Personalisation in Mobility-as-a-Service: Where We Are and How to

Move Forward

Kamaldeep Singh Oberoi

CESI LINEACT, Toulouse, 31670, France

Keywords:

Mobility-as-a-Service, Personalisation, User Preferences, Optimal Route Recommendation, User Privacy,

User Feedback.

Abstract:

Within urban mobility ecosystem, Mobility-as-a-Service (MaaS) has come up as a promising approach to pro-

mote sustainable modes of transport and increase the attractiveness of public and shared multimodal mobility.

It aims to become a viable alternative to personal cars for door-to-door trips. The long-term objective of MaaS

is to change the people’s travel behaviour by nudging them to make sustainable choices. However, changing

people’s travel behaviour is not easy and MaaS has to provide personalised mobility services, catering to the

needs of each individual user, in order to be considered as convenient as a personal car. In this paper, we look

at the existing literature on personalisation in MaaS proposed by the research community as well as different

private MaaS service providers. This brief literature review helps in better understanding the current trends on

personalisation and highlights certain limitations in the way it is incorporated within existing MaaS solutions.

Based on these limitations, we discuss certain challenges which need to be resolved in order to improve MaaS

in the future. These challenges present interesting research directions towards the development of personalised

sustainable urban mobility ecosystem.

1 INTRODUCTION

Over the past few years, European Commission has

been encouraging the development of Sustainable Ur-

ban Mobility Plans (SUMP) (EU Urban Mobility Ob-

servatory, 2023) for European towns and cities to

tackle the challenges of climate change and urban

trafﬁc congestion. SUMP address the complexity of

urban transport, integrate passenger and freight de-

mand, and promote sustainable modes of transport to

improve the overall quality of life.

To concretize the guidelines put forth within

SUMP (EU Urban Mobility Observatory, 2023),

cities need to improve their public transport infras-

tructure, integrate new modes of transport (such as

car and bike sharing, on demand transport, carpool-

ing, etc.) and develop platforms to improve the acces-

sibility of public and shared mobility options. It turns

out that the concept of Mobility-as-a-Service (MaaS)

(Hietanen, 2014) can help overcome the challenges in

implementing SUMP and, for a city, both SUMP and

MaaS can be integrated and developed hand in hand

(Signor et al., 2019).

https://orcid.org/0000-0002-0334-3055

Over the past few years, MaaS has gained signif-

icant interest in the urban mobility and transportation

sector (Hensher et al., 2020). In essence, it aims to

bring together different mobility services, provided

by a single or multiple Mobility Service Providers

(MSPs), on a single platform, with the possibility of

searching, booking, and paying for such multimodal

services directly on the platform (Hietanen, 2014).

Furthermore, it supports the vision of mobility-for-all

and pushes for access-based mobility services as com-

pared to ownership-based mobility models with the

objective of reducing the use of personal cars (Hen-

sher et al., 2020). The concept of MaaS places user

needs and environmental conservation at the center,

while aiming for a seamless integration of multimodal

mobility services.

One of the long-term objectives of MaaS is to

gradually change people’s travel behaviour and per-

suade them to make sustainable transport choices.

According to (Hensher et al., 2020), MaaS has the

ability to inﬂuence people’s travel behaviour, how-

ever, it cannot achieve this objective if people are not

incentivized to choose shared multimodal transport

over personal cars. The authors propose that in ad-

dition to deploying MaaS, cities should also reduce

352

Oberoi, K.

Personalisation in Mobility-as-a-Service: Where We Are and How to Move Forward.

DOI: 10.5220/0012687100003702

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 10th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2024), pages 352-360

ISBN: 978-989-758-703-0; ISSN: 2184-495X

the attractiveness of personal cars by restricting their

use in certain areas and imposing parking regulations.

However, instead of “punishing” people for

choosing personal cars, certain behaviour change

strategies (Durand et al., 2018) can be integrated

within MaaS which encourage people to choose sus-

tainable modes of transport. Such strategies could im-

prove the overall attractiveness of shared multimodal

transport as well as the acceptance rate of MaaS as a

concept.

In this paper, we focus on one such strategy to sup-

port behaviour change, personalisation. We discuss

the aspect of personalisation within existing MaaS

research and MaaS platforms developed by private

MaaS service providers. After brieﬂy presenting the

concept of MaaS in Section 2.1, we discuss the im-

portance of personalisation in Section 2.2, and how

it is integrated in existing MaaS solutions in Section

2.3. From this brief review of the state-of-the-art, we

are able to ﬁnd some limitations in personalisation in

existing MaaS solutions. These limitations are dis-

cussed in Section 2.4. In Section 3, we present cer-

tain challenges in order to improve personalisation in

MaaS. These challenges are related to data availabil-

ity, data standardization and data sharing between dif-

ferent stakeholders, respecting user privacy while pro-

cessing and storing the collected data, integration of

dynamic user preferences, and proactive route recom-

mendations. We also discuss solutions to overcome

these challenges highlighting fascinating research di-

rections. Finally, Section 4 concludes the paper.

