Personalized, Context-aware Intermodal Travel Information

Christian Samsel, Gerrit Garbereder and Karl-Heinz Krempels

Information Systems, RWTH Aachen University, Aachen, Germany

Keywords:

Context-aware Computing, Intelligent Transportation Systems, Recommendation Systems, Web Information

Systems.

Abstract:

The integration of heterogeneous mobility services increases the number of itinerary choices exponentially.

To support travelers with the selection of such an intermodal itinerary this work proposes the use of a rec-

ommendation system. The developed framework rates intermodal itineraries supplied by an external travel

information system based on learned personal preferences and user context (e.g. weather). This rating can be

used by the client application (e.g. a mobile app) for sorting or a ﬁve-star rating. The framework realizes a set

of interfaces to extract feature data of the user context and the possible itineraries and applies a combination

of item-based and context-based recommendation algorithms. As evaluation an online questionnaire (n = 101)

applying the framework was conducted to assess the feasibility of the approach. The number of participants

preferring the personalized and context-aware itinerary presentation compared to the traditional departure

time-based presentation was signiﬁcant. Furthermore it could be veriﬁed that a mobility self-assessment is

suitable as initial training data.

1 INTRODUCTION

Recent technological and socio-economical develop-

ments changed personal mobility signiﬁcantly. Re-

cent examples are the growth in car sharing (Sha-

heen and Cohen, 2007), bike sharing services (Sha-

heen et al., 2010) and services like Uber (Cusumano,

2015). The advent of such transport modalities and

the combination with ordinary modalities give travel-

ers more choices regarding their itineraries. Although

this improves the service coverage and is potentially

more environmentally friendly, the rising complexity

in travel planning is problematic. Modern intermodal

travel information systems (TIS) might offer dozens

of different itineraries for a ﬁxed journey. These

itineraries differ from each other in terms of duration,

price, modality, number of changes, environmental

friendliness, and many more aspects. Comprehending

these differences and hereafter selecting the appropri-

ate itinerary can therefore be a hard task.

This work proposes to enhance travel information

systems by rehashing the available itineraries to bet-

ter suit the traveler, based on his personal preferences,

context information (e.g. weather), and popular selec-

tions. In practice, this could improve the selection in

following ways:

• For persons who tend to select cheap itineraries,

the cheapest one is shown ﬁrst.

• In case of rain, itineraries containing bikesharing

are presented less prominently.

• If multiple people avoid a speciﬁc bus line (maybe

it’s overcrowded), it is presented less prominently.

Systems implementing such tasks, are called Rec-

ommender Systems. Recommender Systems are in-

formation systems which suggest one or more items

from a set of items to a user, based on similarities. The

best known recommendation methods are collabora-

tive ﬁltering (operates on user similarity), item-based

ﬁltering (item similarity), and context-aware ﬁltering

(context similarity).

The remainder of this paper is structured as fol-

lows: In Section 2, related research and existing ap-

plications are discussed. Section 3 describes the pro-

posed approach on a conceptual level, whereas Sec-

tion 4 presents details of the technical realization of

the prototype. The evaluation methodology and re-

sults are presented in Section 5 followed by Section 6

concluding the paper.

2 RELATED WORK

This Section gives an overview on production appli-

148

Samsel, C., Krempels, K-H. and Garbereder, G.

Personalized, Context-aware Intermodal Travel Information.

In Proceedings of the 12th International Conference on Web Information Systems and Technologies (WEBIST 2016) - Volume 1, pages 148-155

ISBN: 978-989-758-186-1

Figure 1: The Qixxit travel information web-platform.

cations and current scientiﬁc work with focus on in-

termodality and personalization.

