to
cluster have enough POIs to create routes while having higher other POIs were favored over the ones that are higher rated but
rated ones preferred over others while increasing the diversity of farther away during the route creation process to ensure more POIs
the tour and keep the distances between them as low as possible. can be added. The results are consistent with increasing number of
After creating the clusters, the top n clusters with highest average days. The means are decreasing with more days, which is expected
rating is selected to create round trips. because more POIs with lower ratings have to be included in the
Figures 3,4 and 5 show 4-day itineraries created with the baseline, routes.
k-means and time-limit approaches respectively. It can be seen that
time-limit approach has more clusters and less data points assigned 4.2 Coefficient of Variation of Average Ratings
to clusters as low rated POIs are discarded beforehand.
For each number of days, we calculate the coefficient of variation
of the average ratings. For a trip with n days, where each day i has
4 OFFLINE EVALUATION
𝑘𝑖 POIs selected, the coefficient of variation CV of average ratings
In order to evaluate the performance of each approach, we first is calculated as:
compare the algorithms in an offline analysis. To do so, we generate 𝑘𝑖
routes with the three algorithms using a dataset with 160 POIs q Í Í
1 𝑛 𝑟𝑎𝑡𝑖𝑛𝑔𝑝
and hotels in Istanbul. The hotels and POIs were selected from the 𝑛 ¯ 2
𝑖=1 (𝑥¯𝑖 − 𝑥) 𝑝=1
𝐶𝑉 = , 𝑤ℎ𝑒𝑟𝑒 𝑥¯𝑖 =
attractions that are listed on Tripadvisor. We use different metrics 𝑥¯ 𝑘𝑖
to better understand the characteristics of the routes created by This measure ensures that each day will have a similar distribu-
each algorithm: mean of average ratings, coefficient of variation of tion of highly rated POIs. The coefficient of variation is selected
average rating and runtime of the algorithms. instead of standard deviation so that the amount of change can be
compared across different algorithms with different means.
4.1 Mean of Average Ratings The coefficient of variation of average ratings of POIs presented
For each number of days, we calculate the mean of the average in Figure 7 shows that for most of the cases, the time-limit approach
ratings. For a trip with n days, where each day i has 𝑘𝑖 POIs selected, had a lower variance in average rating than the baseline and k-
we sum up the ratings of the POIs of a day, divide the number by means approaches. This can be interpreted as that the average
𝑘𝑖 and compute the overall mean. Mean of average ratings ensures ratings of each day in a route is similar and POIs that are more
Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
ACM WSDM WebTour 2021, March 12th, 2021 Jerusalem, Israel 5
Generating Multi-Day Round Trip Itineraries for Tourists
preferred by the user are distributed among different days since each As stated before, one of the main drawbacks of the baseline al-
cluster is limited with an upper limit. Therefore higher rated POIs gorithm is its runtime, as it executes the greedy approach on the
are added to different clusters when the upper limit for a cluster whole dataset for each day of the route. The results in Figure 8
is reached and a more even distribution is obtained. However, the show that as the number of days increases, the number of compar-
coefficient of variation is relatively low for all samples. isons and therefore the runtime increases linearly for the baseline
algorithm. However, the k-means and time-limit approaches use
clustering first and then run the greedy algorithm on these smaller
clusters. The time-limit approach constrains the size of the clusters,
therefore the input for the route creation algorithm was limited to
a smaller number of POIs for each day. For the k-means algorithm,
the runtime decreases as the number of days increases, because
the number of POIs in each cluster decreases with more days and
therefore the route creation algorithm takes less time overall.
Figure 7: Coefficient of Variation of Average Rating of
Routes for Different Number of Days per Route
4.3 Runtime
We also investigated the runtime of each clustering algorithm as
well as the route creation algorithm. We used an Intel Core i7 (3.1
GHz) processor for the performance analysis. In order to get reliable
results, the runtime was calculated as the average of 1000 runs.
Figure 9: Sample Two-Day Trip Created with Istanbul
Dataset from the Questionnaire
Figure 8: Runtime of Each Algorithm for Different Number
of Days per Route
Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
ACM WSDM WebTour 2021, March 12th, 2021 Jerusalem, Israel 6
Elif Erbil and Wolfgang Wörndl
5 USER STUDY
The offline results give an insight about how each algorithm be-
haves to create routes with preferable POIs. In order to evaluate
the different algorithms from a user’s perspective, we conducted
a preliminary user study with the Istanbul dataset. For mitigat-
ing the effect of different user preferences and limitations of the
dataset, each user is presented with predefined routes where each
POI category was given a static rating. POI categories are created
similarly to the categories proposed in [15]. POIs that are in the
top 20 recommended attractions in Tripadvisor for Istanbul were
given higher ratings than all other POIs. A sample route created
using the algorithms can been seen in Figure 9.
