Socializing Public Transportation

Using Situational Context in Public Transportation to Get in Touch with People

Around You

Roman Roor, Olga Birth, Michael Karg and Markus Strassberger

BMW Group Forschung und Technik, Hanauer Str. 46, Munich, Germany

Keywords:

Mobility Patterns, Public Transport, Socializing, Micro Social Network, Recommendation.

Abstract:

Unexpected delays or long traveling times lead often to people who get bored while using public transport.

Whereas some might use their travel time for work or even enjoy the silence, there are still many people that

would welcome an opportunity to use the spatio-temporal proximity to get to know others or meet with friends

that are in the same train. This paper introduces a novel, smartphone based concept to bring people together

while using or waiting for public transport. Based on location and personal preferences, suggestions can be

made for getting in touch with nearby persons. We propose a recommendation system, which identiﬁes the

concrete public transport vehicle and compares the preferences with other users to create recommendations

about people nearby who are also traveling in or waiting for a public transport vehicle.

1 INTRODUCTION

With the megatrend of increasing urbanisation around

the world, more and more people are expected to use

public transportation in the near future. Due to unex-

pected delays or problems on public transport lines,

frequent use of transportation modes like subways,

trains or buses inevitably leads to planned or even un-

expected waiting times at stations, bus stops or sim-

ilar. In particular trips including multiple vehicles

or modes of transport can be subject to unexpected

waiting times due to the possibility of missing a con-

nection trip. Waiting times of up to 30 minutes are

not uncommon, especially when traveling late at late

evening. Commonly, most people waiting at e.g. train

stations today just try to ﬁght their boredom by play-

ing smartphone games or similar. But the upcom-

ing boredom can not only be observed when wait-

ing on e.g. a train or plane, but also during a trip.

While some people might use their travel time for

work or even enjoy the silence, there are still many

people in e.g. a train that are bored and would wel-

come an opportunity to use the spatio-temporal prox-

imity to other bored persons to get to know other peo-

ple or meet with friends that are in the same train.

The missing link to bringing together people with the

same temporal interests is the uncertainty of the in-

tents of the people around, respectively in the same

train. Once this inhibition threshold is overcome by

connecting people with same interests travel or wait-

ing times can be used for socializing instead of be-

ing bored. Due to overall penetration of smartphones

and internet connectivity – even in the subway and on

planes – people are always connected to each other

anywhere. This is an excellent precondition to bring

traveler together. Since built-in smartphone compo-

nents can be used to track nearby traveler, sugges-

tions can be made about friends or other traveler in

the same vehicle, that are available for chatting. This

paper introduces a novel, smartphone based concept

to bring people together while using or waiting for

public transport. Based on location and personal pref-

erences, suggestions can be made for getting in con-

tact with nearby persons. We propose a recommenda-

tion system, which compares location data and prefer-

ences with other users and creates recommendations

about people nearby. This opens up a variety of new

possibilities. For instance, a person missing a con-

necting train can easily experience waiting times of

up to 2 hours until the next train to the destination.

Finding people in a similar situational context could

not only lead to less boredom during waiting time, but

also to shared rides to a common destination. Gener-

ally speaking, our approach enables the use of random

occasions to meet people.

113

Roor R., Birth O., Karg M. and Strassberger M..

Socializing Public Transportation - Using Situational Context in Public Transportation to Get in Touch with People Around You.

DOI: 10.5220/0005492201130118

In Proceedings of the 1st International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS-2015), pages 113-118

ISBN: 978-989-758-109-0

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

2 RECOMMENDATION SYSTEM

Almost every second person living in Germany has

a smartphone today (data from May 2014, (Statista,

2014)), and generally, the number of smartphone us-

age is increasing. Most smartphones have various

kinds of sensor components built in, which can be

used for location tracking. With rising smartphone

usage, services like texting, social network apps and

app usage in general, provide a good basis for loca-

tion and preference based recommendation systems

(Lenhart, 2012). So there is no need for additional

devices, which could be a barrier.

Also the fact that (long distance) public transport

is often delayed or even spontaneously canceled con-

ﬁrms that a recommendation system for bored trav-

eler is a promising concept (AFP/woz, 2010), (Bahn,

2014). In addition, different statistics show that ride

sharing is growing in North America and in Eu-

rope (Chan and Shaheen, 2012), (BlaBlaCar, 2012).

