D5.1 Intent Aware User Interface Model 1
1 Int troductio on
The GAM MBAS projec ct envisions a new class of behaviou ural ‐driven a pplications f for future ge eneration of smart t cities supp porting key domains of urban life s such as pub blic transpor rtation. This vision is realised by developi ng a generic c Internet of f Things mid dleware and d intelligent cloud servic ces which provide adaptive an d predictive information n to people [ [D1.3.1]. The e middlewar re consists o f a set of services for sensing, , processing, , and exploit ting informa ation about how people e behave in an urban context. Work packa age 5 is conc cerned with the develop pment of a g generic user interface se ervice for GAMBAS S application ns and the d design of inte ent ‐aware u user interface es. Intent‐aw wareness me eans that the syst em uses sem mantic infor mation abou ut user beh aviours to p provide perso onalized info ormation
and rec commendatio ons. As the e applicatio on domain of GAMBAS S is public transportat tion, the behaviou urs of intere st relates to the use of p public transp portation and d user interfa ace services relate to transpor rtation and t ravel choice s.
1.1 P urpose
The deliv verable desc cribes the ba asic theoretic c model for the design a nd developm ment of inte nt ‐aware user inte erfaces. For t this purpose , we explore e a novel des sign space fo r novel mob ile travel inf ormation systems encompass ing dimensio ons of perso onal and soc cial travel be ehaviour. Th e key novel ty of our user int erface mod el lies in th he combinat tion of tech hniques to extract user r intents in form of behaviou ural mobility y patterns f from travel histories an nd novel use er interface concepts w which are capable of displayin g and augm enting this i information to serve the e mobility n eeds of trav vellers. In doing so o, our work g goes beyond existing mo obile travel g guides and tr rip planners which mostl ly rely on user ‐agn nostic and ge eneral inform mation. The e exploitation of travel rou utines of use ers for the cr reation of persona lized views o on the trans portation sy ystem is a ne ew approach for improve ed user inter raction in urban sc cenarios.
Figure 1 shows the In ntent ‐Aware e User Interfa ace Service i n the contex xt of the GAM MBAS Middle eware.
The key objectives a nd contribut tions of this d deliverable a are:
Developmen nt of a conc ceptual user r interaction n model wh hich covers different as spects of information needs for t the design o of mobile tra avel guides. The model is based on n a novel
approach a to exploit the relationship between be ehavioural co ontext data a and user inte erfaces to
achieve a the intent ‐aware e user interfa ace concept
Developmen nt of a new interface co oncept for making use of a broade er notion of f context information beyond tim me and locati ion to disco ver and high hlight crowd dedness in th he public
t transport ne etwork Developmen nt of a new c concept for t the persona lization of pu ublic transpo ort systems based on
the t notion o of cognitive maps to ena able travel e experiences w which are cl losely couple ed to the t travellers ac tual behavio our in reality Developmen nt of a socia al interface concepts w hich gives in nsight into s social inform mation of travel t behav viour and hig ghlight socia al relations a among friend ds and like‐m minded peop ple in the t transport sys stems
Prot totype Applic cation
V Validates
Enable es
5 ‐aware Us 4 an Dat
Aut
Inte ent
ser
In
3 Interfaces res a dP
a Mod ro
te
er ted
roce
a ra tio e Priv
ssi
lli e b
acy
n ng le
Adaptive A Data a g
1 Acquisiton
G GAMBAS Com mponents
Servic ce Discove ery and Co ommunic ation
2 M Middlewar re
GAMBAS M Middleware GAMBA AS Approach
Fig gure 1 – GAMB BAS Middlewar re Architecture with User Inte rface Compone ent
In summ mary, the key y contributio ns of this de eliverable are e: the definit tion of an ab bstract User Interface Model (S Section 4), d design sketc hes of user interfaces th hat realise t his model (S Section 4), b behaviour mining m methods for r understand ding and pr redicting use er behaviou r (intent pre ediction me ethods to
discover r routines in n travel beha aviour), as w well as user experience metrics tha at capture th he users’ satisfact tion with pub blic transport t systems.
1.2 Sc cope
This doc cument is t the first del liverable in WP5 and i is concerned d with deve eloping fund damental concepts s and techni iques for int tent ‐aware u user interfac ces. The conc cepts, metho ods and des ign ideas describe ed in this do ocument wil l
be used in n the develo opment of c concrete use er interfaces s for the GAMBAS S application n prototype ( (D5.2). The r result of use r studies to e evaluate use er experience es will be describe ed in D5.3.
With res spect to the e overall GA MBAS system m, the deliv verable exte nds previous s deliverable es of the consorti um ([D1.3.1 ], [D2.1.1]) w with a focus s on human aspects. Wh hile both D1 1.3.1 and D2 2.1.1 deal with me eans to acq uire real‐tim me context data of tra avellers on m mobile devi ices, this de eliverable explores s opportunit ties of how to make e effectively u use of this d data to eng gage users in public transpor rtation and b build intent‐a aware user in nterfaces.