2 MOBILITY-AS-A-SERVICE

2.1 Brief Overview

Inspired by the business model of telecommunica-

tion sector, Hietanen (Hietanen, 2014) proposed the

concept of Mobility-as-a-Service (MaaS) as Netﬂix of

transportation where users would subscribe in order

to use different mobility services provided by a mo-

bility operator who, in turn, would be responsible for

integrating such services onto a single platform. On

the platform, users would be presented with different

combinations of transport modes as well as different

payment options (like mobility bundles, monthly sub-

scription plans, etc.) which the user could choose ac-

cording to his/her preferences and requirements. The

idea was to develop the transportation system with,

to and by the users as an interconnected ecosystem

consisting of transport infrastructure, transportation

services, transport information and payment services

(Hietanen, 2014).

As the concept of MaaS gained traction, more def-

initions were proposed. For example, (Burrows et al.,

2016) deﬁned MaaS as the provision of transport as

a ﬂexible, personalised on-demand service that inte-

grates all types of mobility opportunities and presents

them to the user in a completely integrated manner

to enable them to get from A to B as easily as pos-

sible. Similarly, (Kamargianni et al., 2018) deﬁned

MaaS as user-centric, intelligent mobility manage-

ment and distribution system, in which an integra-

tor brings together offerings of multiple mobility ser-

vice providers, and provides end-users access to them

through a digital interface, allowing them to seam-

lessly plan and pay for mobility.

In order to facilitate the understanding of MaaS,

(Sochor et al., 2018) reviewed multiple deﬁnitions

and proposed a topology to compare existing MaaS

platforms. According to this topology, MaaS could

be perceived at four different levels (from level 1 to

level 4), each improving the services integrated at

lower levels, ranging from the integration of informa-

tion such as route planner, price, and environmental

cost of a route (level 1) and payment options (levels

2 and 3) to the integration of societal goals (level 4)

deﬁned within government policies to incentivize the

use of MaaS at large scale. Similarly to the work of

(Sochor et al., 2018), other topologies that character-

ize MaaS have also been proposed in the literature.

For example, (Opiola, 2018) considers MaaS at six

different levels, with higher levels going beyond mo-

bility and integrating other digital services (like smart

homes and IoT) within MaaS. Taking into account the

cognitive effort required to undertake a multimodal

route, (Lyons et al., 2019) proposed a topology hav-

ing ﬁve levels of integration, with higher levels as-

sociated to the requirement of lower cognitive effort.

It is argued that, in order for MaaS to compete with

the convenience of personal cars, it should be as easy,

convenient, and ﬂexible for everyone.

Out of various deﬁnitions explored in (Sochor

et al., 2018), more than half describe MaaS as a per-

sonalised service for planning, booking, paying and

executing the trips. Furthermore, from the differ-

ent topologies discussed above (Sochor et al., 2018;

Opiola, 2018; Lyons et al., 2019), one can note

that, at higher levels, MaaS integrates multiple modes

of transport with various payment options, while

proposing personalised mobility services catered to

the needs of the user. The importance of person-

alisation within MaaS is evident not only from the

existing literature, but users also want MaaS plat-

forms which provide personalised support, as noted

by (Polydoropoulou et al., 2020) through a user sur-

vey and a focus group based study.

Personalisation in Mobility-as-a-Service: Where We Are and How to Move Forward

353

2.2 Role of Personalisation in MaaS

In the ﬁeld of information systems, personalisation

is deﬁned as the automatic adjustment of informa-

tion content, structure, and presentation tailored to an

individual user (Perugini and Ramakrishnan, 2003).

It takes into account the user’s preferences and past

habits while presenting relevant information or con-

tent at the time it is required (Gao et al., 2010), ideally

without any explicit demand from the user (Mulvenna

et al., 2000). The aspect of personalisation plays a key

role in nudging the user and supporting him to change

his behaviour (Prost et al., 2013) as well as improves

the user’s trust in the system (Briggs et al., 2004).

In case of multimodal transportation, incorporat-

ing personalisation as a tool to persuade users in

changing their travel behavior has been discussed in

the literature (Anagnostopoulou et al., 2018; Hensher

et al., 2020). Furthermore, (Andersson et al., 2018)

reviewed different behaviour change support strate-

gies and concluded that personalisation plays a signif-

icant role in improving user satisfaction. Better satis-

faction motivates the user to keep using the service

which, in turn, has positive long-term effect in chang-

ing user behaviour. More so, personalisation has been

identiﬁed as a core characteristic for any MaaS ser-

vice as it makes it more attractive and improves its

rate of acceptance (Jittrapirom et al., 2017).

The availability of heterogeneous information

about different modes of transport and possible pay-

ment options within MaaS increases the complexity

of using the service and choosing the right combina-

tion. As noted in (Hartikainen et al., 2019), users

face difﬁculties in ﬁnding the relevant information

about, for example, the schedule of public transport

and where to buy tickets. This is especially true for

unfamiliar areas and is a reason for increased travel-

related stress (Hartikainen et al., 2019). Personalisa-

tion of the service helps in reducing the complexity

of multimodal travel by providing right combination

of modes and payment options at the right time. The

trial of MaaS platform NaviGoGo in Scotland (Smith,

2019) showed that personalisation makes traveling

easier and increases user’s self-conﬁdence in travel-

ing using multiple modes. Simplifying multimodal

travel reduces the cognitive effort required to make

different choices, making MaaS as easy as using the

personal car (Lyons et al., 2019).

