Ambient Help · tice and the full blish, to post on fic permission and 12, 2011, Vancouv M...
Transcript of Ambient Help · tice and the full blish, to post on fic permission and 12, 2011, Vancouv M...
AIscMacAafntr
AH
AH
GD
IDapowur
Faffmpthmfnwo
PpncoreCC
ABSTRACT In this paper supports opporcontext-sensitivMultiple videoambiently on a collection of deAfter describinan experiment finding more negative impacraditional help
Author KeywoHelp, Learnabil
ACM ClassificH5.m. Informat
General TermsDesign, Human
NTRODUCTIODespite advanapplications eaprograms can sof commands oways to combunfamiliar arearequire assistan
For most large an exhaustive features. Thesefinding informmaterial on hoperform higherhese standard
many people afound past helpnew version a working and atown [13].
Permission to makpersonal or classronot made or districopies bear this nootherwise, or repuequires prior speci
CHI 2011, May 7–Copyright 2011 AC
J
we present Artunistic learnve learning reos and textual
secondary disesign considerang our implem
which showshelpful inform
ct on the user’s p condition.
rds lity, Understan
ation Keywordtion interfaces
s n Factors
ON ncements in asier to operatstill be a problor functions tobine them intoa of the prognce [21].
applications, set of docum
e references cmation about ow the commar level tasks arhelp systems hare reluctant tp systems inefchance, while
ttempt to figur
ke digital or hard com use is granted ibuted for profit otice and the full ublish, to post onific permission and12, 2011, Vancouv
CM 978-1-4503-0
Justin Mate
210 K{fir
Ambient Help,ning by proviesources while
help resourceplay. We defination for this t
mentation details that Ambienmation, whileproductivity, a
nding, Ambient
ds and presentati
research mte, learning hoem. There are choose from,
o workflows. gram, even ex
many hours armentation for
can sometimesa particular
ands work togre often lackinhave improvedto use them [3ffective and ree others wouldre out what the
copies of all or pawithout fee providor commercial adcitation on the fi
n servers or to red/or a fee. ver, BC, Canada. 267-8/11/05....$10
Ambejka, Tovi G
AutodKing St. E., Trstname.lastn
, a system thiding automatie a user workes are presentne and examinetype of interfacls, we report o
nt Help suppore not having as compared to
t System
on (e.g., HCI)
aking complow to use theoften thousanand many moWhen using
xpert users ma
re spent creatinr the programs be helpful f
command, bether or how ng. Even thougd in recent year37]. Some hav
efuse to give thd rather continuey need on the
art of this work forded that copies aredvantage and thatirst page. To copydistribute to lists
0.00.
bient HeGrossman, Gdesk ResearcToronto, Ontname}@auto
hat ic, ks. ed e a ce. on rts
a o a
ex ese nds ore an ay
ng m’s for but
to gh rs, ve he ue eir
Seeingthan trreasonascreen,videos Howevunderuapplicadifficuapplicathe toowhich explici
Opposiexplicifor helprovidthey neor tranone. Sulearnineven kPreviouof the the sysconsid
Figuapplamba we
In thesupporcontexMultipambienlearnin
r e t y ,
elp George Fitzmch tario, Canadaodesk.com
g other people raditional help able approxim, and at the tion YouTube
ver, all of thutilized since ation so users
ulty finding thation tooltips, ol-level and do
utilize the toitly access the v
ite to the traitly asks for help has been exe help in situaeeded it, such nsitioning fromuch learning c
ng [42] since tknowing there us attempts suDay” have notstem’s terms, ered frustrating
ure 1. Structurlication monitoient help displa
eb help page.
e paper we prts opportunistxt-sensitive leaple videos andntly on a seconng is supporte
maurice
a
work can somresources [45
mation of waime of writingfor the query hese videos
they are noare unaware o
hem. ToolClipbut they are o
on’t always shoool. Also, usevideos by calli
aditional “pullelp resources, axplored [26]. Tations where thas awareness o
m a novice wocould be referrthe user has lwas somethin
uch as Microsot been ambientnot the user’sg and ineffectiv
re of Ambienor on the left, ay on the right c
present Ambientic learning barning resourcd textual help ndary display
ed during natu
metimes be mor5]. Videos can tching anothe
g there are ove“Adobe Tutorrun the risk
ot available wof their existenps [20] show only available ow higher leveers are still ring up the toolt
l” model whean implicit “pu
These systems he user didn’t eof unknown fu
orkflow to a mred to as “uninlearnt somethinng they neededft’s Clippy [26t, requiring use, and as such ve.
nt Help systemdynamically u
containing 5 vid
nt Help, a syby providing ces while a u
resources are[Figure 1]. Op
ural breaks in
re valuable serve as a
er person’s er 350,000 rial” alone.
of being within the
nce or have videos in
for help at l taskflows
required to tip.
ere a user ush” model potentially even know
unctionality more expert ntentional” ng without d to know. 6] and “Tip er input on have been
m. Main updating deos and
ystem that automatic, ser works.
e presented pportunistic
the user’s
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2751
workflow since the information is preloaded for convenient viewing. This design reduces the transaction cost traditionally associated with obtaining help information.