1.3 St tructure
This deli iverable is st tructured as follows. In Section 2w we provide a review of th he state‐of‐t the art in mobile t travel system ms. As part of this anal lysis, we dis scuss existing g mobile ap pplications w which are readily a available on the market, as well as p present an o overview of c current rese arch in mob bile travel This deli iverable is st tructured as follows. In Section 2w we provide a review of th he state‐of‐t the art in mobile t travel system ms. As part of this anal lysis, we dis scuss existing g mobile ap pplications w which are readily a available on the market, as well as p present an o overview of c current rese arch in mob bile travel
relevant t to the use r interface s system, whi ch have bee en defined i in the requi irement spe cification phase o f the projec ct, is discuss sed in Sectio on
6. Finally y, we conclu de this deliv verable with h a short summar y in Section 7.
2 Us ser Interf face Issue es Relate ed to Pub blic Trans sportatio on System ms
The GAM MBAS middle eware and s ervices will be validated d in the cont ext of a pro ototype appli ication in the pub lic transport tation doma in. Although h the exact details of th he validation n plan are s still being worked out (and w will
be publi shed as D6 .1.1) it is c lear that th e prototype e application n will be develope ed for the ci ity of Madrid d, facilitated d by project partners EM MT and ETRA A. In the follo owing we
describe e the Madrid d public tran sport scenar rio and analy yse user inte erface issues s of mobile t transport applicati ions.
2.1 T The Madri d Public T Transport rt System
The com mmunity of M Madrid is loc cated in the e centre of S Spain, covers s an area of f approximat tely 8000 km 2 , and d comprises a population n of about 5 million peo ple, with 3 m million peop le living in th he city of
Madrid. Thus large n numbers of p people from t the surround ding area co mmute on a daily basis a and there is clear demand for r sophisticat ted transpor rt informatio on systems. Although M Madrid has the sixth
1 largest 1 m metro railwa ay in the wo rld, with 13 metro lines and 296 me etro station , the city is suffering
from ah high level of c congestion d during peak h hours [30].
In additi on to the M etro system s, Madrid ha as an extensi ive city bus n network, wh hich covers p practically the who ole city and is s run by GAM MBAS projec ct partner Em mpresa Mun icipal de Tra ansporte (EM MT). Most
lines ope erate everyd day between n 6 am and 1 11.30 pm, w with buses lea aving at inte ervals betwe en 4 and
15 minu utes, depend ding on the time of day y. There are also night b buses, know wn as “búhos s” (owls),
which ru un along 27 d different rou utes. In tota al the EMT fl eet comprise es 2,022 veh hicles. Figure e 2 shows
a repres entation of t the central a nd busiest p part of the M Madrid public transport sy ystem.
Figure 2 – Abs stract Represen ntation of Mad drid Public Tran nsport System.
1 http://w www.citymayo ors.com/trans sport/madrid ‐ ‐metro.html
2.1.1 D Digital Serv vices
A rich di igital ecosyst tem of data and service s exists for t the Madrid t transport sys stem. To kno ow when the next t bus is due, the GAMBA AS partners E EMT and ETR RA have set up an online e service tha at can be accessed d through a w web interfac ce and a mo bile phone. T The web inte erface (show wn in Figure 3) allows users to find bus rou utes, get sche eduled depa arture and ar rrival times a and get estim mated times of arrival (ETA) for r all bus stop ps.
Figure 3 3 – User Interfa ace of EMT Tran nsport Web Ap plication
EMT also o provides a n official iPh one and And droid applica ation with ac ccess to the m most freque ntly used
services (Figure 4 an d Figure 5). Both mobile e applications s provide acc cess to:
A A map of the e Madrid com mmuter netw work (metro and bus line es), location ‐awa are station fi nder, real ‐time jou urney calcula ator, and
w waiting time e and distanc ce calculator .
Figure 4– EMT Android A Application
Figure 5– – EMT iPhone A Application
Third ‐pa arty mobile a applications with extend ed functiona ality are ava ilable for mo ost mobile p latforms. For 2 exam mple, the ‚Bu us Madrid’ application f for Android phones goes s beyond the e EMT applic cation by
providin g personalis sation featur res that allo ow users to define their r preferred b bus stations and bus lines.
Fi gure 6 – Madri id Public Trans portation Syste em (generated from open tra nsport data by y http://openbu usmap.org/)
2.1.2 T Transport Data
An incre easingly rich set of data i is available f for public tra ansport syste ems around the world. C Currently three ty ypes of infor rmation are available fo or the Mad rid public tr ransport sys stem: (I) sta tic route
network k data (i.e. sto ops and rout tes), (II) stati ic time‐table e data, (III) dy ynamic real‐t time bus info ormation (estimat ted time of a arrival = ETA A). Route da ta and sche dules for M adrid are av vailable as o pen data
3 following 4 g the Googl le supported d General T Transit Feed Specificatio on (GTFS) sp pecification and in
2 https:// /play.google.c om/store/app ps/details?id= =es.android.bu usmadrid.apk &feature=rela ated_apps
https:// /developers.go oogle.com/tra ansit/gtfs/refe 4 erence
http://w www.gtfs ‐data a ‐exchange.co om/agency/m madrid/
XML 5 . O Open data is s used by d developers a and compan nies to creat te dedicated d transport mapping services, 6 , such as the Open Bus M Map .
Figure 6 shows a visu ualisation of f the Madrid public trans sport system generated f from openly available data.
2.2 Su urvey of C Commerc ial Mobile e Public T Transport t Applicat tions
A variet ty of smartp phone journ ney planning g application ns exist for Madrid and d many oth her cities worldwi de. Journey planer app ps mostly re ely on static c informatio n, such as p public ‐transf fer time‐ tables 7 , with some (such as th e map syste em provide by Google) using real‐t time inform mation. In addition , there are e mobile ap pplications t that use G PS and acc celerometers s to monito or users’ transpor rtation habit ts to encour rage behavio our changes. . For instanc ce, there are e application ns, which motivate 8 e users to co 9 onsume less f fuel while dr riving .
In order r to clarify fe eatures of e existing publ ic transport apps, we a nalysed a su ubset of all available mobile a applications. As criteria for selecting g application ns for the su rvey we use ed importan ce of the research h project or d download fre equency in t he Android m market. Exce ept for Goog le Now there e are few if any a pplications t that system atically mine e user beha aviour to pro ovide person nalised serv ices. The
summar y of the app lication featu ures is prese ented in Tabl e 1.