Along with improving user satisfaction and reduc-

ing travel-related stress, personalisation plays a sig-

niﬁcant role in making recommendations to users.

Recommendations maybe be about possible routes,

between origin and a destination, with different

modes of transport available or about available pay-

ment options (e.g. mobility bundles, subscriptions,

pay-as-you-go, etc.) according to user requirements

(Arnaoutaki et al., 2021). To make appropriate rec-

ommendations, user preferences are incorporated in

terms of static user proﬁles (Arnaoutaki et al., 2019)

or computed using the past data (de Oliveira e Silva

et al., 2022).

In the following, we discuss some of the existing

research on personalisation within MaaS as well as

the features of existing MaaS platforms. The listed

features personalise the user experience, as noted by

the MaaS operators behind said platforms. Here, we

focus only on information personalisation and not on

the personalisation of payment options. For more in-

formation about the latter, the reader is directed to

(Arnaoutaki et al., 2021).

2.3 Personalisation in MaaS Research

and Existing MaaS Platforms

Within the MaaS research community, some work has

been done to provide personalised mobility services

according to the user’s requirements. For example,

(Melis et al., 2018) developed a microservice archi-

tecture to propose personalised and accessible routes

to the users. The motivation behind using a microser-

vice architecture is to make the software scalable by

distributing its modules as microservices (Dragoni

et al., 2017). The authors demonstrated the applica-

bility of this approach by considering the use cases

of a blind user and a tourist, and discussed the ser-

vices required to deploy a personalised route planner

for both use cases. In the proposed approach, user

preferences (such as audible trafﬁc lights, tactile in-

formation boards, acoustic announcements, etc. for

blind users and scenic routes through historical build-

ings for tourists) were ﬁxed while setting up the user

proﬁle and were taken into account while calculating

the route between an origin and a destination.

Within the (MaaS4EU, 2020) project, (Georgakis

et al., 2020) proposed a personalised multimodal

route recommender integrating the information about

third-party services for bike and car sharing modes

as well as parking facilities. The route recommender

was based on the choice architecture design elements

where the users were provided with default route

choices between an origin and a destination which can

be ﬁltered according to the user preferences (such as

user’s preferred walking and biking distances). Addi-

tionally, (Anagnostopoulou et al., 2020) constructed

the users’ “persuadability proﬁles” based on their per-

sonality types and proposed to attach personalised

messages to nudge users towards sustainable modes

of transport. The persuadability proﬁles described the

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

354

Table 1: Features of some of the existing MaaS platforms which personalise the user experience. Indicated features are

susceptible to evolve over time.

MaaS Platforms

Features

Multimodal

route

planner

User

proﬁle

Sync with

user’s

calendar

Real-time

trafﬁc

information

Store fav.

routes

and

locations

Display

CO2

emissions

Occupancy of

bike stations

and

parking lots

Whim ✓ ✓ ✓

UbiGo ✓ ✓ ✓

Moovit ✓ ✓ ✓ ✓ ✓

Moovizy ✓ ✓ ✓ ✓ ✓

NaviGoGo ✓ ✓

WienMobil ✓ ✓ ✓

Optymod’Lyon ✓ ✓ ✓ ✓

TransitApp ✓ ✓ ✓ ✓ ✓

Tuup/Kyyti MaaS Platform ✓ ✓ ✓ ✓

MobiPalma ✓ ✓ ✓

TripGo ✓ ✓ ✓ ✓

Ecomode ✓ ✓ ✓ ✓

Compte Mobilit

e ✓ ✓ ✓

users’ susceptibility to change their travel behaviour

and were used as a tool for supporting behaviour

change (Durand et al., 2018).

In addition to the ideas proposed by the research

community, some existing MaaS platforms, devel-

oped by private MaaS operators, have incorporate cer-

tain features to personalise the user experience. These

features are presented in Table 1. The list of plat-

forms in Table 1 is not exhaustive but it gives a gen-

eral idea of the level of personalisation in existing

MaaS platforms. It is noteworthy that we only con-

sider the features available on smartphone versions of

the platforms, those available on the web versions are

not considered.

All MaaS platforms have an integrated journey

planner with which users can search for a trip using

a combination of multiple modes available. Most of

the platforms consider public transport, bike and car

sharing, ride-hailing services etc. but some platforms

like Moovit and Ecomode also integrate carpooling

as a possible mode of transport. At the time of in-

stallation, the users are asked to register using their

personal information such as age, email, phone num-

ber etc. (which is common for most platforms), but

some platforms such as WienMobil, NaviGoGo, Op-

tymod’Lyon and Ecomode ask for extra information,

such as if the user owns a bicycle, a car etc. and how

far is the user willing to walk in order to take a public

or shared mode of transport. This information is used

to create the user’s proﬁle which helps in ﬁltering the

route propositions presented to the user.

It is also possible for users to synchronise their

personal calendars with MaaS platforms in order to

receive automatic alerts about their trips which re-

spect their daily constraints. Moreover, the option for

storing preferred routes and locations eliminates the

hassle to search for regular routes before every trip.

