Trust and the Internet of Things

croutonsgruesomeΔίκτυα και Επικοινωνίες

16 Φεβ 2014 (πριν από 3 χρόνια και 3 μήνες)

50 εμφανίσεις

Trust and the Internet of Things
Jon Robinson,Ian Wakeman,Dan Chalmers,and Ben Horsfall
Department of Informatics,University of Sussex,UK
Abstract.We present the design of the shoppingLense and its surrounding infra-
structure.The shoppingLense is designed to allow open collaborative tagging of
patterns within the environment,and for users to browse the environment using
augmented reality.We use trust to control the presentation of the patterns and
anchors within the augmented reality,building upon trust relationships that are
dynamically created and maintained between the users of the system,yet maintain
privacy.
We evaluate our approach through an experimental evaluation of our prototype
in a programcommittee experiment,where the shoppingLense is used to browse
the virtual reviews attached to submitted papers.We show that performance is
significantly improved using a structured presentation based upon trust over an
unstructured randompresentation.
1 Introduction
The world is increasingly a virtual place.In the developed world,it is rare to
find a business that does not have a virtual presence on the Internet.But if we
wish to visit the web-sites relevant to our current location,in most cases,we
must transcribe URLs into the browser,notwithstanding the increasing use of
Bluetooth advertisements and the availability of QR Codes [1] in Japan.One
promising approach to browsing reality is through the use of augmented reality,
where graphical anchors on patterns within displayed video provide links off to
the relevant web-sites and other services.
Our vision is of a world where everything and anything can be used as a pattern,
and where everyone can both create patterns,and can add anchors to any pattern.
If such a free-for-all is possible,how are people to distinguish the patterns and
anchors that are of interest to them?Our view is that this should be through the
use of trust relationships,and ensuring that each pattern and anchor can have a
pseudonymous owner so that the identities can be bound within statements of
trust.We use pseudonyms to provide privacy to the users,whilst allowing ac-
countability in the event of malfeasance.Further,there should be an additional
level of indirection relating pseudonyms to long-lived groups,so providing the
shadow of the future over the actions taken under cover of the pseudonyms,as
described in [2].
In this paper we outline our basic architecture for an open platform to act as
a pattern repository,and to allow annotation of patterns with anchors.We have
built a prototype system to prove our design,based around a standard Tomcat
web server,and a browser based on the ARToolkit which we have named the
shoppingLense.Finally,we describe a short experiment to both demonstrate our
concepts,and motivate our use of trust in the interface.
2 Overall Design
The ARToolkit from[3] is a mature toolkit for overlaying graphics upon a stream
of video.The toolkit uses fiducial patterns,which are pre-compiled into recog-
nisable patterns that the ARToolkit can recognise within frames of video.These
patterns currently use a thick black border,but can use any pattern within the
border such as in figure 1.
Fig.1.An example ARToolkit pattern
Given that any pattern within the ARToolkit border can be compiled into a recog-
nisable pattern,the question arises as to who should be allowed to compile pat-
terns.In a similar manner to the level playing field of the web,we believe that
anyone should be able to compile patterns,and that it should be left to the user
which patterns they choose to display.Similarly,the anchors associated with each
pattern can be installed by anyone,again relying on the user to choose which an-
chors to display on their device.
If anyone can install patterns,and anyone can associate anchors with a pattern,
how are we to prevent the AR-enhanced web falling prey to the same tragedy of
commons as happened to the spam-deluged
1
email system?Our approach is to
build trust mechanisms into the system.We provide a priori policies based on the
1
We define spam in our system as unwanted or irrelevant patterns and anchors,rather than
unsolicited email.
group memberships of the pattern or anchor owner,and a more dynamic mech-
anism based on ratings.By providing an extra level of indirection between the
identity and the trust,we provide more privacy,ensuring that only the informa-
tion necessary to make trust judgements is communicated.
Our aim is to deploy a robust system within a local shopping precinct,allow-
ing the public to download and install a viewer on their mobile phone [4].The
patterns are printed and placed in shop windows or any other suitable location,
and anchors are attached to the patterns.The patterns are uploaded and compiled
through the precinct’s web-site,and anchors are likewise attached to the com-
posed patterns within the web-site.We have developed a web-services based sys-
temto provide the pattern registry,trust model,pseudonymmanagement,and an-
chor and rating database.Importantly,each upload is identified with a pseudonym
held within the web-site,and so each pattern or anchor is digitally signed and
owned by an identity e.g.the shopkeepers will own the patterns located within
their shop windows,whilst customers will own the anchors taggged to the pat-
tern.