App App Static or
Trip R Real ‐
Conte ext Personali isation
Sh haring
name real ‐time
ca alculation time t Visual lization
Resource Informati
based b on
no otificati
Meta aphor
on or rigin and
on o of de estination im mminen
t a arrival
Inter ractive Curre ent user Now .google.com/
Google https://www Real ‐time
Suppo orte
It employ ys the
d, bas sed user cont text to
with h map locat tion can landing/now on Go oole learn its most sup pport be sh hared via
searc ch, freque ent Go oogle
histo ory location s such as
Lat titude
and d hom me locat ion
SeoulBus https://play. Real ‐time google.com/s
curre ent Only rou uting Inter ractive _
tore/apps/de locat ion based on with h map
tails?id=com. current lo ocation sup pport astroframe.s
eoulbus NYC https://play.
static _ _
Only a a static
Bus&Sub google.com/s ma ap is way tore/apps/de
supp ported tails?id=com.
episode6.and roid.nycsubw
aymap
5 http://w wiki.openstree etmap.org/wi ki/Bus_routes s_in_Madrid
6 http://o openbusmap.o org/
8 Live Lon ndon Bus Trac cker, London B Bus Checker, T TubeMap Pro ve,
9 eco:Driv Fiat group drivegai in.com
Automobiles
SG Buses https://play. Real ‐time
Curre ent Only rou uting Inter ractive _
google.com/s locat ion based on with h map tore/apps/de
current
lo ocation
tails?id=com.
iridianstudio. sgbuses
Madrid Real ‐time
Curre ent Only rou uting _ _ _
Metro|B
locat ion based on
us|Cerca https://play.
current lo ocation
nias google.com/s tore/apps/de tails?id=com.
metrodroid. madrid
NYCMat https://play. Real ‐time
Bo oth _
e (Bus & google.com/s inter ractive Subway) tore/apps/de
tails?id=com. m map only
and
densebrain.a ndroid.nycsu bwaymap
Live https://play. Real ‐time
Curre ent Only rou uting Inter ractive _
google.com/s
London locat ion based on
Bus tore/apps/de
Tracker appeffectsuk. bustracker
current
lo ocation
tails?id=com.
One Bus https://play.
Real ‐time x x
Curre ent Routing based Inter ractive _
Away google.com/s
locat ion on curr rent
tore/apps/de
location n
and
tails?id=com.
creati ing
joulespersec
perso nal usbot reminde ers for
ond.seattleb
specific c bus
stop p Table e 1: Feature Co mparison betw ween Transport t Apps
2.3 R Research Topics T
Advance ed interactive e transport applications are now be eing investiga ates by seve eral research h projects in two c categories: (I I) Research t that aims at individual b behaviour m odification a and (II) resea arch that aims at g group behav viour.
2.3.1 P Personal B ehaviour In the fo ollowing, ap plications th hat mine and d exploit pe rsonal beha viour will be e described. In more technica al sense, the se mobile ap pplications s sense the co ontextual info ormation an nd perform r reasoning
based on n the collect ed contextua al informatio on, for perso onal usage.
PEIR [1] , GreenSaw w[6] and Ub iGreen[2] ar re mobile a applications that motiva ate users to o shift to sustaina ble behavio our. PEIR (Pe ersonal Env ironment Im mpact Facto r) [1] uses GPS data o of mobile
phones to collect us sers’ locatio on changes, and by usin g an accele rometer sen nsor data an nd HMM‐ based a activity class sification, de etermines u users’ transp portation m mode. UbiGr reen [2] pro oposes a persona l ambient di isplay on mo obile phones s to give fee dback about t transportat tion behavio ours. This causes p personal aw areness abo out transport tation activi ty and reinf forces user c commitmen t to eco‐
friendly behaviour. J Jigsaw [6] is another ap plication tha at provides a a user interf face for refle ecting on persona l behaviour.
Another research fo ocus, in this a area, is “aut omatic trave el guides”. M Most of these e projects ha ave been designed d for tourists s’ usage, suc h as [10, 11] ]. PECITAS [1 10] propose a route reco ommendatio n feature on a dat tabase that contains tra nsit network k (busses an nd roads) inf formation of f Bolzano, It aly. They profile r outes based d on user sel lection and a are thus abl e to provide e a ranked li ist routes in the user interface e. Garcia et al. [11] prop pose a system m that perfo orms recomm mendation, r route genera ation and customiz zation by id dentifying P OI (point o f interest) o of tourists’ routes with hin public t ransport. RouteCh heckr[15] pe ersonalizes the routing g process o of mobility impaired p pedestrians through collabor ative multim modal annot ation of geo ographical d data. The an notation is based on tw wo parts, namely direct anno otation of g geographical data by us sers and ac cquisition of f directly ob bservable informat tion.
2.3.2 G Group Beha aviour Another category o of applicatio ons focuses on interac ctive service es for group ps of travel llers and combine es group beh haviour mode els with adva anced user in nterfaces. PROCAB B [6] uses GP PS traces for r taxi drivers s to underst tand their ro oute prefere ences and le earn their behaviou ur. It tries to o incorporate e users’ beh avioural atti itude into a route recom mmendation and thus recomm end better routes to ta axi drivers. M Moreover, it t enables re easoning abo out context‐ ‐sensitive
users’ be ehaviour, wh hich includes s actions, pre eferences an d goals.
Tulusan et al. [16] u use the EcoG Gain smart phones app to monitor corporate d driver behav viour and through providing us sers with eco o ‐feedback t o reduce fue el usage.