Travel Information Systems have been scientiﬁ-

cally investigated and technically improved in several

aspects in recent years. Modern travel information ap-

plications like Transit App

, Moovel

or Qixxit

have

different strength and weaknesses (Beutel and Krem-

pels, 2014). Qixxit (see Figure 1) is the newest travel

information platform provided by Deutsche Bahn. It

offers an extensive variety of different modalities pro-

vided by different operators accompanied by a state-

of-the-art user experience. Crucial for the success

of TISes are two main factors: a) the quality of the

provided information in terms of comprehensiveness

and completeness and b) the systems’ usability (Beul-

Leusmann et al., 2014). (Papangelis et al., 2013) state

that travelers are highly frustrated when using pub-

lic transportation because of lacking information pro-

vided. The problem of information and data integra-

tion of heterogeneous mobility services is currently

under investigation (Beutel et al., 2014; Kluth et al.,

2015). A taxonomy of the public transport context is

presented in (Kr

omker and Wienken, 2015). In (Dig-

mayer et al., 2015) the authors identify and model the

phases of an intermodal journey to distinguish user

requirements, whereas in (Vogelsang et al., 2015) a

literature-based study on requirements on information

systems is presented. (Keller et al., 2011) observed

that creating an ideal representation of public tran-

sit trip information on a mobile device is a difﬁcult

task. Using paper prototypes, different approaches to

display intermodal travel chains are compared. The

authors of (Wienken et al., 2014) embedded a mo-

bility planning model hierarchically into an agenda

planning model and identiﬁed common information

needs. (Stopka et al., 2015) conducted an empiri-

http://transitapp.com

https://www.moovel.com/en

http://www.qixxit.de/en

cal study of current intermodal mobility applications

using Apple AppStore and Google Playstore data as

well as an online survey. The work identiﬁes different

user types (e.g. “The Open-Minded Planner”) as well

as different operation modes, e.g. “Route Search”.

The framework proposed in (Pessemier et al.,

2014) detects the current context and activity of the

user by analyzing data retrieved from different sen-

sors available on mobile devices and recommends ac-

tivities. In (Yang et al., 2015) the authors present an

algorithmic framework for personalized and context-

aware driving routing based on trajectories. The driv-

ing behavior (e.g. fuel consumption) is modeled as

context and new routes or rather trajectories are per-

sonalized accordingly. The work presented in (Co-

dina et al., 2015) was conducted in project SUPER-

HUB. A contextual user model was applied to recom-

mend multi-modal journey plans. The system sup-

ports the modalities walk, bike, public and private

transportation, as well as combinations. Sharing ser-

vices are not considered. Ten types of contextual fac-

tors are identiﬁed as relevant, e.g. purpose of journey

and weather. To evaluate the users had to rate jour-

neys using a 5-star scale with regard to a manually

entered context.

3 APPROACH

This section presents the theoretical approach used to

construct itinerary recommendations.

3.1 Modeling Users, Context and

Itineraries

To calculate similarities between items (i.e.,

itineraries) or users (i.e., travelers) it is required

to model them accordingly. These initial models

are kept as simple as possible to allow a reasonable

implementation and to only rely on available data.

For the ﬁrst implementation we opt for a sim-

ple user model only consisting of sex and age group.

The age groups are aligned to the age groups used in

(Follmer et al., 2010).

The context model is simplistic as well. Related

work, e.g. (Pessemier et al., 2014), put a strong focus

on thorough user context detection (e.g., using hard-

ware sensors). This was intentionally not done for this

work, because the context while planning is likely to

differ from the context of the actual travel. Instead

the probable context of time and place of the journey

is processed. That is the time of day and weather of

both start and destination. Temperature is clustered

Personalized, Context-aware Intermodal Travel Information

149

in weather conditions (sunny, rainy, etc.) and by tem-

perature in 5

◦

C intervals. Time is clustered in six hour

intervals.

In contrast, the itinerary model is bit more sophis-

ticated. A itinerary can be described by numerous

features, obviously important ones are overall dura-

tion and price. Lesser important but sometimes men-

tioned might be the slope of a bike leg. For this work

we aimed for a compromise of deemed important and

available information. In literature (Vogelsang et al.,

2015) duration and price are identiﬁed as main fea-

tures for decision making. The price was unfortu-

nately not available through public APIs, but is still

part of the model for a later integration. The simi-

larity s of two durations (d, d

) (for prices analog) is

calculated as:

s(d, d

) =

1 + |d − d

(1)

The reason for this transformation, is that similarity is

usually denoted normalized to [−1, 1] ⊆ R, whereas

−1 denotes opposite and 1 equality. Also relevant for

most travelers are the modalities used. Two itineraries

are considered similar iff the dominant, that is time-

wise longest used, modalities match.