Our design is based on the centralised scheme described in [2].Each pattern or
anchor is cryptographically linked to a pseudonymous or real identity,using the
standard approach of an identity consisting of a public/private key pair,and sign-
ing the digital material with the private key.Each identity can then be a member
of groups known to the user.
The web-site provides the group abstraction,where each group is managed by
one or more identifiable and real people.Users apply through the web-site to
the group manager to have their pseudonym accepted as a group member,and to
provide the necessary digitally signed signature that this is so.Since each group
has their own criteria for deciding whether management of the group is appro-
priate,the actual mechanism is supplemented by several side channels,such as
telephones and email.Users can choose to provide group membership creden-
tials along with the ownership signature over the pattern or anchor when the pat-
tern/anchor is downloaded.
Users can formpolicies based on group memberships,using the credentials of the
owning identity to order the trust placed in each pattern and anchor.An example
of a simple policy can be seen in figure 2.The default trust level is 0,representing
no information.If multiple policies apply,then the following rule is applied:if
any are negative then the lowest trust value is taken (to highlight risk),otherwise
the highest positive value is used.This assumes that the magnitude of a reported
value comes about through building multiple experiences weighted by the level of
trust experienced (through exposure to risk),so that larger values convey a weight
of experience.In this way,the complete policy specification can be applied to all
tags but those groups which have little view on some tag have no impact.In this
way terms of reference for each group do not need to be defined.The memberOf
function tests the membership of the owner of the object,whilst the rating
function retrieves the set of ratings that match the membership argument.We can
then apply functions such as max,min and mean to this set of ratings.
It is clear that not all users would write the policies from figure 2 in a text editor
and maintain them.However,some common patterns of behaviour could be built-
in with switches,and additional policies might be written and shared in the same
way other small scripts,plug-ins and applications are shared (e.g.firefox addons
or iphone applications).Much of this user-specific information could be extracted
from other data,further simplifying policy construction:Traders’ associations
will control their group membership and users may choose to place default levels
of trust in such organisations (the example of 1.0 may not be the value chosen)
and particularly familiar organisations.Groupings of friends could be established
if memberOf(NorthLaineTradersAssociation)
then trust:= 1.0
if memberof(IansFriends)
then trust:= 1.0
if memberOf(EthicalConsumers)
then trust:= 0.8
if max(rating(EthicalConsumers)) > 0.8
then trust = 0.5
Fig.2.Example policy declaration
from social networking sites,phone books etc.Similarly,consumer and interest
groups would be found by the user through other activities and membership might
cause a trust level question to be raised.We have not yet designed a process of
exploring available groups,but many familiar possibilities from the web,e.g.
search,keyword indexes,tags,maps of common interest,and suggesting groups
based on aggregate data are all possible.
A user would thus interact with the systemin the following way:
1.On entering the shopping precinct,they would be greeted with signs remind-
ing themthat they can download and install the application.
2.The application would then contact the precinct’s website,and download all
the patterns and anchors,along with their associated meta-information,such
as the certificate chains.
3.The user can install policies about which pseudonyms and groups are trusted,
using simple membership predicates.
4.The application applies the policies over the downloaded patterns and an-
chors,generating an ordering based on trust.
5.The user in strolling through the precinct can use the application to recog-
nise the pattern from the video feed and display relevant anchors.Which
patterns and anchors that are displayed on the screen is limited to the most
trustworthy items under consideration,as in figure 3.
6.By clicking through into a conventional list and view screen (figure 4),the
user can see all the anchors associated with a given pattern,ordered by their
trust.
We also provide the capability for registered pseudonyms to add ratings to a pat-
tern and anchors,allowing more complex policies based on transitive trust re-
lationships,and the ratings given by a particular group.So,having experienced
a shop a user may feel moved to provide a rating – this is not new and many
sites rate particular goods,manufacturers or shops from user feedback.To this
we have added a mechanism to describe trust in those ratings through the expe-
rience of individuals known by the user and also through individuals who have
similar concerns to the user as expressed in group memberships.This informa-
tion,which expresses trust in tagged entities (shop,goods) is itself subject to trust
relationships.The trust in the information expediates the process of forming an
opinion while in-situ.