Platforms like WienMobile and TransitApp display

the carbon footprint with each route and incentivize

users to choose sustainable options. In order to alert

the user about trafﬁc jams and delays in public trans-

port, platforms such as TripGo, Optymod’Lyon, Mo-

biPalma, and TransitApp take into account the real-

time trafﬁc information. In addition, TransitApp up-

dates the bus schedules automatically when they are

changed, for example, on weekends or during rush

hour trafﬁc.

Complementary to the ones presented in Table 1,

platforms such as MobiPalma, Moovit and Moovizy

integrate additional features to promote inclusive mo-

bility as their services are available for people trav-

elling with wheelchairs or those with visual impair-

ments.

2.4 Limitations in Personalisation in

Existing MaaS Solutions

Although the existing work, both from the research

community and from private MaaS operators, is

promising, there still exist some limitations which

need to be resolved in order to improve the aspect

of personalisation of MaaS. For example, user pref-

erences and requirements need to be better taken into

account and user feedback should be automatically in-

tegrated to enhance the service provided (according to

the deﬁnition of the term personalisation presented in

Personalisation in Mobility-as-a-Service: Where We Are and How to Move Forward

355

Section 2.2). More importantly, user privacy should

not be taken lightly. Users should be consented ev-

ery time before processing their personal data (if done

on third-party servers) or it should be stored and pro-

cessed locally on their smartphones.

In the following, we brieﬂy discuss these user-

centric limitations of existing MaaS solutions:

• User Preferences: Existing platforms are able

to ﬁlter the routes according to user preferences

(such as shorter walking distance, less carbon

footprint, less travel time etc.). However, user

preferences, once integrated, are kept static. To

the best of our knowledge, there does not exist

a platform which automatically updates the user

preferences over time. Furthermore, as discussed

above, (Melis et al., 2018) take into account the

static preferences of a blind user. However, the

user is obligated to declare a personal health con-

dition in order to use the personalised service.

This might have an impact on the acceptability

of platforms which are based on such methodolo-

gies.

• User Feedback: Most of the existing MaaS plat-

forms demand explicit feedback from the users

about their experience with the platform. How-

ever, feedback about the quality of the routes pro-

posed and whether they conform to the user’s re-

quirements and context of travel is not considered.

It has been shown that integrating user feedback

makes the system more user-friendly (Belhajjame

et al., 2011; Klinkm

uller and Weber, 2021), how-

ever, current MaaS platforms are still limited in

this aspect.

• User Privacy: One of the main issue with exist-

ing MaaS platforms is related to user privacy. Pri-

vacy policies of platforms such as Whim, Tran-

sitApp and Moovit are publicly available to en-

lighten the user about storage and potential use

of her data. However, these policies are not easy

to understand and sometimes provide conﬂicting

information (Cottrill, 2020). More importantly,

these policies are limited to the responsibilities of

MaaS service providers, who store the user data

on third-party servers, but do not extend to the

companies maintaining these servers who are ac-

tually responsible for securing the data (Cottrill,

2020). Hence, the actual impact of existing MaaS

privacy policies is fairly limited.

In addition to the above mentioned user-centric

limitations, features enabling the personalisation in

existing MaaS platforms need further improvements

as well. For example, if a user prefers to use her bi-

cycle or personal car in combination with public or

shared modes of transport, she should be provided

with this option. Existing multimodal route planners

need to be improved to make “true” multimodality a

feasible option.

Some existing MaaS platforms such as Ecomode

and Moovit propose carpooling as a mode of trans-

port. Moovit takes into account the needs of users

acting as passengers and users acting as drivers sep-

arately. However, a user who uses carpooling both

as driver and passenger is left out. Putting in place

an efﬁcient carpooling system is a challenge in itself

(partly due to the complexity of ride-matching algo-

rithms (Zafar et al., 2022)), but integrating carpooling

with other modes of transport while taking into ac-

count varying preferences of passengers and drivers

as well as the contextual information (such as if the

passenger has luggage with him and if the driver’s car

has enough space to safely store it during the trip) fur-

ther increases the complexity. There is a need to look

for technical solutions in the form of efﬁcient opti-

misation algorithms which would integrate carpool-

ing as an attractive choice in the multimodal mobil-

ity space. Having discussed some user-centric and

feature-centric limitations in personalisation in exist-

ing MaaS solution, in the next section, we discuss

how these limitations can be resolved to make MaaS a

personalised, secure, ﬂexible and trustworthy choice

for daily use.

3 IMPROVING

PERSONALISATION IN MaaS

As discussed before, personalisation refers to present-

ing the tailored content to the user, according to her

preferences, without explicit demand. Within MaaS,

where mobility services are provided by different ser-

vice providers, public and shared transport infrastruc-

ture is highly dynamic, and user preferences vary over

time, personalising the information in real-time does

not happen without challenges. In this section, we

discuss some of these challenges and possible solu-

tions to overcome them.

3.1 Integration of Dynamic User

Preferences

Existing MaaS solutions take as input the user data

(such as age, gender, preferred walking distance, does

the user own a bicycle etc.) at the time of registration

to set up user proﬁle which is used to ﬁlter out in-

compatible routes. Some of these data (like preferred

walking distance) deﬁne individual user preferences

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

356

whereas other data (such as age, gender etc.) deﬁne

preferences for a category of users. The user cate-

gorisation can help in understanding preferences at a

higher level, however, in order to improve the large-

scale acceptability of MaaS, individual user require-

ments and preferences need to be better incorporated

into the system (Durand et al., 2018). In addition,

user preferences are susceptible to change over time

and they need to be updated in order to recommend

optimal routes and/or payment options to each user.