2.4 C onclusion n
In sum, there are no applicat ions that pr rovide adva anced intera ctive service es by minin ng public transpor rt behaviour . Furthermo re user inter rface approa aches of exis sting or prop posed applica ations do not mak ke use of per rsonalisation n and adapti on and use of context p parameters is s primarily l imited to
location and proxim ity. This pro ovides concre ete evidence e that there is ample spa ace for inno ovation in the follo owing areas:
Incorporatin ng novel con text parame eters into tra ansport appl lications (for r example re elating to q quality of av vailable trans sport options s) User interfa ace persona lisation and adaptation based on individual a and group b behaviour profiles
User interfa aces that p rovide user rs with data a and infor rmation abo out future t transport s scenarios (e. .g. tomorrow w, tonight) so o to support effective tra ansport plann ning ahead o of time.
These a reas of inno ovation dire ectly relate t to the resea arch objecti ves of WP5 5 as specifie ed in the Descript tion of Work .
3 In ntent ‐Awa are User Interfac e Model
In this s ection we d describe a ne ew user inte erface mode l for intent‐ aware servic ces and app plications. The mo del consists s of four lay yers (as rep presented in n Figure 7) where each h layer repr resents a particula ar user inter rface design aspect. Whi ile the unde rlying conce pts and mec chanisms are e generic and not application s specific, the model is tai lored for do mains where e user behav viour plays a key role. In that s sense the us ser interface model refle ects publicat ion transpor rtation as ap pplication do omain for GAMBAS S.
Figure 7 – La ayered User Int erface Model
Each lay er adds a pa articular desi ign aspect; t he final user r interface is the result o of combinati on of the design a spects of (so ome or) all fo our layers. Ea ach layer com mbines three e elements:
‐ Data (about transportati ion and user mobility) ‐
User Interfac ce elements (i.e. widgets s and contro ls) and ‐ A Algorithms ( for generati ng user inter rfaces from d data)
From bo ottom to top these layers s introduce th he following g elements:
‐ T The transpo rt layer (TL) focuses on the represe ntation of p ublic transpo ortation syst tems and forms f the ba aseline for a ll layers abo ove. The tran nsport layer i s where mo ost currently available route finding g and mobile e transport a applications ( (see Section 2.2) operate e.
‐ T The Quality of Transpor t layer (QoT TL) focuses o on paramete rs that affec ct a user’s e motional
e experience o of a public transport sy ystem, such as timeline ss and over rcrowding. T This layer entails e elem ments that h help users m make transp ort decision ns that go b beyond simp ple route
f finding. ‐ T The Persona al Behaviou r Layer (PB L) focuses o on the repr resentation (data and v visual) of
people’s mo obility pattern ns, in particu ular with res pect to the u use of public transport sy ystems.
‐ T The Social Be ehaviour Lay yer (SBL) focu uses on the representati on of mobili ity patterns o of groups of o people (i.e e. friends in terms of soc cial networks s).
A featur re of the GA AMBAS user interface m model is that t the constit tuent layers are inheren ntly data‐ driven; t they create a and exploit data to enab ble novel us er interfaces s and provid de novel serv vices. For this purp pose, the lay yers are stac cked in a wa y such that the knowled dge inferred about the b behaviour of trave llers is incre ementally ex panded from m the lowes t to the mo st upper lay yer. While th he lowest layer is concerned w with informa ation about t the transpor rtation netw work (which is available from the transpor rtation netw work provide er), the high her layers e enrich this b basic inform mation and integrate
persona l trip histor ies and crow wd data to highlight th he role of h umans as tr ravellers in different contexts s. In the foll owing, we w will describe the design principles fo or the user i interface fo r each of these lay yers in detai l.
3.1 T Transport Layer
3.1.1 M Motivation n and Object tives The tran nsport layer serves as a foundation of the inten nt ‐aware use er interface system. Thi s layer is concerne ed with key information about the tr ransport net work such a s the locatio on of bus stat tions, the routes o of bus lines a nd time‐tab les. Classical route plann ners and trav vel recomme enders heavi ly rely on this data a to compu ter their se rvices, and most public cly available transport a applications don’t go beyond t this layer.
In GAMB BAS we use basic transp port data in t two ways: t o identify tr ansport opti ions and to generate user inte erfaces that allow peopl e to make b basic transpo ort decision, such as wh en and whe ere to get from At o B.
3.1.2 U User Interf face Service es This laye er supports o one fundame ental service :
D. The layer com mputes and visualise rou utes and time etable inform mation to ge t from A to B B.
User Inte erface Servic ce 1 (Route F Finding): The e user specifi ies a start lo cation S and d a destinatio on
3.1.3 U User Interf face Sketch es Route fi nding is a w well ‐covered use case in n transporta tion system s and most applications s provide effective e and usable visual repre esentations. A As far as the e Transport L Layer is conce erned we do on’t see a need to o deviate fr rom establis shed practic ces. Figure 8 and Figu ure 9 provi ide potentia al design
approac hes for this l layer that sh ow how rout te informatio on can be vis sualised on m mobile devic ces.
Figure 8 – Typical visu ualization of th he transportatio on layer (adapted fro om http://live.t transloc.com )
Figure 9 – Simple met tro map visuali ization on smar rtphone (ad dapted from htt tps://play.goog gle.com/store/ /apps/details?i id=us.pandav.N NYC)
3.1.4 D Data and A Algorithms We mak ke use of st atic and dyn namic transp port data o n this layer. . Static data a describes t the basic structure e of the tran nsport netwo ork in terms o of the locatio ons of all sta ations as wel l as the time etables of all bus li ines. This en ables us to l learn about the trajecto ries of differ rent bus line es (which sta ations are
linked to o each other r), possible in nterchange f facilities as w well as to inf form about d departure an nd arrival times of f buses at pa articular stat tions. Dynam mic data enr iches the sta atic data wit th up‐to‐date e reports about th he real‐time status of bu uses. These reports inclu ude data abo out the serv vice conditio n (e.g. in service o or out of ser rvice) of a b us, the locat tion of buse s, and the e estimated tim me of arrival l (ETA) at bus stat ions. While t this layer do oes not direc ctly involve prediction ta asks, the ava ailable data becomes relevant t for predictio on on the hig gher layers o of our model l.