Establishing a real time ride sharing network could

also reduce CO2 pollution. Mobile apps like Tinder,

Lovoo, Grindr and Cuddlr demonstrate that geolo-

cation based matchmaking has a huge potential. 11

times a day on average, people use such kind of apps

and spent between 7.2 and 8.5 minutes per login using

the app (Bilton, 2014). Combining the idea of match-

making with an recommendation system for travelers,

based on short geolocation distance with an intelligent

algorithm for ﬁnding potential conversation partners

with the same transportation route and preferences, a

solution for potential boredom during trips can be cre-

ated.

Figure 1: Workﬂow Recommendation System.

We propose a recommendation system based on

the workﬂow shown in ﬁgure 1. First of all the user

needs to set some preferences according to the at-

tributes of the desired user group. As soon as the sys-

tem knows the user, the person can plan routes while

the system continuously keeps track of the user’s lo-

cation. By gathering information the system can es-

timate the current and target state. By merging the

states and evaluating the differences, recommenda-

tions are made.

2.1 Micro Social Network

One focus of the proposed idea is to connect people

considering their personal preferences. To make it

work, we need to know who the people are. So ev-

ery user needs to register at the service and partici-

pate within the network. It is also conceivable to of-

fer registration using OAuth / connect APIs of major

social networks to be able to collect more information

and preferences and reduce conﬁguration costs for the

user. Intelligent suggestions can only be made if the

service knows the bias of the users.

The intention is not to create another social net-

work where the users maintain their proﬁles addition-

ally to the existent major social networks but only

provide a minimum set of their bias and one or more

photos. This micro social network is more provider

driven to be able to build graph based databases about

social contacts according to users who are friends and

potential matches while the user is being tracked.

The way to connect others is to press ”Like” on

target proﬁle. Only after the user behind the targeted

proﬁle also has pressed ”Like” they are connected and

can start to communicate. The targeted user does not

know in advance if he/she was liked.

2.2 Habits

Humans tend to pattern daily actions into sequences

which they repeat at particular times in particular

places (Townsend and Bever, 2001). Furthermore,

most people spend a large proportion of their time at

just a few locations, of which the home of a person

as well as his/her workplace(s) have a high impact

(Herder et al., 2014), (Song et al., 2010), (Gonzalez

et al., 2008). Such regularities in persons’ mobility

patterns can be used to predict likely next locations

of a person as well as the corresponding route and

mode of transport that is taken to the destination. To

perform such predictions, there exist various methods

based on heuristics (Froehlich and Krumm, 2008),

different variants of Bayesian networks (Simmons

et al., 2006) or similar. Most of the approaches use

motion tracking data from smartphone sensors and

apply different clustering algorithms about commonly

visited locations, such as a persons home and work-

VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems

114

place. Furthermore, statistical models about spatio-

temporal relations between locations are generated to

enable predictions of likely next locations of a person

as well as the corresponding departure time, mode of

transportation and the used routes.

Figure 2: Two different persons traveling the same section

at the same time.

Such information about habitual trips of a person

can be used to intelligently create recommendations

for possible meetings with people, the user is con-

nected with. Figure 2 illustrates a situation in which

such a recommendation could be useful. Person A

usually leaves his/her home at around 10:51am each

Saturday to visit his/her grandmother by train. Person

B, to whom person A is connected by the micro social

network (as described in section 2.1), joined the same

train earlier at 10:42am with the same destination. In

this case, our system could inform both persons about

a possible meeting opportunity. The same scenario

would not only be thinkable for routes with the exact

same destination, also routes with partial similarities

are possible candidates for meetings.

3 IDENTIFICATION OF

CONTEXTUAL DATA

To make our proposed idea work, we need to know

two additional things about the person using our sys-

tem. On the one hand, we need to have a better under-

standing of the localization of a person who is using

public transportation. On the other hand, our system

should be aware of the remaining time the particular

person will spend traveling in or waiting for bus, train

etc. First we will describe, how we can identify the

precise localization of a person using different types

of public transport. Further we will take a closer look

on how we manage to get the remaining travel time.

3.1 Localization

As we suggest a solution, which notiﬁes a person

about possible passengers or friends who are also

traveling in a particular public transport vehicle and

are also looking for a conversational partner, our sys-

tem should be able to identify the concrete means

of transport in which those two people are traveling.