After reviewing related work, we discuss design considerations for an ambient help system, present our implementation, and report on a study that shows our system supports finding more useful information, while not having a negative impact on the user’s productivity, compared to a traditional help system.
RELATED WORK
Contextual and Video Based Help Since the outset of on-line help, there have been explorations into contextual [2,12,29] and video-based [5, 27,29] delivery, with less research exploring both [20,1]. Shneiderman [44] argues that graphical demonstrations can be the most direct way for novices to learn procedural knowledge. However, its implementation was expensive and time consuming when the ideas were first proposed in the early 90’s [27], and in general, on-line documentation was “crippled” by hardware and software limitations [15]. The evolution of CPU power and video hosting services has made video documentation a viable possibility [31].
In terms of publishing, it is often easier to record a 5 minute tutorial video than it is to take the necessary screenshots and write up an easy to follow and useful tutorial [39]. However, with text based help a quick glance might provide the needed assistance, where a video may require a longer viewing period to get the same information [16]. It has also been found that some users naturally prefer videos while other prefer text-based help [20].
Despite psychology research which exposes the benefits of contextual assistance [3], help systems are generally offered through completely separate components. This discontinuity can result in delayed, disruptive, inconsistent, and obtrusive help systems [2,30].
Previous research in animated assistance [23,38,1] has explored contextual demonstrations. One good example is Google Sketchup which has an “instructor” panel providing tool-based contextual animated assistance whose content changes as the user switches tools. While their assistance consists of very basic animations which only illustrate how the tools function, we use full motion videos which show how the tools can be used in a workflow.
Ambient Information Systems and Displays Ambient and peripheral displays [8,35,40,46] support the monitoring of noncritical information with the intent of not distracting or burdening the user. One typical example of an ambient display is Portholes [10] which periodically takes a video snapshot of members of a community and presents them as a grid of images for each user to view and have an awareness of who is in and available for communication.
Pousman and Stasko [41] describe four dimensions of consideration when designing an ambient display:
information capacity, notification level, representational fidelity, and aesthetic emphasis. In our approach, we wish to present contextual learning resources in an ambient display where the information capacity and representational fidelity is high and always available to be absorbed by the user during breaks of work activity. Also, we do not interrupt the user, so the notification level is low.
Evaluation of the usefulness of ambient displays is often challenging [24,33,35,43]. The evaluation criteria typically centers around noticeability, comprehension, relevance, division of attention, and engagement. While these are all important, we are particularly interested in increasing the temptation to learn something new, perhaps serendipitously. This is similar to over the shoulder learning [45] where users can see how other users work.
One concern with an information rich ambient display is that it may induce information overload [28]. While this could be true, the very nature of our ambient display is that it contains optional information that can be ignored during intense work or during high levels of information processing in the main display.
Buxton describes a framework for modeling systems which factor in not only human-human and human-computer interaction but a second dimension of foreground and background activity [7]. Here, foreground activities are in the “fore of human consciousness – intentional activities” such as typing in a computer. Background activities take place “in the periphery ‘behind’ those in the foreground” such as being aware of someone in the next office typing. Our intent is that Ambient Help easily transitions between background and foreground activity.
Interruptability and Notification A great deal of research has studied technology-mediated interruptions and notification of system or application events [1,17,22,36]. The concern is that interruptions can be an annoyance, adversely affect a user’s task completion time, and make them vulnerable to mistakes. Thus, researchers have explored ways to mitigate the problems by deferring interruption until users are more open to the information during their work. Attentive user interfaces attempt to recognize some basic user states (presence, head position, mouse activity) and act differently according to user behavior [14]. Our goal is not to interrupt the user, particularly during their intense work.
Transaction cost A key goal for our system is to reduce the transaction cost of a user accessing help and learning resources while working. Research has explored the dynamics of switching between work and help [4] and shown that small interaction improvements can cause categorical behavior changes that far exceed the benefits of decreased task times [6,18].
The Cocoa Buzz system [11] runs on a user’s extra display, or a large shared display, and automatically cycles through different information sources (such as news sites)
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2752
customized by the user. About every minute it updates and generates a new information collage. Our system differs in that the learning resources are contextually retrieved.
Instead of a user explicitly entering in search terms, implicit and activity based web searching is being researched. Maekawa et al. [32] have explored query free web searching based on activity detected by sensors and internet-connected domestic appliances. For example, while washing a coffee maker, a web page is automatically retrieved on how to clean it using vinegar. Henzinger et al. [25] automatically select web pages that a user might want to see which are relevant to a TV program they are currently watching. This implicit style of searching system matches our approach but differs in that we focus on software learnability [21].
DESIGN CONSIDERATIONS In the following we look at a number of design goals we considered while creating the Ambient Help system to be as useful as possible, while not being disturbing or distracting.