3.2 Q Quality ‐of‐ ‐Transpor rt Layer
This laye er comprises s data, algor rithms and d design elem ents to prov vide users ri ich opportu nities for making effective tra ansport deci sions based on quality of transport t data. Onlin ne and mob bile route planners s give people e informatio on about rou ute options w where route recommend dations are based on objective e criteria su uch as time of departure e, estimated d time of ar rival, travel duration, a nd costs.
Howeve r, such metr rics neglect important a aspects of th he travel exp perience and d ignore tha at people often m ake travel d ecision base ed on emotio onal and su bjective crite eria. Transpo ort systems research has unco overed ‘pain n points’ suc h as overcro owding and delays that m make people e less incline ed to use public tr ransport. Qu uality of tra ansport is a new conce ept that refe ers to emot tional and s ubjective aspects of the user’s s transport e experience th hat affect pe eoples’ trans port behavio our. We use the term ‘quality o of transport’ ’ (QoT) to re fer to user‐c centric meas ures of the t transport exp perience. Qo oT can be based on n objective d data (measu red by senso ors) or on su ubjective fee lings (collect ted for exam mple from user opi nions publish hed on twitt er feeds).
In line w with the cont text acquisiti ion work in W Work Packag ge 2 we focu us on crowd level as prim mary QoT attribute e. In dense u urban areas such as Mad drid and Lon ndon overcro owded statio ons, buses a nd trains are an i nherent exp perience of t travelling, w which is one of the reas ons many tr ravellers pre efer non‐ shared m modes of tr ansportation n such as ca ars. Overcro owding also interferes w with more tr raditional travel ex xperience m metrics, as pr rolonged wa aiting times caused by c crowded bus s lines migh t lead to longer jo ourney time s than expec cted. In the past, data a about crowd levels has n not been ava ailable to users (al lthough stat istical data f from surveys s has been t taken into a ccount by tr ransport ope erators in
scheduli ng transport t options). T The objective e of this laye er is therefo re to realise e user interfa aces that expose Q QoT measure es to users a nd helps use ers to make t travel decisio ons based on n QoT param meters.
3.2.1 N Novel Conc cepts This laye er introduces s three new c concepts:
Quality ‐o of ‐Transport t: a term tha at refers to measures o of a transpo rt system th hat affects t he travel experien nce (such as overcrowdin ng and delays s).
Quality ‐o of ‐Transport t Map: a tran nsport map a augmented w with contextu ual QoT data a. Quality ‐o of ‐Transport t prediction: the ability to o predict Qo oT attributes for a future e point in tim me, based
on an an nalysis of hist torical patte rns.
3.2.2 U User Interf face Service es This laye er supports t wo user‐inte erface servic es:
D. The layer com mputes and visualise QoT T attributes for routes to o get from A to B.
User Inte erface Servic ce 2 (QoT Se earch): The u user specifie es a start loc ation S and a destinatio on
User Int terface Servi ce 3 (QoT T Temporal Exp ploration): T The layer en ables users to explore past and predicte ed future QoT T attributes o of the transp port network k.
3.2.3 U User Interf face Sketch es The prim mary user in nterface elem ment of this layer is the e QoT map. T This is a visu ual represen ntation of QoT attr ributes of th he transport network. Fi igure 10 and d Figure 11 s show two ex xamples of how QoT
10 vi sualises cro wd ‐level on route segm ments by va rying the thicknes ss of line se egments; Fig gure
paramet ters can be visualised. F Figure
11 visu ualises crowd d ‐levels on bus stations s by using c circles of varying s sizes. Future e work will e xplore the ad dvantages a nd disadvant tages of each h alterative d design.
Figure 10 0 – An example e of how a QoT T map could be realised. ( ad dapted from ht ttp://content.s tamen.com/ze ro1 )
Figure 1 1 – An example e of how a QoT T map could be realised. (adapted f rom http://hai irycow.name/c commute_map /map.html)
3.2.4 D Data and A Algorithms The QoT T layer relies on quality d ata that can come from two potentia al sources.
‐ f from direct measureme nts of crow d levels pro vided by the e context ac cquisition fra amework and/or a
‐ f from ticketin ng informatio on provided by transport t operators. As part of this wor k package w we will inves stigate ways s to integrat te and mash h ‐up data fr rom both
sources. This will en nable us to p predict crow wd levels for a future dat ta and time from histor ical data. The desi ign of user in nterfaces for r predicted f future scena rios is one o of the key as spects of the e work on this laye r.
Data rep presentation and predict ion algorithm ms for this la ayer are desc cribed in deta ail in Section n 5.
3.3 P ersonal Behaviour B r Layer
The Per sonal Behav viour Layer is designed to create a a personaliz zed experien nce of the t transport network k by anticipat ting travel pr references (i ntents) and mobility pat terns of indi ividual users .