For our solution, we assume that we don’t need to

equip the trains or buses with further hardware, to

make the positioning better and that it can work with

the sensors of a smartphone. Based on that assump-

tion, current systems lack of functionality, to identify

the precise public transport a person is using. Those

systems map only low-level sensors to a generalized

high-level behaviour, e.g. walking, running or driv-

ing, without the concrete distinction between, which

public transport vehicle, route and station is actually

used (Partzsch and Foerster, 2012), (Patterson et al.,

2003), (Reddy et al., 2008). The methodology is

mostly based on a classiﬁcation model made out of

mobility patterns based on historically collected data.

While it is easy to distinguish between motorized and

non-motorized states, it is difﬁcult to differentiate be-

tween the various motorized states.

Proximity sensing, trilateration or dead reckoning

are the the most common technologies, when it comes

to identify the position of a person. During a trip, a

person can use different kinds of transportation types,

including ”on foot”, using motorbike, car or public

transport. The focus of our proposed system is to

identify the concrete public transport vehicle a person

is currently using as well as the public transport states.

Limitations in areas where for example no GSM or

GPS connection exists, makes it almost impossible

to identify the concrete public transport a person is

currently traveling with just by using the smartphone

sensors. That case applies especially in underground

transportation systems. But we need that kind of in-

formation, to be able to understand whether someone

is traveling in a particular subway using which di-

rections and heading towards which station. Current

research also shows, that as soon contextual data is

used, the success rate is much higher (Patterson et al.,

2003). They showed an increase on the prediction ac-

curacy from 60 per cent to 78 per cent with just addi-

tion additional bus stops, bus routes and parking lots.

Thus, we built a knowledge-based system to over-

come these limitations and used context-information

to identify the current public transport vehicle, the di-

rection and station a person is heading to.

For our knowledge-based system we built a data

base with a representation of a digital map of all kinds

of public transport and stations of Munich, Germany.

Second, we considered to add additional contextual

information like entrances, stairs, elevators to identify

a concrete station, but decided to stay with the stations

as the only information for better public transport ve-

hicle identiﬁcation, because it was the only source

of contextual data that could be exported from map

content databases, which are publicly available (e.g.

SocializingPublicTransportation-UsingSituationalContextinPublicTransportationtoGetinTouchwithPeopleAround

You

115

Google Maps). we name that contextual information

keypoints. The main focus of keypoints is the rela-

tions to each other. This information was mapped to

a directional graph, wheres every edge is able to rep-

resent multiple lines. Figure 3 shows such an sam-

ple keypoint representation of Munich, wheras the

map represents the real-world example and the graph

shows the abstract representation of that real-world

example. The yellow points represent tram stations,

magenta are bus stations and blue are subway stations.

The small blue points are the entries of a particular

subway station. If a traveler is close to a subway en-

trance, the keyoint logic would return a high proba-

bility for that subway station.

Figure 3: Abstract graph visualization between multiple

keypoints.

After the initialization of the digital map, in-

cluding the contextual information described before,

we’re tracking, by using smartphone sensors, all tar-

geted key-points, if they are approached by the trav-

eler. When the probability of one targeted key-point

falls, then it is sorted out and the key-point resets it-

self to its initial state. As soon as the traveler gets

close enough to the targeted key-point, it switches to

the ”check if stops”-state (compare ﬁgure 4). At this

point a decision can be made, and all lines that are as-

signed on that edge are possible public transportation

candidates. With the addition of live departure times

at a station, the likelihood of one line can be shifted

to the most recent line departed. Without the live in-

formation these are excluded as soon as they split up

their route. This closeness factor has about four times

the size to the nearby factor because the traveler is

moving in a faster pace and the algorithm needs to

determine if the traveler will indeed slow down and

stop. If the traveler stops, an ”on public” notiﬁca-

tion is sent, as shown in ﬁgure 4. The next targeted

key-point is notiﬁed. This is done by an own state to

prevent multiple ”on public” notiﬁcations. Hereby it

prioritizes the current possible lines but if additional

lines stop at the current station, or there are transitions

available, they are also tracked in the background, be-

cause the traveler might change the vehicle.

Figure 4: States and transitions in the keypoint logic.