Ambient Nature A primary consideration is for the help content to be displayed ambiently; that is, in an always-on display which does not require user input to function. In this way, the time-cost to access the help content is reduced since user interaction is not required to call up the information. Since the display is persistent, the user can look over for help information at any time, and it should be designed in a way so there is no cost associated with not using the display; i.e., it can be easily ignored. Similarly, to minimize distraction the system should never interrupt the user.
Relevance of Content For an ambient system to be useful, it must be able to find and display content which is relevant to the user. This is different from Tip of the Day where you have small pieces of help information, or standard help systems which give access to more content, neither of which use any knowledge of the user’s context. Besides attempting to show help content which is a good match for what the user is looking for, it is also useful to present generally useful or interesting content to enhance the variety of the system and provide the opportunity for exploration and serendipitous learning [34].
Transition Between Ambient and Interactive Ideally there should be no explicit switch required to move between ambient and interactive modes of operation. Users should be able to move between the main and help windows freely without concern for the state of the help system.
Multiple Forms of Information Traditionally, product support materials have been delivered in mostly text, with some diagrams mixed in. A limited number of help videos are now being published by the software manufacturers, but with the recent advancements in the ease of posting videos to the internet (YouTube, etc.) many individuals and small third-party companies have started posting their own tutorial videos.
Videos have the benefit of being able to demonstrate the concept being described. Textual resources on the other hand have the benefit of being more easily skimmed or consumed quickly. Besides official product documentation, there are also significant web-based textual resources such as discussion boards and tutorial web pages which many users turn to for information.
With their being benefits to both video and textual help information, we recommend supporting both.
Channels of Information Ambient display systems can range from displaying a single stream of information up to a great number simultaneously. By displaying only one channel, that single item takes up the entire information bandwidth of the display, and as such, when the user looks over they are assured to see it. With many channels on the same display, each individual channel must be of smaller size and possibly lower fidelity. With more channels the chances are greater that useful content will be displayed. However, if too many things are going on, that useful content might be missed. In the end, designers should ensure a balance between the extremes of too little and too much information.
Location An ambient help system could be implemented on a secondary monitor or within the primary monitor. In our experience we have found that users are quicker to dismiss items which take space away from their primary application. Instead of using a secondary display on the same computer, the system could also run on a secondary system, such as an Apple iPad or similar device.
Content Variance If the content of the ambient display never changes, users will stop looking over because they have already seen the information. On the other hand, if the content changes too often it could be a distraction, or relevant information could be swapped out while the user is trying to view it. As such, there should be a balance between updating too slowly leaving the information stagnant, and updating so frequently that the content doesn’t have time to be used.
SYSTEM DESCRIPTION
Overview Given these design considerations we now present our exploration into offering ambient help. The Ambient Help system is displayed on a secondary monitor placed beside the main monitor [Figure 1]. The user sits directly in front of the main application monitor with the secondary monitor angled off to the side, keeping the application in the center of focus while the ambient display is visible in the periphery. The system automatically presents help content relevant to the commands the user is working with.
Ambient and Interactive Modes On the left side of the screen are relevant help videos, and on the right side is a web page [Figure 2]. We display closed captioning information as an overlay for videos with
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2753
thwrimf
OamvsBtow
Wcrthaovtolith
he informationwithout the audreference pagess a page for
making it easiefor a smoother
F
Our system reaudio and dimmmain applicatiovideos are shoseconds, with Being connecteo use the appli
without specific
When the cursocursor focus isremained dimmhe application
and muted at thoption to “checvideo playing ao the main aistening to thehe video will p
Figure 3. E
n available makdio playing. Ws as the sourceeach comman
er for users to transition betw
igure 2. Ambien
educes potentiaming the videoon. We also reown (only cha 1 second sm
ed to the main ication contextc input from th
or is within thes shown at fumed. By defau
window, the vhe reduced frack” a video [Fat a reduced vapplication. The video while wplay to complet
lements of the A
king it possibleWe choose to ue for the web pnd with consis
find informatiween pages.
nt Help interfac
al distractionsos while the cueduce the fram
hanging framesmooth fade bapplication al
t to show relevhe user.
e help windowll brightness w
ult, when the cvideos return tame rate. HoweFigure 3] whicvolume when this allows thworking, and ation without be
Ambient Help v
e to follow alonse the comman
pages since thestent formattinion, and makin
ce.
s by muting thursor is over th
me rate at whis once every etween framesllows the systevant help conte
w, the video wiwhile the othecursor returns o being dimmeever, there is thch will keep ththe focus returhe user to kealso ensures theing replaced.
video player.