2.1 Interface Design
The shoppingLense interface is based around a video view,highlighting patterns
that are recognised,and annotating the pattern with the name that was attached to
the pattern by its owner,as illustrated in figure 3.We then display the most trusted
anchors according to the user’s policy around the pattern.By trial and error,given
the expected viewing distance of up to ten meters,we decided to limit the number
of anchors displayed around a pattern to four.
When an anchor is selected,the mode of the viewer is changed to that of a stan-
dard list box,displaying the text or URL associated with the anchor,as illustrated
in figure 4.All the other anchors linked to the pattern are displayed in a list,
ordered by the trust rating of the user.
The level of trust in an anchor is indicated through colour coding (green being
most trusted,orange neutral and red least trusted),the font weight of the text,and
the size of the text.Anchors have a title and a body text.The body text is currently
restricted to be either plain text or a URL.If the anchor is linked to a URL,then
the URL is transferred to the appropriate browser when the anchor is clicked.
Our initial prototype is built on a tablet PC (as seen in figure 5,and work is
under way to port the shoppingLense to Windows CE and other mobile telephony
platforms.The pattern repository is fully functional,based around Tomcat.We
are currently improving the robustness of the repository for deployment in a real
shopping precinct.
3 Validation Experiment:A Shadow Program
Committee
To validate our design,we have constructed an experiment to determine whether
using trust to order the patterns and anchors is better than providing the anchors
in a randomorder.Our experiment is based around the idea of using the shoppin-
gLense to browse the reviews upon a set of candidate papers for a conference so
as to decide which papers are to be accepted,and which rejected.
We chose 6 papers from a recent workshop,3 of which were accepted and 3
which were rejected.We used a population of 12 PhD students working within
the general area of the workshop to generate 2 reviews per paper.The authors then
generated one antagonistic reviewand one spamanchor for each paper,providing
a total of 4 genuine reviews,1 antagonistic review and 1 spam comment per
paper
2
.The four reviews from the student were given a trust rating of 1,the
antagonistic review was given a trust rating of 0,and the spam assigned a value
of -1.The six papers were printed out and attached to display boards,along with
an abstract pattern,as shown in figure 6.
The experimental setup was used in the following two trials using unpaired and
paired evaluations.In both trials,the subject had a preliminary briefing and period
to familiarise themselves with the tool before undertaking the experiment in an-
other room.Each participant was asked to enter the room,use the shoppingLense
to access the potted reviews and decide which three were accepted,and which
were rejected.The total time taken to come to a decision,along with logs of
clicks and currently viewed image were recorded.The shoppingLense was con-
figured to present the anchors either in an random order (unstructured),or in an
ordering based on the trust rating of the review (structured).In the unstructured
view the reviews were still presented with colour and font variations to indicate
trust levels in the review.Hence it is only the ordering of the reviews due to trust
which might effect the outcome.The spatial positioning of the display boards was
permuted across participants.
2
e.g.Greetings for you!Someone has sent you a greeting card.Visit our web-site to find out
who!
Fig.3.The Browser Display augmented video view
Fig.4.The Browser Display - list view
Fig.5.Paper examination usage
Unpaired Trial The original pool of PhD students that generated the reviews
were asked to reviewthe six papers,deciding which of the six were to be accepted
or rejected.Of course there would be some familiarity,but not with sufficient
papers to make an immediate selection.The students were primed on the use of
the shoppingLense,and were then allowed to investigate the papers freely using
the shoppingLense.Ten students were able to complete the trial (the other two
being unavailable while the trial was running).Five used the structured view,
whilst the other five used the unstructured view.Four students were native English
speakers,whilst the others had been in England for at least two years.
Whilst there was too much variation between subject competencies in language
and computer efficiency to allow direct comparison in either total time taken or
number of clicks,all the students reported that the tool was easy to use.We took
note of some suggestions about font and colour usage,and refined the shoppin-
gLense before undertaking the following paired trial.
Paired Trial In this experiment,we asked 14 academics and post-doctoral re-
searchers to evaluate two sets of three papers using both structured and unstruc-
tured orderings of the shoppingLense.Which ordering came first was balanced
across participants.All participants were well-acquainted with the programcom-
mittee concept.Two of the participants were excluded from analysis,since they
misconstrued the task they were undertaking,based on their comments during
and after the experiment.In the analysis,the dependent variable was the time
taken to come to a decision,whilst the independent variables were the orderings.