One way to take into account individual user pref-

erences is for the MaaS platform to observe the user in

his daily routine and collect data about his trips using

sensors (such as GPS and accelerometer) embedded

in his smartphone. These data can be processed to ex-

tract the user’s regular movement behaviour and auto-

matically learn his preferences (such as his preferred

departure time, preferred mode of transport, number

of times he changes the mode of transport etc.).

There exist numerous approaches to process GPS

and accelerometer data collected from the smart-

phones (Zheng, 2015; Hemminki et al., 2013; Zhang

et al., 2023). Within the context of MaaS, the user’s

movement data could be processed in real-time in an

online fashion by considering the incoming data as

a stream (Gaber, 2012). Although stream data min-

ing has been applied to numerous use cases, there ex-

ist certain challenges, such as handling delayed in-

formation, two-phased (ofﬂine and online) data pro-

cessing, analysing complex data, etc. which need to

be resolved in order to apply data stream mining to

real-world applications (Krempl et al., 2014). Some

of these challenges can be overcome using adap-

tive multi-agent system based methodologies, as dis-

cussed in (Grachev et al., 2020).

Once the GPS and accelerometer data is processed

to compute the user’s regular movement behaviour,

his travel preferences can be extracted from this in-

formation. The Recommendation Systems research

community has made huge contributions in model-

ing and updating user preferences while incorporating

user feedback (de Gemmis et al., 2009). For exam-

ple, collaborative-ﬁltering based techniques consider

user ratings given to various items by different users

and tries to group users with similar interests so as

to recommend similar items to them. Content-based

recommender systems look at the content of the items

being recommended and the attributes describing an

item, and check if they match with user interests. It

has been observed that incremental preference learn-

ing, where the preferences are updated in real-time is

better than batch processing (Gallacher et al., 2013)

for highly dynamic use cases (such as MaaS). Hence,

such methods need to be incorporated within MaaS to

learn user preferences in real-time and personalise the

service provided.

3.2 Proactive Route Recommendations

Once the user preferences have been extracted from

the user’s travel data, they can be employed to recom-

mend optimal routes to the user. Personalised MaaS

should recommend such routes proactively, without

any explicit demand from the user (as discussed in

Section 2.2). From the user’s regular movement be-

haviour, it can be anticipated that he/she needs to take

a trip, between an origin and a destination, at a given

time. Then, various possible routes between origin

and destination can be ranked according to the user’s

preferences.

Recommender systems use ranking models to ei-

ther calculate the rank (score) of a particular item

(Score-based ranking) or use supervised machine

learning where the score function is learned using

some labeled training data (Learn-to-Rank approach)

(Zehlike et al., 2022). Although these methods pro-

vide accurate recommendations, they might lead to

“overly accurate” recommendations over time (Mc-

Nee et al., 2006). The problem is that, generally, the

recommender systems are developed to be accurate

and over long-term they tend to recommend similar

items which the user has used in the past.

This approach is problematic, especially for

MaaS, since the objective of MaaS is to promote sus-

tainable modes of transport and gently “nudge” the

user to change her behaviour. If the route recom-

mender system integrated within MaaS keeps making

recommendations similar to the user’s actual and past

behaviour (which might not be sustainable enough),

it would not be useful. Hence, from time-to-time

MaaS route recommendation system needs to make

unexpected and novel but relevant recommendations

to promote sustainable options. This kind of recom-

mendation is referred to as serendipitous recommen-

dation (Ziarani and Ravanmehr, 2021). The challenge

in implementing such serendipitous route recommen-

dation system is the potential trade-off between what

the user needs at a given moment depending on her

past behaviour and preferences, and what and how

often novel recommendations can be made to change

her behaviour. This presents an interesting research

opportunity for the MaaS community.

3.3 Privacy-Aware Data Storage and

Processing

While learning user’s regular movement behaviour,

extracting his/her preferences and recommending op-

Personalisation in Mobility-as-a-Service: Where We Are and How to Move Forward

357

timal routes, his/her privacy should not be put aside.

With this privacy-centric vision for MaaS, the ques-

tion of data storage and processing presents a huge

challenge. One solution is to store (with user’s con-

sent) and process the data on third-party servers.

However, this does not fully respect the criteria of

user privacy and also suffers from latency issues. A

better approach for large data storage and processing

is based on Mobile Edge Computing (MEC) (Mach

and Becvar, 2017). MEC, a standard edge computing

architecture, is different from traditional cloud com-

puting since, with MEC, the data is processed in close

proximity (in terms of mobile network topology) to

the device which collects the said data. The idea is to

exploit the capabilities of nearby (to a source device)

mobile IoT-enabled devices to collectively process the

data and send it back to the source device. This leads

to improved latency and faster overall processing but

there still exist certain privacy concerns (Ranaweera

et al., 2021).