ng nd ere ng, ng
he he ch 5
s). em ent
ith ers to ed he he
rns ep
hat
In our of a nethe sasuccesrelevanconstanvideos
Based we cho“primavideos even thsense ous to dchanceworkinto be tthe privideo manuait, mak
If a vreplacethe vidthumbnvideo, the rele
AutoCThe AAutoCsystemAutoCor othnotificahad recfrom twYouTuand an
AutomIdeallywith Cvideo were mallow ufor now
Knowifind mdirectlythus olookingdisplayeasier comma
system, the disew command, ame commandsion, or the cunt. Between cont so users are without fear o
on the channeose to show 5ary” position (a
(at 490 by 278he small videoof what is goindisplay a num
es of presenting and the dispthe most relevimary positionin the top,
ally promote aking it larger an
video the usered, it can be redeo player. Thenails which alland a button
evance algorith
CAD Plug-in Ambient Help
CAD plug-in um hooked into CAD to get a nherwise starteations the plugcently used andwo sources: a
ube search for n AutoCAD trai
matic Video Any, all tutorial viChronicle-like
commands ocmodified, etc. users to recordw, nearly all vi
ing which commore germaney to positions i
overcoming thg for commandy markers for for users to n
ands of interes
splay may onlyand even then
d has been isurrently playinommand issuanable to read th
of them being r
els of informati5 videos. One at 970 by 536 8) below [Figu
os are large enong on in the v
mber of differeing a useful play is updatin
vant is automatn, in theory plargest, positismaller video
nd easier to see
r did want totrieved by clice interface alsolow navigationto move to thhm described l
p system wausing the Obje
the CommandSotification whe
ed to use a g-in tracked wd was currentlylocal cache of“AutoCAD T
ining DVD (10
nalysis ideos ever crea[19] metadata
ccurred, settingHopefully ap
d videos with tideo help conte
mmands happee videos. Addin the video to
he need to mad invocations. the commandsnavigate to pat were used.
y change at then, it may remassued multipleng videos are snces the contehe help page orreplaced.
ion design convideo resides resolution) and
ure 2]. At this sough for the usvideo while stient clips to inone. While thng, the video dtically transitioplacing the mion. The userto the top by c
e the details.
o watch happking the “Prevo includes a timn to different pohe next video slater.
as implementectARX frameStarted messenever a user ccommand. W
which commandy using. The vif videos obtainTutorial” (1,3707 videos) [Fig
ated would be a indicating whgs were changpplications wilthis type of meent is untagged
ened where allditionally, we display relevaanually scrub In the video ti
s [Figure 4] tharts of the vi
e beginning in stable if e times in sufficiently ent remains r watch the
nsideration, in the top
d 4 smaller size we feel ser to get a ll allowing
ncrease the he user is determined oned up to
most useful r can also clicking on
pens to be v” button in meline and oints in the selected by
ed as an ework. The sages from clicked on,
With these ds the user ideos came ned from a 79 videos), gure 2].
marked up here in the ged, layers ll someday etadata, but d.
lows us to can seek
ant content, the video
imeline we hat make it deo where
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2754
Acapcmassim
PDdjutrinthhaSin[mou
Ccaailicwo
OICbfimr
As a proof-ofcommand scrapa program to aproduced Autocommand strematerial for thiassumptions thscale, and the system such asmprove the alg
Figure 4. V
Possible ApproDifferent appldetermining whust been issuedell the active
rendered darknformation is he tool is acti
highlighted to another darker Some productsndicating whiFigure 5c]. By
mouse button wone could deteused.
Figure 5. Excommand usaOffice, (C) Au
Commands in clicking on a ba shortcut, entautomated scrips issued, all cine window [F
command usagwindow in Autopen in all of th
Optical CharactIn the AutoCADCommand:namebeen issued. Wframes with thmage for each
run some stan
f-concept to sping process coanalyze our cooCAD tutoria
eam tags for s process allow
hat all videos welements wouls Sikuli [47] ogorithm to loca
Video timeline sh
oaches for Deteications suppohich tool is ad. For example
tool by lookker than the
persistent andive. The buttoone colour wcolour when
s such as Autoich tool or my combining thwas clicked (permine which
xample avenueage in: (A) Adoutodesk Maya, (
AutoCAD canbutton, typing itering a commpt. However, reommand histo
Figure 5d] creage. It is possiboCAD, but mohe videos in ou
ter RecognitionD command lie indicates thWe started byhe OpenCV tooh of the 4 visibdard OCR pa
how one wayould be autom
ollection of 10al videos and
each. Limitiwed us to makewould have theld be in the saor Prefab [9] cate and recogni
howing comman
ecting Commanort unique w
active or whiche, in Adobe Phking at which
others [Figd available forons of many pwhen the mou
they are clickodesk Maya h
menu item the his informationperhaps from t
tools and m
es for determinobe Photoshop, (d) Autodesk Au
n be entered ininto the comm
mand alias, or egardless of hoory shows up iting a good op
ble to close theost users leave ur sample colle
n (OCR) ne history, a li
hat the commay capturing inolkit. We cut ble text lines aackages (ABBY
y in which thmated, we creat07 professionald generate thing our soure the simplifyine UI at the samame positions. could be used ize UI elements
nd markers.