To evaluate the behavior of different algorithms with a different
number of days, each algorithm is used to create a two-day and
three-day trip from selected hotels. To decrease the effect of the
hotel locations on efficiency of routes, we selected two different Figure 10: Questionnaire Results for Two-day Trips
hotels in different locations. In total, each user was presented with
12 routes. For each route, the user was asked to answer the following
questions by rating each option ranging on a Likert scale ranging
from strongly disagree (1) to strongly agree (5):
(1) There is an adequate number of points-of-interests for each
day
(2) Points-of-interests are distributed evenly among each day
(3) An adequate amount of time is spent on visiting points-of-
interests rather than travelling between them
(4) The points-of-interests recommended are interesting
(5) I’m satisfied with the overall recommendation
In total, the questionnaire was filled out by 22 people (50% female,
50% male). The user study participants were identified by sharing
the questionnaire on social media and among acquaintances. In
order to understand the success of the routes, participants were
chosen among people who have lived in Istanbul and who thus have
knowledge about the city and the presented POIs. The distribution
of the ages of the participants were 0-20 (13.6%), 21-30 (59%), 31-40
Figure 11: Questionnaire Results for Three-day Trips
(4.5%), 41-50 (4.5%), 51-60 (9%) and 61-70 (4.5%). Figures 10 and 11
show the results of each algorithm for each question for two-day
and three-day trips respectively. For two-day trips the time-limit
approach performed better than both other algorithms in over-
all recommendations (ø: 3.70, 𝜎: 1.06), POIs are distributed evenly 6 CONCLUSION AND FUTURE WORK
among days (ø: 3.80, 𝜎: 1.16) and recommended POIs are interest- In this research, we propose a new route creation process for multi-
ing (ø: 4.32, 𝜎: 0.79). For three-day trips, the time-limit approach day round trips, starting and ending at the accommodation of the
performed better than other algorithms in all five categories. user. Our approach tags each POI according to their importance
In order to understand if the time-limit algorithm is significantly for the user, then clusters the POIs to days by limiting each cluster
better than other algorithms, we performed a one-way ANOVA with the maximum amount of time the user spends on travelling
test. The results of the ANOVA test with the confidence interval per day, starting from the POIs with the highest importance to the
of 𝛼 = 0.05 shows that for two-day trips the POIs recommended user. By doing so, the top clusters that include high importance
are significantly more interesting for the time-limit approach than POIs that are closer to each other are selected. One assumption
both the baseline and k-means approaches, whereas the rest didn’t of the algorithms is that there are enough places highly rated by
yield results significant enough for comparison. For three-day trips the user to create routes for the given number of days for the
the ANOVA results shows that both the POIs are significantly more route. By creating clusters and running the round trip algorithm,
interesting and also the overall recommendations are significantly which is a greedy approach that gives an approximate solution
more appealing for time-limit approach than both the baseline for the orienteering problem, both the runtime is decreased and
and k-means approaches. Also for the three-day trips, the POIs POIs with higher importance are prioritized over POIs closer to the
are distributed more evenly among each day by the time-limit accommodation. The proposed solution allows parallelization of the
approach compared to k-means approach but not significant enough route creation process, which can further increase the performance
to compare to the baseline algorithm. of the algorithm.
Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
ACM WSDM WebTour 2021, March 12th, 2021 Jerusalem, Israel 7
Generating Multi-Day Round Trip Itineraries for Tourists
According to the offline results and the user study, the selection [16] Wolfgang Wörndl. 2016. Solving Tourist Trip Design Problems from a User’s
of recommended POIs for the route and the overall recommenda- Perspective. In Mensch und Computer 2016 – Workshopband, Benjamin Weyers
and Anke Dittmar (Eds.). Gesellschaft für Informatik e.V., Aachen. https://doi.
tion quality of the proposed time-limit approach is better than the org/10.18420/muc2016-ws05-0003
recommendations given by the baseline algorithm and k-means
clustering. However, the current approach follows a very simple
method in terms of rating and ranking POIs and was not personal-
ized to user interests, which is also a factor that needs to be taken
into account. For the future work, the route creation algorithm can
be combined with a recommendation algorithm to create personal-
ized scores for the POIs. By this way the algorithms can be tested
by a more extensive user study with personalized routes.