For MaaS applications, MEC provides an interest-

ing alternative to store and process huge amounts of

data (Xie et al., 2022). However, MEC is a relatively

recent concept and comes with its own challenges

such as optimised resource allocation, transparency,

etc. (Ahmed and Rehmani, 2017). Overcoming these

challenges present the objectives of recently launched

EU funded (EMERALDS, 2023) project.

Recently, Federated Learning (FL) based methods

have also been proposed to resolve the problem of

user data privacy in MaaS (Chu and Guo, 2023). FL is

a type of machine learning approach which is specif-

ically designed to preserve data as well as model pri-

vacy (Zhang et al., 2021). In FL, the machine learning

model is trained locally on the device which collects

the data. In this manner, the data and the model are

kept private and only the results are shared with other

devices through a central server (in case of centralised

FL) or directly over the network (distributed FL). Al-

though interesting, FL based approaches are not with-

out problems (Lyu et al., 2020) which require further

investigations, especially in case of MaaS.

3.4 Data Availability, Standardization

and Sharing

Finally, one of the major challenges in personalis-

ing mobility services is the lack of available real-

time data (Maas, 2022). The data collected by differ-

ent mobility service providers is not made available

since most of these service providers are private en-

tities and start-ups which have their ﬁnancial needs

in mind, and sharing their data with their competitors

does not help their business. As mentioned in the re-

port published by the House of Commons Transport

Committee of the U.K. Parliament (House of Com-

mons Transport Committee, 2018), it is the responsi-

bility of the public authorities to get private businesses

onboard and incentivize them to share their data with

each other in order to develop an intelligent mobility

ecosystem.

In order to share the data collected by different

partners and promote interoperability, data standards

need to be set up and common application program-

ming interfaces (APIs) need to be developed to pro-

cess the collected data (Polydoropoulou et al., 2020).

Such APIs should provide access to real-time data

about the state of road trafﬁc, occupancy and loca-

tion of all public and shared modes of transport avail-

able, users’ location and travel-related data etc. to

all the involved stakeholders in a secured manner. At

this stage, the responsibilities of all public and private

stakeholders about data sharing and data ownership

should also be speciﬁed. The availability, standard-

ization and sharing of real-time data between different

service providers will help in improving the feature-

centric limitations discussed above.

Furthermore, it is also possible that the available

mobility services evolve over time with the addition

of new forms of mobility proposed by new mobility

partners. For their easy integration within existing in-

frastructure, micro-service architecture for data stor-

age, processing and sharing could be a viable option

(Dragoni et al., 2017; Maas, 2022).

4 CONCLUSION

Mobility-as-a-Service (MaaS) has grown into a

promising solution integrating multimodal transport

services, including public and shared modes, onto

a single platform with the possibility of searching,

booking and paying for using such mobility services.

It aims to promote sustainable mobility, change peo-

ple’s travel behaviour over time, and reduce their de-

pendence on personal cars. To become as convenient

as personal cars, MaaS needs to propose personalised

mobility services tailored to the needs of each indi-

vidual user.

In this paper, we presented a brief literature review

on the aspect of personalisation in existing MaaS re-

search and MaaS platforms. Exploring various deﬁ-

nitions of MaaS, we found that most of them describe

it as a personalised service. This review then goes on

to highlight current trends on personalisation within

MaaS and lists some existing limitations related to

integrating user preferences, user feedback and user

privacy. Overcoming these limitations in the context

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

358

of MaaS comes with certain challenges. The paper

presents technical solutions which need to be stud-

ied by the research community and integrated within

MaaS platforms. For example, instead of only relying

on static user preferences to ﬁlter routes, MaaS plat-

forms should also include their dynamic aspect. User

preferences should be learnt and updated over time as

the user continues to interact with the platform so as

to make appropriate recommendations whenever re-

quired. Trade-off between recommendations accord-

ing to user behaviour and sustainable mode choices

needs to be further studied so as to nudge the user

towards environment friendly mobility options. In-

stead of storing and processing user data on third-

party servers, system architecture inspired from Mo-

bile Edge Computing or data processing algorithms

based on Federated Learning should be integrated

within MaaS to store and process user data while pre-

serving user privacy. The proposed solutions present

exciting future research directions for improving per-

sonalisation within MaaS.

REFERENCES

Ahmed, E. and Rehmani, M. H. (2017). Mobile edge com-

puting: Opportunities, solutions, and challenges. Fu-

ture Generation Computer Systems, 70:59–63.

Anagnostopoulou, E., Bothos, E., Magoutas, B., Schram-

mel, J., and Mentzas, G. (2018). Persuasive tech-

nologies for sustainable mobility: State of the art and

emerging trends. Sustainability, 10(7).

Anagnostopoulou, E., Urban

c, J., Bothos, E., et al. (2020).

From mobility patterns to behavioural change: lever-

aging travel behaviour and personality proﬁles to

nudge for sustainable transportation. Journal of In-

telligent Information Systems, 54(1):157–178.

Andersson, A., Winslott Hiselius, L., and Adell, E. (2018).

Promoting sustainable travel behaviour through the

use of smartphone applications: A review and devel-

opment of a conceptual model. Travel Behaviour and

Society, 11:52–61.

Arnaoutaki, K., Bothos, E., Magoutas, B., Aba, A., Es-

zterg

ar-Kiss, D., and Mentzas, G. (2021). A recom-

mender system for mobility-as-a-service plans selec-

tion. Sustainability, 13(15).