nds ways of visual
h command hhotoshop you ch tool button gure 5a]. Thr the entire timproducts becomuse if over, anked [Figure 5bave a status b
cursor is ovn with when ththe audio trackenu items we
ning tool and (B) Microsoft
utoCAD.
n multiple wayand line, issuinissued from a
ow the commanin the comman
ption for trackine command linit open, and it
ection.
ine of the formand “name” hndividual imagout areas of thand attempted YY FineReade
his ed lly he
rce ng me
A to s.
lly has an is
his me me nd b]. bar ver he k), ere
ys: ng an nd nd ng ne is
mat has ge he to er,
Simplemost Oscannevideo fcharact[47] wrecognalphabsettingsame astep aloour dicimage pixel rof the greaterdiction100% a
Video At the was cacommacomma
Each v,
wherecomma
.
.commeFor eaadded model,worth which mentiovideo multipllinearlyhave jufor vidsome tare excthe selare sele
This iswell profess
eOCR, ocropusOCR packages ed text, they peframes—generters correctly.
where it appenized correctly.bet of charactergs as a default amount as the tong the line coctionary of renand moving
rounding, then pixel values l
r than 99% nary of the coaccuracy in co
Relevance beginning of
alculated for alland most highands with a dec
video v’s relev was calculate
,
∑
. is and name c
, . op
ents and commach user comm
to account f, having the comore than hais in turn mo
oned in the keyis already p
lied by a facy down to 0.0 ust started pladeos to be repltime. Videos acluded from thections. The vected to be pla
s just one ratinwith our msionally author
s, and others) are tuned to re
erformed quite rally recogniziThis is consis
ears that mosy. To improve rs using the saminstallation of text in the videomparing each ndered characteit around slighusing a square
let us find the accuracy. Th
ommands in Aommand recogn
f each commanl available videhly and weigcreasing power
,
vance score fed using the fol
. 10 ..
∑ ..
1 if the title ofc, otherwise .perate on the
mand stream (iment the videofor generally ommand used iaving the comore valuable thywords, descri
playing in thctor starting after 5 minute
aying priority, laced after the
among the moshe calculation
videos with the ayed when a ne
ng scheme, whimixed collectired tutorial vid
to extract the ecognize printepoorly on the
ing less than 5stent with prevst of the texton this, we reme font and an
f AutoCAD, sceo. Then, we wcharacter in th
ers. Enlarging htly to accouned difference cmatching charhis, combine
AutoCAD let unition.
nd, a relevanceeos weighting
ghting the prer function:
6
for a given collowing formul
8
7 6 0.02
12
f the video conit is 0. , .
e keywords, dif available) reo had, 0.02 ppopular video
in the commanmmand name inhan having theiption or comme display, itsat 1.2 which es. This gives vand then make
ey have been pst recent 50 to s to support v highest releva
ew command is
ich we have foion of YouT
deos.
text. Since ed and then text in our
50% of the vious work t was not
endered the nti-aliasing
caled to the were able to he image to
the source nt for sub-comparison racter with d with a us achieve
e score, R, the current
evious five
ommand c, la:
ntained the Similarly,
and description, espectively. oints were
os. In this nd stream is n the title,
e command ments. If a s score is
decreases videos that es it easier playing for be viewed
variation in ance scores s started.
ound works Tube and
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2755
EWwfd
PTberhthyhr
AT36Frthmth
DAwaatrsnthd
HTcsA
EXPERIMENT We wanted to work-like situafrom the systemdistraction and
Participants Twelve paid vbetween the agemail list of Aurequired to be how long they hey typically
years, and 3 tohad used Autoreceiving forma
Apparatus The experimen3.16GHz quad-64-bit Edition. FX 5600 and wresolution of 1heir chair and
monitor with thhe right [Figur
Design A repeated mewith the indepeand help systemand help systerial conditions
statistics the number of videhe number of
data stream the
Figur
Help Systems The system descondition. As system consistAutoCAD 201
evaluate Ambations, could dm. In addition, productivity c
volunteer partges of 23 and utoCAD users
proficient wihad been usinuse it, with r
o 7 days/weekoCAD in a pral classroom tr
t was conducte-core desktop The graphics
was running tw920 by 1200. d keyboard dhe second monre 1].
easures withinendent variablem (ambient, m
em were counts lasted 15 mexperimenter
eos watched, nuseful/interesty were found i
re 6. Setup of m
scribed in this a baseline c
ted of three 1 online docu
bient Help to derive some ins
the study allowompared to tra
ticipants (8 m30 were recru
. To be eligiblith AutoCAD ng the programresults rangingk respectively. rofessional setraining with the
ed in an encloscomputer runn card was a N
wo 24” Dell moParticipants w
directly in fronitor angled an
n-participant des being task (
manual). The oterbalanced. E
minutes. Duringjudged and
number of helpting items founin.
manual help cond
paper was useomparison, thwindows: Yo
umentation we
see if users, structional valuwed us to look aditional help.