3.3.1 M Motivation n and Object tives The nee d for a pers onalized tra vel informat tion system is motivated d by the fact t that people es’ travel behaviou urs are highl y different in n daily life. S Specifically, w we can obser rve that ther re exist highl y diverse travel ha abits
a) acro oss time and d space and d b) for diffe erent passen ngers. For in nstance, at w weekdays users oft ten rely on p public transp ortation for commuting to work, wh ile leisure ac ctivities and shopping are usua ally more pop pular at the weekend. At t the same t ime, users liv ving in certa in districts o only use a
well ‐def ined subset of the tran nsport optio ons offered by the ent ire transpor rt network. A global immutab ble view on the transpo rt system, w which covers s all station a and routes o of the transp portation infrastru ucture, is the erefore not c consistent w with the way people use a and access t the transpor t system. Howeve r, current tr ansportation n informatio on systems a re heavily d esigned on t this principle e, lacking any beha avioural ‐driv ven personal ization conc cepts. As a co onsequence, , people are often burde ened with
too muc ch or irreleva ant informati on in their c current trave el situation. The goa l of this laye er is to reve rse the exist ting practice e of how tra vellers are in ntegrated in nto travel
informat tion systems s. Key to this s approach is s the discove ery and expl oitation of t the personal routines of trave llers. Based on the know wledge abou ut personal travel behav viour, person nalized view ws on the
transpor rt system can n be created d to give peo ople useful in nformation, w which is rele evant to thei ir current and futu ure mobility needs. By establishing a close em motional rela tionship of passengers with the transpor rt system, th he public tr ansportation n system be ecomes muc ch more attr ractive and provides incentive e for its freq quent usage, , since passe engers will p perceive them mselves to b be an integra al part of the bus n network.
3.3.2 N Novel Conc cepts This laye er introduces s three nove l concepts:
Personal l Travel prof file: this laye er records ho ow people u use a public transport sy ystem over time and builds a statistical m model of how w an individua al uses publi ic transport. A user trans sport profile contains informat tion about h ow often an d when an in ndividual use es bus stops and bus rout tes.
Personal l Travel Beh haviour pred diction: a use er transport t profile is u used to pred dict future t transport behaviou ur (intent) a nd identify e elements of the route ne etwork (such h as bus stop ps, bus lines) ) that are most rel evant to use ers in a given situation.
Cognitiv e Map: a co ognitive map p is a visual representat ion of the t ransport net twork that h highlights element ts of the tran nsport netwo ork that are m most relevan nt to a user i n a given con ntext (deter mined by current t time, curren t location an nd transport profile).
3.3.3 U User Interf face Service es
The pers sonal behavi our layer sup pports two u user ‐interfac e services:
User Int terface Servi ice 4 (Person nalized Trav vel Recomme endation): U Using time a and place as s context paramet ters the use er is getting g access to a cognitive e map that summarizes s the most relevant informat tion for mak ing travel ch oices releva nt to a user in a proactiv ve manner.
User Inte erface Servic ce 5 (Tempo oral Explorat tion of Perso onalized Trav vel Recomm mendations): The user specifies s a time and d/or place of f interest an nd is shown a Cognitive Map that s summarizes t the most relevant t information n for making travel choice es at that tim me and/or pl ace in the fu uture.
3.3.4 U User Interf face Sketch es The prim mary user e lement of t his layer is the cognitiv ve map. Figu ure
13 s how two possible design ske etches for a cognitive m map. Figure 12 exploits s traces of mobility to indicate frequent t routes in t the transpor rt system an nd is thus a direct repr esentation o of a user’s t transport profile. Figure
12 and Figure
13 is on a higher r level of ab bstraction an nd shows ho ow a transpo ort profile is s used to highlight t elements o of the transp port system that are of high relevan nce to a use er while hidi ing those
element ts that are of f less relevan nce (given th he user’s cur rrent tempo ral and spat tial context). By using predictio on it is possi ble to pinpo int transport t elements t that will be o of importanc ce an hour, a a day or a week int to the future e. Because o of the much‐ ‐reduced co mplexity of a cognitive map, it can easily be augment ted with rea al ‐time depar rture and arr rival informa ation, and Q oT data, wit hout overloa ading the user wit h too much information . Future wor rk will explor re the advan ntages and d isadvantage s of each alterativ ve design.
Figure F 12 – Des sign Sketch for Cognitive Map 1 (adapted from m http://conten nt.stamen.com )
Figure F 13 – Des sign Sketch for Cognitive Map 2
3.3.5 D Data and A Algorithms Key to e enhancing th e degree of personalizat tion of publi ic transport systems is a a good under rstanding of the u users’ mobili ty patterns. Specifically, , the cogniti ive travel m ap sketched d above is b uilt from
knowled dge about th he user’s typ pical bus usa age patterns . We will th erefore inte egrate data i inference algorithm ms to extract t the users’ t travel prefer rences from past bus trip ps. For this p urpose, we w will make use of c ontext data collected by y the data a cquisition fr ramework on n their mobi ile devices. T This data
encomp asses inform mation about t the detaile d journeys p people make e over the co ourse of a w week. This includes the source and destina ation station ns of their jo ourneys, as well as chan nges at inte rmediate encomp asses inform mation about t the detaile d journeys p people make e over the co ourse of a w week. This includes the source and destina ation station ns of their jo ourneys, as well as chan nges at inte rmediate
A popular route is one e that the us ser frequentl y travels on and, theref fore, is of high im portance fo r him in hi s daily life. The popula ar routes w will
be explo oited to ide ntify the geograp hical space inside which h the user’s mobility ta kes place. H However, sin nce the popu ularity of routes m may
be diffe erent for a g given day an nd time, this s constitutes s dynamic k knowledge w which will evolve o over time. Th herefore, the e patterns w will also have e an associat tion to the d days and tim mes when
they bec come relevan nt. This will allow us to a adapt the us ser interface to different t structures i n human life, whe ere especiall y working d uring the we eek and leis sure activitie s at the wee ekend are a relevant discrimin nating facto r. Temporal patterns in the travel h habits can al so be used as triggers t to decide
when us sers should b be presented d travel‐relat ted informat tion. Therefo ore, the user r trip historie es will be also ana alysed to dis cover popul ar times at which a den nse concentr ration of sim milar departu ure times can
be used then to info orm the con ntrol logic a nd visual appeara nce of the m mobile travel application.
be o observed. T he predictio on results w will
More de etails about d data and algo orithms for t the personal behaviour la ayer can be f found in Sec ction 5.