Arnaoutaki, K., Bothos, E., Magoutas, B., and Mentzas,

G. (2019). Personalization and recommendation tech-

nologies for maas. In 2nd International Conference

on Mobility as a Service.

Belhajjame, K., Paton, N. W., Fernandes, A. A., Hedeler,

C., and Embury, S. M. (2011). User feedback as a ﬁrst

class citizen in information integration systems. In

5th Biennial Conference on Innovative Data Systems

Research, pages 175–183.

Briggs, P., Simpson, B., and De Angeli, A. (2004). Per-

sonalisation and Trust: A Reciprocal Relationship?,

pages 39–55. Springer Netherlands.

Burrows, A., Bradburn, J., and Cohen, T. (2016). Journeys

of the future: Introducing mobility as a service. Tech-

nical report, Atkins Ltd.

Chu, K.-F. and Guo, W. (2023). Privacy-preserving fed-

erated deep reinforcement learning for mobility-as-a-

service. IEEE Transactions on Intelligent Transporta-

tion Systems, pages 1–15.

Cottrill, C. D. (2020). Maas surveillance: Privacy consid-

erations in mobility as a service. Transportation Re-

search Part A: Policy and Practice, 131:50–57. De-

velopments in Mobility as a Service (MaaS) and In-

telligent Mobility.

de Gemmis, M., Iaquinta, L., Lops, P., Musto, C., Narducci,

F., and Semeraro, G. (2009). Preference learning in

recommender systems. In ECML/PKDD-09 Work-

shop on Preference Learning, pages 41–55.

de Oliveira e Silva, R. A., Cui, G., Rahimi, S. M., and

Wang, X. (2022). Personalized route recommendation

through historical travel behavior analysis. GeoInfor-

matica, 26(3):505–540.

Dragoni, N., Giallorenzo, S., Lafuente, A. L., et al.

(2017). Microservices: Yesterday, Today, and Tomor-

row, pages 195–216. Springer International Publish-

ing.

Durand, A., Harms, L., Hoogendoorn-Lanser, S., and Zijl-

stra, T. (2018). Mobility-as-a-service and changes in

travel preferences and travel behaviour: a literature re-

view. Technical report, Ministry of Infrastructure and

Water Management, The Netherlands.

EMERALDS (2023). Extreme-scale Urban Mobility

Data Analytics as a Service. https://www.emeralds-

horizon.eu/. Accessed January 2024.

EU Urban Mobility Observatory (2023). Sustainable

urban mobility plans. https://urban-mobility-

observatory.transport.ec.europa.eu/sustainable-urban-

mobility-plans. Accessed January 2024.

Gaber, M. M. (2012). Advances in data stream min-

ing. WIREs Data Mining and Knowledge Discovery,

2(1):79–85.

Gallacher, S., Papadopoulou, E., Taylor, N. K., and

Williams, M. H. (2013). Learning user preferences

for adaptive pervasive environments: An incremental

and temporal approach. ACM Transactions on Au-

tonomous and Adaptive Systems, 8(1).

Gao, M., Liu, K., and Wu, Z. (2010). Personalisation in web

computing and informatics: Theories, techniques, ap-

plications, and future research. Information Systems

Frontiers, 12(5):607–629.

Georgakis, P., Almohammad, A., Bothos, E., et al. (2020).

Heuristic-based journey planner for mobility as a ser-

vice (maas). Sustainability, 12(23).

Grachev, S., Skobelev, P., Mayorov, I., and Simonova, E.

(2020). Adaptive clustering through multi-agent tech-

nology: Development and perspectives. Mathematics,

8(10).

Hartikainen, A., Pitk

anen, J.-P., Riihel

a, A., R

anen, J.,

Sacs, I., Sirki

a, A., and Uteng, A. (2019). Whimpact:

Insights from the world’s ﬁrst mobility-as-a-service

Personalisation in Mobility-as-a-Service: Where We Are and How to Move Forward

359

(maas) system. Technical report, MaaS Global Ram-

boll.

Hemminki, S., Nurmi, P., and Tarkoma, S. (2013).

Accelerometer-based transportation mode detection

on smartphones. In Proceedings of the 11th ACM

Conference on Embedded Networked Sensor Systems.

Association for Computing Machinery.

Hensher, D. A., Ho, C. Q., Mulley, C., Nelson, J. D., Smith,

G., and Wong, Y. Z. (2020). Understanding Mobility

as a Service (MaaS): Past, Present and Future. Else-

vier.

Hietanen, S. (2014). Mobility as a service - the new trans-

port model? Eurotransport-ITS & Transport Manage-

ment Supplement, 12(2):2–4.

House of Commons Transport Committee (2018). Mobil-

ity as a service : Eighth report of session 2017–19.

Technical report, U.K. Parliament.

Jittrapirom, P., Caiati, V., Feneri, A.-M., Ebrahimighare-

hbaghi, S., Gonz

alez, M., and Narayan, J. (2017). Mo-

bility as a service: A critical review of deﬁnitions,

assessments of schemes, and key challenges. Urban

Planning, 2(2):13–25.