3.2 Creating Recommendations

The essence of this work is recommending itineraries

for travelers. To create recommendations a combi-

nation of different techniques, i.e., algorithms, is uti-

lized. These algorithms have no inter-dependencies

and are therefore both replaceable and extensible.

As already mentioned the itinerary duration is one

of the most important decision factors. It is safe

to assume that travelers will always select the faster

itinerary of otherwise identical ones. Accordingly, we

use the travel duration for an initial score. The list of

itineraries i ∈ I is sorted ascending by duration and

then rated (r

) using:

= (1 −

|I| −1

) ∗ 0.5

with i

∈ (i

, ..., i

|I|

), 1 ≤ k ≤ |I|.

(2)

The actual duration is not used here, but instead the

ordering of possible itineraries. The weight is modi-

ﬁed using the factor 0.5, so the impact on overall rec-

ommendation score is lower compared to the follow-

ing recommender.

The way carsharing services work, it is advisable

to treat it specially regarding recommendations. Car-

sharing requires a formal registration with the opera-

tor to check the travelers drivers license. Based on the

fact that the majority of traveler are not registered, we

assume that only traveler who already used carsharing

want recommendations containing carsharing legs.

Let p

be the frequency of carsharing travels,

where as let p

be the frequency of non-carsharing

travels.







, p

6= 0

0, p

= 0

1, p

= 0

(3)

For the following recommendations the overall sim-

ilarity between itineraries is required. We already

introduced feature similarity in Section 3.1 but still

need a way of combining them. For combined sim-

ilarity only features present in both sets are consid-

ered. The similarity between itineraries is deﬁned as

s( f , f

). Let F be the set of features of i ∈ I, and let

be the set of features of i

∈ I.

s(i, i

) =

∑

∀ f , f

∈F

s( f , f

)

, with F

= F ∩ F

. (4)

The weighting of feature similarities is conceivable

but was not used for this work.

To enrich the similarity with context-based infor-

mation we extend the used function with context at

the origin (C

) and the context at the destination (C

)

of itinerary i ∈ I. C

, C

likewise for itinerary i

∈ I.

s(i, i

) =

∑

∀ζ,ζ

∈Z

s(ζ, ζ

) +

∑

∀γ,γ

∈Γ

s(γ, γ

) +

∑

∀γ,γ

∈Γ

s(γ, γ

)

|Z| + |Γ

| + |Γ

with Z = F ∩ F

, Γ

= C

∩C

, Γ

= C

∩C

(5)

The ﬁnal personal item-based rating, is calculated by

multiplying the personal preference p

u,i

for item i

by the similarity s(s, s

) whereby P ⊆ I ist the set of

learned preferences. Additionally we only consider

items with similarity values of 0.75 or more (P

). Ini-

tial tests of systems showed that lower similarities de-

note completely unrelated itineraries.

= {i

∈ P ∧ s(i, i

) > 0.75},

∑

∈P

u,i

∗ s(i, i

(6)

Combining Equation (5) and Equation (6) results in

the ﬁnal item-based and context-aware recommender.

The algorithm is designed to deal with adding or re-

moving of itinerary and context features by only cal-

culating similarities if the respective feature exist for

both items.

3.3 Filtering Results

As ﬁnal step the results are ﬁltered depending on the

conﬁguration, i.e., the request. It is possible to ﬁl-

ter the results based on a score threshold or to ﬁlter

WEBIST 2016 - 12th International Conference on Web Information Systems and Technologies

150

App

PCR TIS

Query (+ Context)

Itineraries (+ Scores)

Query

Itineraries

Figure 2: Information Exchange with PCR.

results qualitatively. For example, a ﬁlter could elim-

inate itineraries with pedestrian legs longer then 5km,

as suggested in (Follmer et al., 2010).

4 REALIZATION

In this section the actual implementation of the pro-

posed framework is described. The system was de-

veloped in Java Enterprise Edition 7, as application

container Red Hat Wildﬂy 8

is used. To enable easy

modularization, Java Context and Dependency Injec-

tion (CDI) is employed. As data storage MongoDB

is used. For build and deployment Maven

, Vagrant

and Puppet

are used.

4.1 System Architecture

To simplify the integration into existing travel infor-

mation systems, the personalized, context-aware rec-

ommendation (PCR) system can also transparently

work between an existing (mobile) travel information

application and an existing backend as shown in Fig-

ure 2.