male, 4 femaluited through e the users weand were ask
m and how oftg from 4 to All participan
tting as well e software.
ed office usingning WindowsNVIDIA Quadonitors each at
were seated wiont of the mand positioned
design was us(intense, casuaordering of ta
Each of the fog each trial th
counted werp pages viewend, and in whi
dition.
d as the ambiehe manual heouTube and thbpages, and th
in ue at
le) an
ere ed en 10 nts as
g a s 7 dro t a ith ain to
ed al) ask our he re: ed, ch
ent elp he he
interfac[Figuresame mintegramanua
TasksThe intypes ofocusedinstancand thparticutime br
For thdocumdesigncabinetmotivacabinet
The taexperiecabinetwere acould perform
For thbluepridrawinwere torequiretwo coThe baAutoCRECTAwould commacommauser pocommaensure user, ththis lishelp sywhen t
ProcedUpon ato see helpfulexplain
ce to an Aue 6]. In this wmaterial as thated into AutoCally found by th
ntense and casuof working cod on the task ce, they are whose where thular rush (for erowsing the int
he intense tasment with a gr
s [Figure 7]. Tts as quickl
ation, the pats was to be aw
ask was desigenced AutoCAts without mu
also designed tcount and c
mance in both
Figure 7. Port
he casual taskints for a hou
ngs in any mano add a patio oement was thaommand lists: asic list conta
CAD commanANG, which a
know. The rands which areands, but are uopulation. Theands that the pthe commands
hey were instrst they had not ystems when uthey are doing
dure arrival the part
if “a secondl to people wned that we w
utoCAD profway the manualhe ambient heCAD and the che participants.
ual tasks wereonditions: thoseand cannot aff
working towardhe user is wexample, they ternet or talkin
sk, users weraphic underlaThey were insly and accurarticipant whowarded an extra
gned to be siAD users wouuch trouble. Tto each be of scompare the help system co
tion of the inten
k, participantsuse and told tnner they chosor some landscaat they use 3 cbasic comman
ained 20 of thnds such as are commands rare list contae among the 1
used by a smalle idea being thparticipants wes selected fromructed to only used before. T
users may explexploratory lea
ticipants were dary display owhile using A
would be presen
fessional trainl condition hadelp system, bucontent would .
e created to sime where the u
fford to be distds a very tightorking, but nmight spend s
ng to colleague
re given an ay of 14 kitchstructed to recrrately as poso completed a $100 gift car
imple enough uld be able toThe cabinets tsimilar compleuser’s cabine
onditions.
nse task docume
s were givento modify or se. Some examaping elementscommands fronds, and rare che most frequ
LINE, ERAthat all AutoC
ained a collect100 most frequl (<5%) percenhat these wouldere not familia
m this list were choose comm
This allowed uslicitly need helarning.
told that we wof informationAutoCAD”. It nting the help
ning DVD d all of the ut was not need to be
mulate two ser is very tracted (for t deadline), not in any some extra s).
AutoCAD hen cabinet reate these ssible. As the most
rd.
that most o draw the themselves
exity so we et drawing
ent.
n a set of extend the
mples given s. The only
om each of commands. uently used ASE, and CAD users tion of 10
uently used ntage of the d be useful ar with. To new to the
mands from s to test the lp and also
were testing n could be
was also content in
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2756
twfththwtossua
RBw
UTpnca(ath0s
WtisinittathaAath
ImAa[8rdoim
wo different wfound any infohem. They werhe content on
wanted to, and o let us know
study the exasecondary monuser to explain anything useful
Results Because the twwas performed
Useful Items FoTo assess the eproviding usefunumber of uscondition [Figuanalysis of var(F1,11 = 12.6, ambient and 1.2he intense task
0.17 for manusignificance (F
Figure 8. Aveof the four tasreport standa
With the cursorime the users
second monitonteresting to ntems while usask although thhe second disp
and 18.4%). TAmbient Help and they are abhan they would
mpact on PerfoAs a measure oat the number oFigure 10]. Th
8.63 for manuarespectively. Tdifference to bour belief that mpact when th
ways and we wformation that re encouraged
n the secondarif they found
w so we couldminer saw thnitor for 10 sewhat they wer
l.
wo tasks were for each of the
ound effectiveness oful informationseful or interure 8]. For the riance showed
p < .005) w25 items for mk the means weual. This diffeF1,11 = 2.83, p =
erage number osk/help system
ard error).
r tracking data spent on aver
or for each ote that users f
sing the Ambiehey spent abouplay for both hhis result is csystem to have
ble to do that wd in a tradition
ormance of performanceof cabinets comhe mean resulal with standar
The analysis obe significant
Ambient Helhe user is imme
would be lookinwas useful o
to “glance at, ry monitor” aanything usefu
d keep a recorhe participant econds, he wore looking at a
so different, se task condition
f the Ambient n to the user, wresting items
casual task, rea main effect
with means of manual, a 2.6 tiere 0.92 items
erence howeve .121).
of useful items conditions. (No
we can also lorage in AutoCcondition [Fi
found 2.6 timeent Help systeut the same amhelp system coonsistent withe users find mo
without spendinnal manual help
e in the intensempleted with elts were 7.88 rd deviations o
of variance did(F1,12 = 0.974lp would not ersed in an inte
ng to see if thor interesting or interact wit
as much as thful or interestinrd. If during th
looking at thould prompt thand if they foun
separate analysns.