3.4 So ocial Beh aviour La ayer
The Soc cial Behaviou ur Layer is designed to o create aw areness of travel prefe erences and mobility patterns s within a use er’s social ne etwork.
3.4.1 M Motivation n and Object tives Social no orms are the e standards w we use to jud dge the appr ropriateness of our own actions, and it is now widely a acknowledge ed that makin ng pro‐envir ronmental so ocial norms m more visible e is an impor rtant part of the c challenge of f promoting g sustainable e behaviour r. In laborat tory studies s and more applied, practical l settings, pr roviding peo ople with evi idence of w hat others a around them m are doing has been shown to o have a sign nificant effec ct on behavio our, for exam mple the use of public tra ansport.
Online s ocial networ rks such as F Facebook an d Google+ a are already b built upon th he social met taphor of sharing a and discover ring content t among frien nds. Howeve er, the appli cation of thi is paradigm to public transpor rtation has n not been exp plored yet. T The social be ehaviour laye er encompas sses design e elements to estab lish social no orms in publ ic transporta ation and ma ake people a ware of how w their social network
uses pub blic transpor rt. This will e encourage tr ravellers to r rely on publi c transport m more often based on the incre eased value o of the travel information n system.
3.4.2 N Novel Conc cepts The Soci al Behaviour r Layer introd duces three novel conce epts.
Social Tr ravel Profile: : a social tra avel profile e encompasse es informatio on about ho ow often and d when a social gr oup uses bu s stops and b bus routes.
Social Tr ravel Behavi iour Predicti ion: a social travel profi le is used to o predict be ehaviour (int tent) and identify elements of f the route network (su uch as bus s tops, bus lin nes) that are e most relev vant to a group of f users in the e future.
Social Tr ransport Ma ap: a social t ransport ma ap is a visua l representa ation that hig ghlights how w a social group is using a publ lic transport system.
3.4.3 U User Interf face Service es The soci al behaviour r layer suppo orts two user r ‐interface se ervices:
User Int terface Serv vice 6 (Socia al Transport t Usage Dis scovery): Us ing time an nd place as s context paramet ters the user r is getting ac ccess to a so ocial transpor rt map that s summarizes how the use er’s social network k is using a pu ublic transpo ort system.
User Inte erface Servic ce 7 (Tempor ral Exploratio on of Social T Transport Us sage): The us ser specifies a time of interest and is show wn a social transport m map that su mmarizes h ow the use r’s social ne etwork is
predicte ed to be using g a public tra ansport syste em at the ind dicated time .
3.4.4 U User Interf face Sketch es The prim mary user int terface elem ment of the s social behav iour layer is the social tr ransport ma ap. Figure
14, Figur re 15 and Fig gure 16 show w possible de esigns of a so ocial transpo ort map. Figu ure 14 repre sents the overlap in the areas of interest (defined by the reachab ble travel des stination wit thin short tim me) for a number of users (in this case jus st two). Figu re 15 shows a social ma p that is gen nerated by o verlaying
transpor rt traces of members o of the user’s s social netw work. Figure e 16 shows a social ma ap as an aggregat tion of cogni itive maps o f individual u users (again, , in this exam mple just two o users). Fut ure work will expl ore the adva antages and disadvantag ges of each a lterative des sign.
Figure 14 – Envisioned d design of a So ocial Transport t Map 1.
Figure 15 – Envisione ed design of a S Social Transport t Map 2 (ada apted from sou rce: http://ma cwright.org/ru unning/)
Figure 16 – Envisione ed design of a S Social Transport t Map 3
3.4.5 D Data and A Algorithms The soci ial layer targ gets the tra vel behavio ur of group s who have strong soci al ties amon ng them. Therefor re, it has the e most comp prehensive d ata requirem ments conce rning the am mount of use er ‐related informat tion. From a a data proce essing point of view, the e social layer r represents s an extensio on to the persona l layer descr ribed previou usly. While t the personal layer extrac cts the users s’ travel patt terns, the social la yer deals wi ith a set of d different, bu ut possible r related trave el patterns. S Since this da ata is not
available e on the dev ice of the qu uerying user, , it has to be e retrieved fr om the user r’s friends. Th herefore, this laye er includes re emote acces ss mechanism ms to coord inate the on n ‐demand da ata retrieval process. Howeve r, the sharin ng of travel data obeys precise priv vacy restrict tions as defi ined by use r ‐specific policy ru ules, where u users state t heir prefere nces of who o else may ac ccess their d ata. Hence, only that
data can n be retrieve ed which bel ongs to user rs who have defined acc cess permissi ions for the querying users.
In order to explore t the travel ha abits of socia al friends, in nformation a about their r regular route es will be required d. In order to o address th is issue, we exchange o nly a summa ary of the m most significa ant travel habits b etween the users. This is done to a avoid transfe erring low‐le vel trip histo ories so as t to reduce the proc cessing and c communicat ion load on the mobile d devices. The e retrieved tr ravel data co onsists of the user r’s identity a and the rou tes most oft ften taken b by this user, as inferred from the p prediction algorithm ms on the pe ersonal laye r. The social l layer will th hen use app ropriate met trics to deci de which routes h have to be d displayed. Fo or instance, this could be those ro outes having g similar sou urces and destinat ions in rela ation to the e trips of a a user. The e routes can n then be used to en nrich the transpor rtation netw work with clu ues on the fr riends’ trave el behaviour and will be visualized a as part of the user r’s interface.