4.2 Data Acquisition and Feature

Extraction

Itineraries or rather routing information, are the most

important input data. The creation of itineraries us-

ing routing algorithms for public transport and such,

is not in the scope of this work, instead the PCR sys-

tem relies on existing travel information systems. The

PCR supports the integration of multiple routing ser-

vices by offering a plugin API. Currently, Google Di-

rections

and MapQuest

are supported via respective

RESTful API clients. Both services offer itineraries

consisting of pedestrian legs, public transport legs,

private car legs as well as combinations. To enhance

the variety, it is appropriate to include more uncom-

mon modes of transportation, e.g. carsharing, as men-

tioned earlier. Unfortunately, TISes like Qixxit (see

http://www.wildﬂy.org

http://maven.apache.org

http://www.vagrantup.com

http://puppetlabs.com/puppet/what-is-puppet

https://www.google.com/maps/dir/

http://www.mapquest.com

Section 2) or MobilityBroker (Beutel et al., 2014)

supporting holistic intermodal itineraries did not of-

fer an open API at the time the system was developed.

To still allow such itineraries, the itineraries supplied

by Google Directions and MapQuest were augmented

with realistic carsharing legs for the evaluation.

The Context Data injection is modularized. For

testing and demonstration of the PCR system, time

of day and weather data supplied by OpenWeath-

erMap

is supported. For start and destination of

itineraries the corresponding weather information are

fetched and cached for follow-up requests. Besides

weather information various other context informa-

tion sources are conceivable to use.

4.3 Recommendation System

The Recommendation Engine is built on top of

Apache Mahout

, which is a framework for devel-

oping scalable machine learning and recommenda-

tion applications. Mahout can be used in conjunc-

tion with Apache Hadoop

and/or Apache Spark

for distributed computing, but the use is optional. For

the relatively small data pool and number of users, we

opted to not use any kind of distributed computing.

4.4 Test and Evaluation Client

To test the server-based framework an Android mo-

bile application was developed for demonstration.

Because the graphical user interface was not in scope

of this work, the application only uses a text based

interface. For information on the design of mobile

travel information application, refer to Section 2. The

application allows registration, login, as well as the

main functionality: the actual travel query. After the

input of start, destination and departure time the user

is presented with a list of possible itineraries sorted

by the recommendation rating. He or she can select

one of them as usual to train the system. In case of

anonymous usage, which is also possible, personal-

ized information is not considered.

To evaluate the system, a special web client was

developed using AngularJS (Figures 3 and 4). The

http://openweathermap.org

http://mahout.apache.org

http://hadoop.apache.org

http://spark.apache.org

Personalized, Context-aware Intermodal Travel Information

151

Figure 3: User-Study Learning Phase.

collected data is saved in a MongoDB database for

later analysis. The evaluation component is modular-

ized so it can be easily removed for production oper-

ation.

5 EVALUATION

Aim of the user study is to decide whether the ap-

proach is feasible and to gain general insight. The

methodology for the evaluation is described in Sec-

tion 5.1 and the results are presented in Section 5.2.

5.1 Methodology

The user study was developed as a web based appli-

cation. The application consists of three phases:

1. Collection of demographic data and mobility self-

assessment,

2. Training phase (6 selections),

3. Evaluation phase (3 selections).

Demographic data questions are the usual refer-

ence data (e.g., age, sex, professional and educa-

tional background). The mobility self-assessment

consists of questions whether the participants con-

sider themselves as “car-person”, “train-person” or

“bike-person” (multiple answers possible).

Both the training and evaluation phase do not al-

low the traveler to query arbitrary journeys but instead

query a ﬁxed set of scenarios. A scenario consists of a

Figure 4: User-Study Evaluation Phase.

start and destination, ﬁxed contextual information and

description for the participant, e.g.: “You are in city

A work-related and want to meet for dinner in city

C. The weather is sunny. Please select your preferred

itinerary.” The graphical representation resembles a

state-of-the-art TIS with a time-scaled, vertical, icon-

based visualization of travel chains (Vogelsang et al.,

2015).