Using this localization methodology, we’re now

able to recognize with a success rate of 95 per cent the

concrete public transport a person is in, the concrete

line and direction and at which station the person is

currently. If various people use our system, we can

identify two people (at least) who are traveling in the

same public transport. Having additional information,

like if the two people are friends, we can inform these

people that there is someone they know in the same

train or bus etc. Then they can either ride the rest

of the trip together. Thus, we have built a technical

solution for bringing friends, who are unknowingly

riding the same public transport, together (compare

ﬁgure 5).

Figure 5: Friends, who are using the same public transport

can travel together (Scenario A).

VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems

116

3.2 Deviation Detection

The methodology we use to detect deviations allows

us to understand where and for how long a person will

need to wait on the chosen route. Basically, a route

deviation can happen in two dimensions, active and

passive. An active deviation can for example occur

if the user didn’t left at a predetermined time. Also

the user can leave at the right time, but getting lost

or going too slow and thus missing the connection.

Another example for an active deviation is, if the user

accidentally takes the wrong public transport vehicle.

These deviations are capable of being inﬂuenced so

the user can be actively warned or informed about any

anomaly to undertake countermeasures.

The passive deviation occurs for example if a pub-

lic transport delay or even cancellation happened. If

the user is traveling by car, then a trafﬁc jam has also a

inﬂuence on the deviation. The only deviation which

is important for our presented system is, if the public

transport is late and the person has to wait for or in it.

To identify such a deviation, we need the target and

current state in regards to public transport. The target

state is the time and place a public transport should

be according to the schedule. This is what our sys-

tem gets as the route parameters, which needs to be

observed. The current state of the public transport is

the time and place the vehicle is in real-time. If we

have access to real-time data about the whereabouts

of various public transports, we can easily identify

deviations by comparing the target with the current

state. If we don’t posses this real-time data, we have

to work with the identiﬁcation of the whereabouts of

the public transport using our knowledge-based sys-

tem as described above. This can only work, if we

have many users who are using our system, so that

we know the current states about any public transport

without having access to real-time data.

Figure 6: Strangers who want to travel with the same train

can get notiﬁed if the train delays or gets cancelled so that

they can arrange an alternative to travel to the desired desti-

nation by using car sharing or taxi (Scenario B).

As can be seen in ﬁgure 6, person A and person

B want to travel with the same train, which leaves at

10:51 and arrives at 11:01. Person A has another trip

before the train and as person A arrives, he/she notes,

that the desired train has a delay. Our recommenda-

tion system doesn’t only know about the train delay

but also about the habits of person B. Thus, the sys-

tem assumes, that person B wants to travel to a des-

tination using the same train. As both, person A and

B, had pressed ”like” before, the system suggests that

both can either share a car, which is near the train sta-

tion or a taxi to split costs.

4 USE CASE

In the sections before, we have described various

technological solutions for different kinds of prob-

lems. Now we want to bring all of them together and

describe, how they can interact in order to make our

proposed system work.

After a user registered on our system and entered

preferences concerning the potential co-travelers, the

user can now select a route using our system, which

should have at least one public transport route seg-

ment. If the user selected one of the presented route

suggestions, we can start tracking to get a better un-

derstanding of the personal mobility patterns. With

the help of our knowledge-based system described

above, we are able to identify the concrete public

transport vehicle, direction and station, at which a

person is currently heading to. Based on the pref-

erences towards other people and the connection to

existing social networks, we know who the person

would like to travel with. If our knowledge-based

system for identifying the concrete means of public

transport found that, at least, two people are travel-

ing with the same public transport vehicle and if the

set preferences towards each other are positive, we

can show that there is someone in the public trans-

port whom both probably would like to travel with.

If we know, with the access towards various social

networks, that those two people are friends, and each

other pushed the ”like” button in our system when was

asked for meeting that friend, we can notify them.

We also collect information about behavioural

patterns, like when does the user goes where using

which type of transportation. After a while, we have

learned the typical destinations and traveling habits

and thus can predict future trips. In the event of a

public transport deviation, e.g. a train delay, we know

that there are people who either also wanted to take

the same public transport vehicle or are heading to-

wards the same destination. Depending on the set

preferences (either by pressing ”like” in our system

or via social network connection), our system would

pop suggestions like using a car or taxi together in

order to pay less.