Currently, the algorithm only takes user ratings into account
when clustering POIs and creating routes. In the next step, the
working hours of the POIs can be incorporated into the routing
algorithm as well the clustering algorithm, so the user can visit
these POIs at the suggested times. Also this could be used to further
customize trips such as adding restaurants to routes for lunch time
or creating certain breaks for the users between POIs. Another
possible addition is using the context information to rank POIs or
create different routes according to different contextual factors.
REFERENCES
[1] Marcel R. Ackermann, Johannes Blömer, Daniel Kuntze, and Christian Sohler.
2012. Analysis of Agglomerative Clustering. Algorithmica 69, 1 (Dec 2012),
184–215. https://doi.org/10.1007/s00453-012-9717-4
[2] Ricardo Anacleto, Lino Figueiredo, Ana Almeida, and Paulo Novais. 2014. Mobile
Application to Provide Personalized Sightseeing Tours. Journal of Network and
Computer Applications 41 (May 2014), 56–64. https://doi.org/10.1016/j.jnca.2013.
10.005
[3] Ray Deitch and Shaul P. Ladany. 2000. The one-period bus touring problem:
Solved by an effective heuristic for the orienteering tour problem and improve-
ment algorithm. European Journal of Operational Research 127, 1 (2000), 69–77.
https://doi.org/10.1016/S0377-2217(99)00323-9
[4] Dominique Feillet, Pierre DEJAX, and Michel Gendreau. 2005. Traveling Salesman
Problems With Profits. Transportation Science 39 (05 2005), 188–205. https:
//doi.org/10.1287/trsc.1030.0079
[5] Croes G. A. 1958. A Method for Solving Traveling-Salesman Problems. Operations
Research 6, 6 (1958), 791. https://doi.org/10.1287/opre.6.6.791
[6] Ander Garcia, Olatz Arbelaitz, Maria Teresa Linaza, Pieter Vansteenwegen, and
Wouter Souffriau. 2010. Personalized Tourist Route Generation. In Current Trends
in Web Engineering, Florian Daniel and Federico Michele Facca (Eds.). Springer
Berlin Heidelberg, Berlin, Heidelberg, 486–497. https://doi.org/10.1007/978-3-
642-16985-4_47
[7] Damianos Gavalas, Vlasios Kasapakis, Charalampos Konstantopoulos, Grammati
Pantziou, and Nikolaos Vathis. 2016. Scenic route planning for tourists. Personal
and Ubiquitous Computing (10 2016). https://doi.org/10.1007/s00779-016-0971-3
[8] Damianos Gavalas, Charalampos Konstantopoulos, Konstantinos Mastakas, and
Grammati Pantziou. 2014. A Survey on Algorithmic Approaches for Solving
Tourist Trip Design Problems. Journal of Heuristics 291, 20 (2014). https://doi.
org/10.1007/s10732-014-9242-5
[9] Anil Jain and Richard Dubes. 1988. Algorithms for Clustering Data. Vol. 32.
https://doi.org/10.2307/1268876
[10] Xin Jin and Jiawei Han. 2010. K-Means Clustering. Springer US, Boston, MA,
563–564. https://doi.org/10.1007/978-0-387-30164-8_425
[11] Yohei Kurata and Tatsunori Hara. 2013. CT-Planner4: Toward a More User-Friendly
Interactive Day-Tour Planner. 73–86. https://doi.org/10.1007/978-3-319-03973-
2_6
[12] Gilbert Laporte and Silvano Martello. 1990. The selective travelling salesman
problem. Discrete Applied Mathematics 26, 2 (1990), 193 – 207. https://doi.org/
10.1016/0166-218X(90)90100-Q
[13] Francesco Ricci. 2002. Travel Recommender Systems. IEEE Intelligent Systems 17,
6 (Nov. 2002), 55–57.
[14] Wouter Souffriau, Pieter Vansteenwegen, Joris Vertommen, and Greet Van-
den Berghe. 2008. A Personalized Tourist Trip Design Algorithm For Mobile
Tourist Guides. Applied Artificial Intelligence 22 (10 2008), 964–985. https:
//doi.org/10.1080/08839510802379626
[15] Wolfgang Wörndl, Alexander Hefele, and Daniel Herzog. 2017. Recommending
a Sequence of Interesting Places for Tourist Trips. Information Technology &
Tourism 17, 1 (01 Mar 2017), 31–54. https://doi.org/10.1007/s40558-017-0076-5
Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).