In the training phase the participant selects a suit-

able itinerary out of the presented ones, based on e.g.

duration, used modal types or number of changes,

just as in the real world (see Figure 3). The selec-

tion for each question is then used to train the recom-

mender system. Unfortunately, prices are not shown,

as none of the supported travel information systems

offer price information. This would be a major con-

cern for the productive operation of the system but

does not limit the evaluation as the price information

is just an additional itinerary feature similar to the du-

ration.

In the evaluation phase (see Figure 4) the partici-

WEBIST 2016 - 12th International Conference on Web Information Systems and Technologies

152

100

150

200

Identical

Classic

PCR

# Selections

Figure 5: Total variant selection.

pant can select his or her preferred itineraries sorting.

He or she can select between the PCR-based sorting

and a traditional sorting (departure time-based). The

position of both variants is randomized.

The recommendation knowledge base is freshly

initialized for every participant to allow independent

and reproducible results.

Of 101 participants who completed the question-

naire, 52 were female. Participants ages ranged from

18 to 82 with an average of 33.6 years. 95% of the

participants own a drivers license, 50.5% hold a least

college degree. The overall answer time per person

was around 10-15 minutes.

5.2 Results

The central question of this evaluation is, whether

travelers prefer a personalized, context-aware

itinerary presentation compared to the traditional,

departure-time based presentation. To answer, par-

ticipants could choose between these variants in the

evaluation phase of the questionnaire. The results, in

number of selections, are shown in Figure 5. Identical

means that both variants (PCR and traditional) had

identical sorting, so no improvement was gained by

PCR. With a signiﬁcance level of 95%, 65% of the

selections are for PCR, which proves that travelers

prefer a personalized, context aware itinerary selec-

tion. The number of selections is three times the

number of participants because every participant had

three choices.

A common problem for recommendation systems

is the cold start problem, which is to give meaningful

recommendations to novel users for whom the system

has not learned any preferences (Lam et al., 2008).

For productive use, it is impractical to let new users

go through a lengthy learning phase of six selections

as carried out in the evaluation questionnaire. Instead

we assume answering a few self-assessment questions

to be more suitable usability-wise and enough infor-

mation for initialization. In Figure 6 the modality

selection and the respective modality self-assessment

is depicted. Travelers, who consider themselves as

Accept Unsure

Reject

100

Self-assessment

Selection in %

Train Car Bike

Figure 6: Modal use for respective self-assessment.

Accept Unsure

Reject

100

Self-assessment

Variant in %

PCR Classic Identical

Figure 7: Relative variant selection per self-assessment.

“train-person”, selected a train connection in 72.5%

of the cases, whereas travelers who rejected this, only

selected a train connection with a likelihood of 37%.

The relative high number of car and train selections

compared bike selections is caused by the nature of

the scenarios. Four out of six trips cover a long dis-

tance (50km - 600km) and one of the remaining two

is a job interview which generally discourages to use

a bike. Based on the correlation of the mobility self-

assessment and actual selection, one can assume that

self-assessment is usable as initial personal training

data. That is, the recommender systems behave as if

the new user had chosen a respective itinerary previ-

ously based on his self-assessment.

Figure 7 shows the relation between the share of

selections of PCR vs. the traditional presentation and

the self-assessment answer. The interesting result is

the bend for the answer “unsure”; Travelers, who are

unsure about their own mobility proﬁle, are less likely

Personalized, Context-aware Intermodal Travel Information

153

to select PCR. That leads to the conclusion, that trav-

elers with a volatile travel behavior are harder to pre-

dict and therefore harder to present with satisfying

recommendations. This is somewhat expected, but

still noteworthy.

6 CONCLUSION

This work presented a system for enhancing the

itinerary selection by employing personalized and

context-aware recommendations. The system was im-

plemented as a web service with a modular archi-

tecture to easily extend existing travel information

systems. The recommendation engine is based on

Apache Mahout and uses item-based, context-aware

scoring. To evaluate the system an online question-

naire with both a learning and testing phase was

used. The questionnaire results (n = 101) showed that

participants prefer a personalized and context-aware

sorted itinerary selection.

Discussion and Outlook

Although the presented evaluation showed that the

approach is technically working and the enhanced

itinerary selection is superior, some aspects are not

fully covered. Not all mobility services, most no-

tably ﬂights and carpooling, are currently supported.