SocializingPublicTransportation-UsingSituationalContextinPublicTransportationtoGetinTouchwithPeopleAround

You

117

5 CONCLUSIONS

One possibility to launch this intelligent mobility rec-

ommendation system would be in combination with

a cloud backend and a mobile app. The system

logic and the algorithms would reside in the back-

end. Therefore it is possible to create thin mobile apps

for different mobile OS’ with no need for expensive

high-power CPUs in mobile phones. The mobile app

should be only a client for the cloud backend, which

is more like a GUI. The main task is to transmit infor-

mation for computation, receive the results as well as

notiﬁcations and visualize these information. Due to

this approach no additional devices are needed and the

acceptance barrier would be lowered. ”Yet another

app!” could be a barrier for the user, if this idea is not

going to be integrated into existing navigation apps.

Another app must be installed and conﬁgured (regis-

tration, settings etc.) before usage. Moreover privacy

is also a valid reason to be concerned to start using

this system. The user has to let the system record the

tracking information, compute mobility patterns and

store personal preferences and relations to other peo-

ple, who are connected with the user.

If the user is willing to let this happen the system

can support him to get a new experience about spare

time usage and socializing while traveling. Even in

reducing costs and saving the environment in case of

delayed or cancelled public transports by sharing non

public transport vehicle costs.

We also think about to use the described subsys-

tems in an inter modal navigation system. The focus

would be on detecting anomalies of different trans-

portation types, which should be used to arrive at the

destination point. The current time and the schedules

of the different transportation vehicles must be con-

tinuously observed and deviations have to be detected

in real time to warn the traveler or do a recalculation,

if there is a high risk to miss a connection. This pro-

vides a high potential for efﬁcient route planning and

time saving especially in todays fast progression of

urbanization.

REFERENCES

AFP/woz (2010). Im winter durchschnittlich 114 zugaus-

ﬂle pro tag bei der bahn. Hannoversche Allgemeine

Zeitung.

Bahn, D. (2014). Pnktlichkeitsentwicklung 2014.

Bilton, N. (2014). Tinder taps an age-old truth. The New

York Times, page E1.

BlaBlaCar (2012). Ridesharing statistics in europe.

Chan, N. D. and Shaheen, S. A. (2012). Ridesharing in

north america: Past, present, and future. Transport

Reviews, 32(1):93–112.

Froehlich, J. and Krumm, J. (2008). Route prediction from

trip observations. Technical report, SAE Technical Pa-

per.

Gonzalez, M. C., Hidalgo, C. A., and Barabasi, A.-L.

(2008). Understanding individual human mobility

patterns. Nature, 453(7196):779–782.

Herder, E., Siehndel, P., and Kawase, R. (2014). Predict-

ing user locations and trajectories. In User Model-

ing, Adaptation, and Personalization, pages 86–97.

Springer.

Lenhart, A. (2012). Teens, smartphones & texting. Pew

Internet & American Life Project.

Partzsch, Ina; Duerrschmidt, G. M. O. and Foerster, G.

(2012). Comparison of vehicle-based and smart-

phone generated acceleration data to determine mo-

tion states of passengers.

Patterson, D., Liao, L., Fox, D., and Kautz, H. (2003).

Inferring high-level behavior from low-level sensors.

In Dey, A., Schmidt, A., and McCarthy, J., editors,

UbiComp 2003: Ubiquitous Computing, volume 2864

of Lecture Notes in Computer Science, pages 73–89.

Springer Berlin Heidelberg.

Reddy, S., Burke, J., Estrin, D., Hansen, M., and Srivas-

tava, M. (2008). Determining transportation mode on

mobile phones. In Wearable Computers, 2008. ISWC

2008. 12th IEEE International Symposium on, pages

25–28.

Simmons, R., Browning, B., Zhang, Y., and Sadekar, V.

(2006). Learning to predict driver route and destina-

tion intent. In Intelligent Transportation Systems Con-

ference, 2006. ITSC’06. IEEE, pages 127–132. IEEE.

Song, C., Qu, Z., Blumm, N., and Barab

asi, A.-L. (2010).

Limits of predictability in human mobility. Science,

327(5968):1018–1021.

Statista (2014). Anzahl der smartphone-nutzer in deutsch-

land in den jahren 2009 bis 2014 (in millionen).

Townsend, D. J. and Bever, T. G. (2001). Sentence compre-

hension: The integration of habits and rules, volume

1950. MIT Press.

VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems

118