Help system fwe looked at th
found in eaepeated measufor help syste3.25 items f
imes increase.for ambient an
er did not rea
found in each ote: Error bars
ook at how mucCAD and on thigure 9]. It s as many usef
em in the casumount of time oonditions (17.7
the goal of thore useful item
ng any more timp situation.
e task, we lookeach help systefor ambient anof 4.54 and 3.7d not show th, ns) supportinhave a negativ
ensive task.
ey to
th, ey
ng, he he he nd
sis
for he ch
ure em for In nd ch
ch he is
ful ual on
7% he
ms, me
ed em nd 73 he ng ve
Figupartieach
Figu(Not
After someDisagr“The cThe reDisagrmajoritsystemwho fo
Figuwas h
Participthe excasualambien
ure 9. Cursor icipants. Percen
h monitor.
ure 10. Number te: error bars re
each conditionstatements on
ree, and 7-Stroncontent of the eesults have beree (1, 2), Nety of users (8
m helpful in theound the manua
ure 11. Results helpful”.
pants were alsxtra monitor dl tasks, 3 of 12nt help system
movement for ntages show fra
of cabinets draeport standard
n the participn a Likert scangly Agree). Oextra monitor ween aggregateither (3, 4, 5)
8 of 12) said e casual task al system help
for “The conte
so asked if thedistracting”. I2 users agreed [Figure 12].
each conditioaction of time s
awn in the intenerror)
ants were askale (between
One of the statewas helpful” [F
ed into three ) and Agree (they found thcompared to oful.
ent of the extra
ey “found the In both the inthis was the c
on of all spent on
nse task.
ked to rate 1-Strongly
ements was Figure 11]. categories:
(6, 7). The he ambient only 1 user
monitor
content of ntense and case for the
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2757
Wrasc
Wf
B“[thimtrfewh
CFph[af
Figure 12. Rewas distracting
While not a lorating of the diambient help insaid the interfcompared to on
Figure 13. Rewas distractinfirst vs. secon
With these resufind the interfac
Besides distrac“content of theFigure 14]. Feheir productivmportance in trying to be as
for this task essentially the when not in uhindrance to a u
Figure 14. Rehindered my p
Content SourceFor each of tpages), the nuhigher in the Figure 15]. Fo
ambient and mafor help pages t
esults for “Theg”.
ongitudinal stuistraction durinnterface compaface was distranly 1 during the
esults for “Theng” during thend task.
ults, we believce distracting g
ction, participe extra monitoew users felt thvity in eitherthe intense tas productive as
the ambientsame subjectiv
use, the ambiuser than a mo
esults for “Theproductivity”.
es the two conteumber of itemambient than
or videos the manual respectivthe means wer
e content of the
udy, if we lookng the first tasared with theiracting during e second task [
e content of the e ambient help
ve that many given time to g
pants were alsor hindered thehat either help r task. This isk where the ps possible. It it and manuave rating, suggent display isnitor showing
e content of the
nt sources (vms viewed wn the manual means were 4.vely (F1,11 = 5.re 4.83 and 2.8
extra monitor
k at each usersk they used thr second, 5 use
their first tas[Figure 13].
extra monitor
p condition in
users would nget used to it.
so asked if their productivitysystem hinderis of particul
participants wes promising th
al systems hagesting that evs no more of static content.
extra monitor
ideos, and hewas significant
help conditio.08 and 1.67 f.04, p < .05) an3 (F1,11 = 8.25,
r’s he ers sk,
not
he y” ed lar ere hat ad en
f a
elp tly on for nd , p
< .05).the parat the c
The nuambienpages (pagesthe rasignificvs. 0.1compa.05). Wvideos Severa20 seco“videoinformcompleto play
Figufounerror
Overalcould bcould bfact, o“Oh nthey wo
Figucould
DISCUOur ssignificinformobjectiintenseusers’ ouseful time anperformreactio
It is importanrticipant spent content, not tha
umber of usefunt help system(3.17 vs. 0.92)reached signif
atio of usefucantly higher 84: F1,11 = 7.5
ared to manualWe believe thi
since the statial times we sawonds and then
o was taking tomation from a v
ete tagging of y the video at th
ure 15. Numbernd useful in eachr bars report st
ll when users be useful, 7 obe useful com
one user duringo, my videos aould be really
ure 16. Results d be useful”.