On average,participants completed the reviewing task significantly faster using
the structured presentation (mean=220s,standard error=22s) than using the un-
structured presentation (mean=250s,standard error=24s) at a significance level
of (p = 0.079,t(11) = −1.934,r = 0.503).
Fig.6.The Experimental Layout
This experiment showed considerable practice effects.We undertook a mixed de-
sign ANOVA,comparing the first and second trials as the repeated measures,and
the order of the structured/unstructured presentations as a between-group variable
within SPSS.The effect of trials shows significance F(1,10) = 9.28,p < 0.012,
whilst the combined interaction shows significance F(1,10) = 6.561,p < 0.028,
as illustrated in figure 7.It can be seen that the structured/unstructured differ-
ence was most obvious upon initial presentation,and that practice effects then
dominated upon the second trial.
3.1 Lessons
What does this tell us about moving forward with the shop based system?The
first phase of the trial provided some important usability testing within a small-
scale trial.The second phase tells us that the approach of using trust to influence
display is quickly understood (faster than learning effects) and does not hinder
the user.While a stronger positive effect might be more exciting,this is a useful
positive outcome.We believe that the benefit of the ordering would become more
apparent with a greater volume of different opinions to process.If we assume that
a large volume of largely untrusted reports might be generated over time in the
shopping scenario,the ability to find trusted sources (directly trusted or through
social/interest networks) would be valuable,whilst retaining the open authorship
model.Given this assumption,our experiment design presents a low complexity
situation in which any benefit would quickly be hidden by any negative aspects of
the system.However,we see a clear (if not large) and significant benefit,which
gives us confidence to proceed with introducing greater data volume,more com-
plex policies and more complex experimental situations.Further,it was only the
150170190210230250270290310
Time1Time2
Marginal Mean Task Time
Structured/Unstructured
Unstructured/Structured
Fig.7.Marginal means for trials versus ordering
ordering of the reviews which was varied.If we had removed other visual indica-
tions of trust in the review it is likely that greater effort would have been put into
the different reviews by the participants and so strengthening our result further.
4 Related Work
Much of the work on adaptive hypermedia uses some measure of relevance to a
user to determine link selection [5].For instance,the work by Sinclair et al in
[6] displays the links using augmented reality,based on the conventions existing
in standard hypermedia systems.Our approach differs fromadaptive hypermedia
norms in that we use trust in a public authoring model,building from the work
on rating schemes in ubiquitous computing,such as Quercia et al [7].This itself
builds on work in the WWW,where ratings are used to build trust in contribu-
tions to the web-site [8],and more recent work in peer to peer networks such as
Eigentrust [9] and the early work by Aberer [10].
There is a large body of work on mobile augmented reality eg [11–15].Of par-
ticular interest to our goal of creating an augmented shopping precinct is work
using AR to supplement the consumer process.Most such work has focused on
leaving control to the retailer and the location administrator,such as the shop
based vision in [16],or has concentrated on presenting work as a narrative [17,
18].Whilst such systems may be initially attractive,without open input,they may
not grow and continue to be useful after the original designers have left.
5 Conclusion
We have presented the design of an open augmented reality tagging system built
upon the ARToolkit,using the group memberships of the pattern and anchor own-
ers to derive the level of trust that the user should invest in the viewed object.We
believe that this will provide an effective mechanism to control the pollution ef-
fects of spamon collaborative AR.
We have presented an experiment showing that using trust to structure the pre-
sentation of the data results in higher task performance within the confines of
displayed paper browsing,but that practice effects dominate after users become
accustomed to the technology.We are deploying the systemfor further user trials.
Acknowledgements
We are very grateful to Ann Light for many useful discussions,the members of
our lab for reviewing and using the tool,to the willing participant academics
within our department,to the members of the program committee of the Mid-
dleware for Pervasive and Ad-Hoc Computing workshop,and to the anonymous
authors who allowed re-reviewing of their papers.
References
1.ISO,ed.:Information technology – Automatic identification and data cap-
ture techniques – QR Code 2005 bar code symbology specification.2 edn.
Number 18004.ISO/IEC (2006)
2.Wakeman,I.,Chalmers,D.,Fry,M.:Reconciling privacy and security in
pervasive computing.In:5th International Workshop on Middleware for
Pervasive and Ad-Hoc Computing (MPAC07),Los Angeles,Ca (November
2007)
3.Billinghurst,M.,Kato,H.:Collaborative augmented reality.Communica-
tions of the ACM45(7) (July 2002) 64–70
4.Wagner,D.,Schmalstieg,D.:First steps towards handheld augmented reality.