Even more important, price information is not cov-

ered, because of the lack of public APIs that offer this

information. We can assume that these features are

required for productive operation of such systems to

be on par with existing TISes.

The system in its current form does utilize col-

laborative ﬁltering, i.e., population data. Also the

employed models are simple compared to models

used in established recommender systems domains

like eCommerce. Extending the models and algo-

rithms used, can improve the recommendation re-

sults fruther, but requires a signiﬁcant amount of data,

which was not available for evaluation. Because of

the modular design of the system, the implementa-

tion of enhanced models and algorithms is straight-

forward.

A potential gap in the current implementation is

the way that travelers explicitly state that they are us-

ing a speciﬁc itinerary. For an occasional travel, this

is conceivable. But for a daily commute, travelers

might not interact with a travel information system

at all and instead opt for an itinerary purely based on

experience. Still, these information might be valuable

both for the personal proﬁle of the traveler and for

population data. To solve this, an automatic tracking,

potentially combined with a travel assistance system

(Samsel et al., 2015), is required.

Personalization (and tracking even more) always

has implications towards privacy. Travelers have to

be informed about the workings of such a system and

should always have the option to use it anonymously.

This way a traveler can make a distinct decision if he

or she wants to trade privacy for comfort (Kowalewski

et al., 2015).

Last but not least, the presented evaluation was

conducted based on hypothetical scenarios without

any actual travel conducted. It is still unknown how

such a system performs in productive operation and if

traveler actually considered it useful in the long run.

ACKNOWLEDGEMENTS

This work was funded by the German Federal Min-

istry of Economic Affairs and Energy for project Mo-

bility Broker (01ME12136).

REFERENCES

Beul-Leusmann, S., Samsel, C., Wiederhold, M., Krem-

pels, K., Jakobs, E., and Zieﬂe, M. (2014). Usability

evaluation of mobile passenger information systems.

In Design, User Experience, and Usability. Theories,

Methods, and Tools for Designing the User Experi-

ence - Proceedings of the third International Confer-

ence, DUXU 2014, Crete, Greece, June 22-27, 2014,

pages 217–228.

Beutel, M. C., Gokay, S., Kluth, W., Krempels, K., Samsel,

C., and Terwelp, C. (2014). Product oriented integra-

tion of heterogeneous mobility services. In Proceed-

ings of the 17th IEEE International Conference on In-

telligent Transportation Systems (ITSC 2014), pages

1529–1534.

Beutel, M. C. and Krempels, K. (2014). Encompassing

payment for heterogeneous travelling - design impli-

cations for a virtual currency based payment mech-

anism for intermodal public transport. In Proceed-

ings of the 3rd International Conference on Smart

Grids and Green IT Systems (SMARTGREENS 2014),

Barcelona, Spain, 3-4 April, 2014, pages 305–310.

Codina, V., Mena, J., and Oliva, L. (2015). Context-

aware user modeling strategies for journey plan rec-

ommendation. In Ricci, F., Bontcheva, K., Conlan,

O., and Lawless, S., editors, User Modeling, Adap-

tation and Personalization, volume 9146 of Lecture

Notes in Computer Science, pages 68–79. Springer In-

ternational Publishing.

Cusumano, M. A. (2015). How traditional ﬁrms must

compete in the sharing economy. Commun. ACM,

58(1):32–34.

WEBIST 2016 - 12th International Conference on Web Information Systems and Technologies

154

Digmayer, C., Vogelsang, S., and Jakobs, E. (2015). De-

signing mobility apps to support intermodal travel

chains. In Proceedings of the 33rd Annual Interna-

tional Conference on the Design of Communication

(SIGDOC 2015), Limerick, Ireland, July 16-17, 2015,

pages 44:1–44:11.

Follmer, R., Gruschwitz, D., Jesske, B., Quandt, S., Lenz,

B., Nobis, C., K

ohler, K., and Mehlin, M. (2010).

Mobilit

at in Deutschland 2008 [Mobility in Germany

2008]. Methodenbericht. Berlin, Germany.

Keller, C., Korzetz, M., K

uhn, R., and Schlegel, T. (2011).