Kamargianni, M., Matyas, M., Li, W., Muscat, J., and Yfan-

tis, L. (2018). The maas dictionary. Technical report,

MaaSLab, Energy Institute, University College Lon-

don.

Klinkm

uller, C. and Weber, I. (2021). Every apprentice

needs a master: Feedback-based effectiveness im-

provements for process model matching. Information

Systems, 95:101612.

Krempl, G.,

Zliobaite, I., Brzezi

nski, D., H

ullermeier, E.,

Last, M., Lemaire, V., Noack, T., Shaker, A., Sievi, S.,

Spiliopoulou, M., and Stefanowski, J. (2014). Open

challenges for data stream mining research. SIGKDD

Explor. Newsl., 16(1):1–10.

Lyons, G., Hammond, P., and Mackay, K. (2019). The im-

portance of user perspective in the evolution of maas.

Transportation Research Part A: Policy and Practice,

121:22–36.

Lyu, L., Yu, H., and Yang, Q. (2020). Threats to federated

learning: A survey. arXiv preprint arXiv:2003.02133.

Maas, B. (2022). Literature review of mobility as a service.

Sustainability, 14(14).

MaaS4EU (2020). Horizon2020 MaaS4EU Project.

https://cordis.europa.eu/project/id/723176. Accessed

January 2024.

Mach, P. and Becvar, Z. (2017). Mobile edge comput-

ing: A survey on architecture and computation of-

ﬂoading. IEEE Communications Surveys & Tutorials,

19(3):1628–1656.

McNee, S. M., Riedl, J., and Konstan, J. A. (2006). Be-

ing accurate is not enough: How accuracy metrics

have hurt recommender systems. In CHI ’06 Extended

Abstracts on Human Factors in Computing Systems,

page 1097–1101.

Melis, A., Mirri, S., Prandi, C., Prandini, M., Salomoni,

P., and Callegati, F. (2018). Integrating personalized

and accessible itineraries in maas ecosystems through

microservices. Mobile Networks and Applications,

23(1):167–176.

Mulvenna, M., Annual, S., and Buchner, A. (2000). Per-

sonalization on the net using web mining. Communi-

cation of the ACM, 43(8):122–125.

Opiola, J. (2018). Deﬁning levels of maas. Trafﬁc Technol-

ogy International, pages 70–71.

Perugini, S. and Ramakrishnan, N. (2003). Personalizing

web sites with mixed-initiative interaction. IT Profes-

sional, 5(2):9–15.

Polydoropoulou, A., Pagoni, I., and Tsirimpa, A. (2020).

Ready for mobility as a service? insights from stake-

holders and end-users. Travel Behaviour and Society,

21:295–306.

Prost, S., Schrammel, J., R

oderer, K., and Tscheligi, M.

(2013). Contextualise! personalise! persuade! a mo-

bile hci framework for behaviour change support sys-

tems. In Proceedings of the 15th International Con-

ference on Human-Computer Interaction with Mobile

Devices and Services, page 510–515. Association for

Computing Machinery.

Ranaweera, P., Jurcut, A. D., and Liyanage, M. (2021).

Survey on multi-access edge computing security and

privacy. IEEE Communications Surveys & Tutorials,

23(2):1078–1124.

Signor, L., Karjalainen, P., Kamargianni, M., Matyas, M.,

et al. (2019). Mobility as a Service (MaaS) and Sus-

tainable Urban Mobility Planning. Technical report,

ERTICO – ITS Europe.

Smith, D. (2019). Navigogo: Scotland’s ﬁrst maas pilot.

Technical report, ESP Group.

Sochor, J., Arby, H., Karlsson, I. M., and Sarasini, S.

(2018). A topological approach to mobility as a ser-

vice: A proposed tool for understanding requirements

and effects, and for aiding the integration of societal

goals. Research in Transportation Business & Man-

agement, 27:3–14.

Xie, H., Song, X., and Zhang, H. (2022). Maas and iot:

Concepts, methodologies, and applications. In Zhang,

H., Song, X., and Shibasaki, R., editors, Big Data and

Mobility as a Service, pages 229–243. Elsevier.

Zafar, F., Khattak, H. A., Aloqaily, M., and Hussain, R.

(2022). Carpooling in connected and autonomous ve-

hicles: Current solutions and future directions. ACM

Computing Surveys, 54(10s).

Zehlike, M., Yang, K., and Stoyanovich, J. (2022). Fairness

in ranking, part ii: Learning-to-rank and recommender

systems. ACM Computing Surveys, 55(6).

Zhang, C., Xie, Y., Bai, H., Yu, B., Li, W., and Gao, Y.

(2021). A survey on federated learning. Knowledge-

Based Systems, 216:106775.

Zhang, D., Chang, Z., Yang, D., Li, D., Tan, K.-L., Chen,

K., and Chen, G. (2023). Squid: subtrajectory query

in trillion-scale GPS database. The VLDB Journal.

Zheng, Y. (2015). Trajectory data mining: An overview.

ACM Transactions on Intelligent Systems and Tech-

nology, 6(3).

Ziarani, R. J. and Ravanmehr, R. (2021). Serendipity

in recommender systems: A systematic literature re-

view. Journal of Computer Science and Technology,

36(2):375–396.

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

360