In:Seventh IEEE International Symposiumon Wearable Computers.(2003)
5.Kobsa,A.,Koenemann,J.,Pohl,W.:Personalised hypermedia presentation
techniques for improving online customer relationships.Knowl.Eng.Rev.
16(2) (2001) 111–155
6.Sinclair,P.,Martinez,K.,Millard,D.E.,Weal,M.J.:Links in the palm of
your hand:tangible hypermedia using augmented reality.In:HYPERTEXT
’02:Proceedings of the thirteenth ACMconference on Hypertext and hyper-
media,New York,NY,USA,ACM(2002) 127–136
7.Quercia,D.,Hailes,S.,Capra,L.:Mobirate:making mobile raters stick to
their word.In:UbiComp ’08:Proceedings of the 10th international confer-
ence on Ubiquitous computing,NewYork,NY,USA,ACM(2008) 212–221
8.Dellarocas,C.:Immunizing online reputation reporting systems against un-
fair ratings and discriminatory behavior.In:EC ’00:Proceedings of the
2nd ACMconference on Electronic commerce,New York,NY,USA,ACM
(2000) 150–157
9.Kamvar,S.D.,Schlosser,M.T.,Garcia-Molina,H.:The eigentrust algorithm
for reputation management in p2p networks.In:Proceedings of the Twelfth
International World Wide Web Conference.(2003)
10.Aberer,K.,Despotovic,Z.:Managing trust in a peer-2-peer information
system.In Paques,H.,Liu,L.,Grossman,D.,eds.:Proceedings of the
Tenth International Conference on Information and Knowledge Management
(CIKM01),ACMPress (2001) 310–317
11.Feiner,S.,MacIntyre,B.,Hollerer,T.,Webster,A.:A touring machine:Pro-
totyping 3d mobile augmented reality systems for exploring the urban en-
vironment.In:ISWC ’97:Proceedings of the 1st IEEE International Sym-
posium on Wearable Computers,Washington,DC,USA,IEEE Computer
Society (1997) 74
12.Wagner,D.,Pintaric,T.,Ledermann,F.,Schmalstieg,D.:Towards massively
multi-user augmented reality on handheld devices.In:Third International
Conference on Pervasive Computing (Pervasive 2005),Munich,Germany
(May 2005)
13.Newman,J.,Ingram,D.,Hopper,A.:Augmented reality in a wide area sen-
tient environment.In:ISAR ’01:Proceedings of the IEEE and ACM In-
ternational Symposiumon Augmented Reality (ISAR’01),Washington,DC,
USA,IEEE Computer Society (2001) 77
14.Reitmayr,G.,Schmalstieg,D.:Location based applications for mobile aug-
mented reality.In:AUIC ’03:Proceedings of the Fourth Australasian user
interface conference on User interfaces 2003,Darlinghurst,Australia,Aus-
tralia,Australian Computer Society,Inc.(2003) 65–73
15.Kimura,H.,Tokunaga,E.,Nakajima,T.:System support for mobile aug-
mented reality services.In:SAC ’07:Proceedings of the 2007 ACMsympo-
siumon Applied computing,New York,NY,USA,ACM(2007) 1616–1623
16.Zhu,W.,Owen,C.B.,Li,H.,Lee,J.H.:Personalized in-store e-commerce
with the promopad:an augmented reality shopping assistant.Electronic Jour-
nal for E-commerce Tools and Applications 1(3) (2004)
17.Hansen,F.A.,Bouvin,N.O.,Christensen,B.G.,Gronbaek,K.,Pedersen,
T.B.,Gagach,J.:Integrating the web and the world:contextual trails on the
move.In:HYPERTEXT ’04:Proceedings of the fifteenth ACMconference
on Hypertext and hypermedia,New York,NY,USA,ACM(2004) 98–107
18.van Doorn,M.,van Loenen,E.,de Vries,A.P.:Deconstructing ambient intel-
ligence into ambient narratives:the intelligent shop window.In:Ambi-Sys
’08:Proceedings of the 1st international conference on Ambient media and
systems,ICST,Brussels,Belgium,Belgium,ICST (Institute for Computer
Sciences,Social-Informatics and Telecommunications Engineering) (2008)
1–8