Nutzerorientierte Visualisierung von Fahrplaninfor-

mationen auf mobilen Ger

aten im

offentlichen

Verkehr [User-oriented visualization of train sched-

ule information on mobile Devices for public

transportation]. In Mensch & Computer 2011:

uberMEDIEN|

UBERmorgen - 11. fach

ubergreifende

Konferenz f

ur interaktive und kooperative Medien,

Chemnitz, Germany, September 11-14, 2011, pages

59–68.

Kluth, W., Beutel, M. C., G

okay, S., Krempels, K., Samsel,

C., and Terwelp, C. (2015). IXSI - Interface for X-

Sharing Information. In Proceedings of the 11th In-

ternational Conference on Web Information Systems

and Technologies (WEBIST 2015), Lisbon, Portugal,

20-22 May, 2015, pages 293–298.

Kowalewski, S., Zieﬂe, M., Ziegeldorf, H., and Wehrle, K.

(2015). Like us on Facebook!–Analyzing user prefer-

ences regarding privacy settings in Germany. Proce-

dia Manufacturing, 3:815–822.

omker, H. and Wienken, T. (2015). Context elicita-

tion for user-centered context-aware systems in public

transport. In Human-Computer Interaction: Interac-

tion Technologies - Proceedings of the 17th Interna-

tional Conference, HCI International 2015, Los An-

geles, CA, USA, August 2-7, 2015, pages 429–439.

Lam, X. N., Vu, T., Le, T. D., and Duong, A. D. (2008).

Addressing cold-start problem in recommendation

systems. In Proceedings of the 2nd International

Conference on Ubiquitous Information Management

and Communication (ICUIMC 2008), pages 208–211,

New York, NY, USA. ACM.

Papangelis, K., Sripada, S., Corsar, D., Velaga, N., Ed-

wards, P., and Nelson, J. D. (2013). Developing a

real time passenger information system for rural areas.

In Human Interface and the Management of Informa-

tion. Information and Interaction for Health, Safety,

Mobility and Complex Environments - Proceedings of

the 15th International Conference, HCI International

2013, Las Vegas, NV, USA, July 21-26, 2013, pages

153–162.

Pessemier, T. D., Dooms, S., and Martens, L. (2014).

Context-aware recommendations through context and

activity recognition in a mobile environment. Multi-

media Tools Appl., 72(3):2925–2948.

Samsel, C., Dudschenko, I., Kluth, W., and Krempels, K.

(2015). Using wearables for travel assistance. In

Proceedings of the 11th International Conference on

Web Information Systems and Technologies (WEBIST

2015), Lisbon, Portugal, 20-22 May, 2015, pages

635–641.

Shaheen, S. A. and Cohen, A. P. (2007). Growth in world-

wide carsharing: An international comparison. Trans-

portation Research Record: Journal of the Trans-

portation Research Board, 1992(1):81–89.

Shaheen, S. A., Guzman, S., and Zhang, H. (2010). Bike-

sharing in Europe, the Americas, and Asia. Trans-

portation Research Record: Journal of the Trans-

portation Research Board, 2143(1):159–167.

Stopka, U., Pessier, R., and Fischer, K. (2015). User

requirements for intermodal mobility applications

and acceptance of operating concepts. In Human-

Computer Interaction: Design and Evaluation - Pro-

ceedings of the 17th International Conference, HCI

International 2015, Los Angeles, CA, USA, August 2-

7, 2015, pages 415–425.

Vogelsang, S., Digmayer, C., and Jakobs, E. (2015). User

requirements on intermodal traveler information sys-

tems. In Proceedings of the IEEE International Pro-

fessional Communication Conference (IPCC 2015),

Limerick, Ireland, July 12-15, 2015, pages 1–9.

Wienken, T., Mayas, C., H

orold, S., and Kr

omker, H.

(2014). Model of mobility oriented agenda planning.

In Human-Computer Interaction. Applications and

Services - Proceedings of the 16th International Con-

ference, HCI International 2014, Heraklion, Crete,

Greece, June 22-27, 2014, pages 537–544.

Yang, B., Guo, C., Ma, Y., and Jensen, C. S. (2015). To-

ward personalized, context-aware routing. The VLDB

Journal, 24(2):297–318.

Personalized, Context-aware Intermodal Travel Information

155