More de etails about d data and algo orithms for t the personal behaviour la ayer can be f found in Sec ction 5.
4 Sy stem Arc chitectur re
Over the e course of the project, , we will de evelop a mo obile travel s system whic ch integrates s the key concepts s of the use r interface m model prese ented before e. In this sec tion, we intr roduce two different views on n this system m. First, we d discuss an int ternal persp ective which h represents the key com mponents required d to implem ent our syst tem. Then, w we present an external perspective e and talk a bout the
integrati ion of the int tent ‐aware i nterface syst tem into the e overall arch hitecture.
4.1 U User Interf face Syste em
The arch hitecture of our user in nterface syst tem follows s a model‐vi iew ‐controlle er design pa attern as shown in n Figure 17. It is based o n a set of dif fferent funct tional compo onents, whic ch manage a accessing, processi ng and disp laying of co ntext data. Instances of f the archite ecture will be e deployed on every mobile u user device p participating in the GAM MBAS system . The data fl ow among d different com mponents in our sy ystem is sho own in Figur re
17. In the e following, we present the compo onents of ou ur system architect ture more in n detail.
Figur re 17 – Archite cture of the Us ser Interface Sy ystem
4.1.1 P Presentatio on The pres sentation lay yer impleme nts the inter rface to the u users and is focused on t the represen ntation of various travel infor mation. Hen nce, it corre esponds to t the view pa art in the m model ‐view‐c controller pattern. The presen tation comp ponent provi ides interact tive user inte erface eleme ents that all ow users to requ est and ret trieve desir ed travel in nformation. The interfa ace will be e designed to make informat tion access a and cognitive e processing g as easy as p possible, usin ng the conce epts describe ed earlier (cognitiv ve map, soci ial map, etc. .). The interf face will not t only includ de static tran nsport data, but also behaviou ur ‐driven inf formation ab bout the tra avel habits o of each user. . Therefore, we like to p point out that the view looks different for r every user in the syste m. The comp ponents will also enable e users to change p preferences settings, suc ch as the priv vacy policies stating the d data sharing g restrictions .
Figur re 18 – User Int terface System m Data Flow Dia agram
4.1.2 C Controller The con troller medi ates betwee en the prese entation and d the data p roviding com mponents. R Regarding this inte eraction, the e controller is intended to act in a two‐way d irection, dep pending on whether changes in the user r interface a are reactive or proactive e in nature. On the one e hand, it in ntercepts events f from the p presentation component t to retriev ve and man nipulate the e current vi ew. This represen nts a reactive e strategy fo or adapting t the user inte erface since t the interacti ion is initiate ed by the user. On n the other h hand, the co ontroller also o continuous sly observes relevant co ntext chang es of the user to t trigger autom matic change es in the use er interface. I In this case, the controlle er anticipate es further needs o of the user t that may re equire specif fic informati ion and a p roactive stra ategy is cho osen. For
instance e, by noticing g that the us er is heading g towards to o a bus statio on, the arriva al times of ne ext buses at this st tation can be e delivered t o the user.
4.1.3 D Data Infere ence The data a inference c component consists of a a set of algo rithms for m mining behav vioural patte erns from the data a available i n the GAMB BAS system. The compo onent contin nuously mon nitors change es in the user’s c ontext to g ather inform mation abou ut the user’ s daily rout ine. The tra aces of past t context changes will be proc cessed to fit t a stochastic c model whi ich summari zes the past t user behav viour in a statistica al representa ation. The st tochastic mo odel is an ab bstract repre esentation o of the user b behaviour and muc ch more co mpact than the origina l user trace s. Based on the model, , travel patt terns are derived which are th hen used to personalize the user int erface view and create p proactive tri ggers for deciding g when to sho ow travel‐re lated inform mation.
4.1.4 D Data Acces s The user r interface sy ystem is hea avily data‐dri iven and use es knowledg e from vario ous sources t to inform the mob bile travel ap pplication. Fo or this purpo ose, a compo onent is req uired to pro ovide mechan nisms for accessin g local and r remote data sources. Loc cal data refe ers to the con ntext inform mation direct ly sensed on the m mobile device e, comprisin g real‐time c context data , e.g., the cu urrent locatio on, as well a s historic informat tion in form of past user r traces. In c contrast, the e remote dat ta sources ar re distribute ed among
different t stakeholde ers in the GA AMBAS syste em. They inc lude service s and databa ases of the t transport network k provider, e xternal aggr regation serv vices to infe r crowd‐leve el statistics a as well as tra avel data
from the e friends in a a social netw work. Since t he access to o remote dat ta is highly se ensitive from m privacy perspect tive, this com mponent bui ilds on the p privacy and s security mec hanisms tha t have been outlined in D3.1.1 1.
4.1.5 E Experience e Logging
The expe erience logg ing compone ent comes w with a set of tools to mon nitor pattern ns in the usa ge of our mobile t travel applic cation. Basic cally, it logs s data abou t how the user interac cts with the e system, requests s and consu mes differen nt types of information n. The comp onent is des signed to ru un in the backgrou und of the a application a and intercep t all relevan nt interaction ns observed at run‐time . We will exploit t the logs as p part of our e evaluation pr rocess to ass sess the valu ue of our int terface conc cepts. For instance e, based on a an analysis o of the intera action logs, w we can unde erstand how w much time the user