USSION & FUTtudy has shocantly improve
mation during ively, or subjee tasks. A longown machine ito see how thend explore themance and lons users have
nt to note that hsome amount
at it was simply
ul items found wm for both vide), however onlyficance (F1,11 =
ul items founfor help page
56, p < .05) as help (0.425 vs
is ratio is higic pages are eaw participants
move on to thoo long”. The tvideo can hope
the video conhe point of the
r of videos and h of the help sytandard error)
were asked ifof 12 agreed th
mpared to 0 forg the manual are gone now. useful right no
for “A system l
TURE WORK own that the ed the ability o
casual workectively, hinderger term study, in their standarese results holde impact of thlearning. Give to new help
here “viewed” of time actual
y shown on the
was also higheeos (1.0 vs. 0.5y the difference= 7.44, p < .05nd to items es than for vid
well as for ams. 0.203: F1,11 =
gher for help pasier to consumlook at a videhe help page btime it takes tofully be reduce
ntent allowing relevant inform
help pages viewystem condition
f they felt eithhe Ambient Hr manual. [Figuhelp conditionI wish I had t
ow”.
like <Ambient/M
Ambient Heof the user to k, while genring performanor one perform
rd environmend over a longee system on hen the oftenp systems and
means that lly looking e screen.
er using the 5) and help e with help 5). Overall viewed is
deos (0.448 mbient help = 8.17, p < pages than me quickly. o for 10 to
because the o get useful ed by more the system mation.
wed and s. (Note:
her system elp system ure 16]. In
n remarked them back,
Manual>
elp system find useful
nerally not nce during med on the nt would be er period of higher-level n negative d interface
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2758
designs, it would be interesting to see in what scenarios users would choose to keep the ambient display open.
We measured the number of useful or interesting items that the participants found with the system. It would be valuable to explore the relationship between information that is interesting to the user now and the potential it has to become useful in the future.
A potential concern with such an active help system is that it may be distracting to users. We found that after their second 15-minute exposure to our system most users did not find it distracting. However, we believe the system could be even less distracting if we choose not to animate the position changes of the videos from the bottom to the primary position while in the ambient mode. There may be other ideas such as pausing the videos completely while over the main application, or updating less frequently in general which would be interesting to explore.
We chose to show five videos and one web-based help page but a study to examine how many channels of information supports the best mix of providing information without the user becoming overloaded would be valuable. We suspect this might come to the finding that different users would like to see different volumes of information and would lead to the question of how to design the interface to support varying amounts of channels.
While we only used two sources of videos and one source of text-based help for this prototype, future work could explore using more sources and how to best integrate them into a single system. Along with more sources a more sophisticated algorithm could be developed for finding relevant content. Additional pieces of context which could be used to improve context matching include: which Ribbon tab is active, which dialogs are open, and the actual content of the document. Although not accessible through the public API, YouTube has the ability to automatically generate closed captioning data for their videos. This text might be a valuable resource for trying to automatically determine what is going on in each of the clips.
By virtue of being an always-on display, the transaction cost to access the help information is already very low. We could decrease the cost of interacting with the content by introducing keyboard shortcuts to perform some of the more common actions such as listening to a video. It would also be nice to reduce the amount of time required for a user to scan a video for useful content. We have thought of introducing some graphic overlays to the videos to more obviously show which commands are being used, and a longer selection of the closed captioning text to serve as a “glanceable summary” so they can be more quickly scanned for useful information.
CONCLUSION Design considerations for an ambient system that provides help content using a push rather than pull delivery model were discussed. The resulting Ambient Help system offers
the potential to increase the user’s learning in multiple ways: both “intentional” learning, where the user knew they needed help, and perhaps as importantly, “unintentional” learning, where the push model presents information the user wouldn’t even know to look for. In the casual task, which simulates what we envision to be the primary environment for using this type of system, Ambient Help led to 2.6 times as many useful insights as the control. Besides being helpful during casual tasks, users generally agreed that the system did not negatively impact their productivity during intense work and this result was supported by the quantitative data. Although implemented in AutoCAD, we believe Ambient Help is a general solution which could be made to work with almost any software application where learning is an important issue.
ACKNOWLEDGEMENTS We thank Ramtin Attar for designing the AutoCAD tasks.
REFERENCES 1. Adamczyk, P. D., Iqbal, S. T., and Bailey, B. P. (2005). A
method, system, and tools for intelligent interruption management. In Proceedings of the 4th International Workshop on Task Models and Diagrams. 123-126.
2. Ames, A. L. (2001). Just what they need, just when they need it: an introduction to embedded assistance. ACM SIGDOC. 111-115.
3. Anderson, J. R., Boyle, C. F., Farrell, R. and Reisner, B. J. (1984). Cognitive principles in the design of computer tutors. Cognitive Science Society. 2-9.
4. Andrade, O., Bean, N., and Novick, D. (2009). The macro-structure of help. ACM SIGDOC 2009. 143-150.
5. Baecker, R. (2002). Showing instead of telling. ACM SIGDOC. 10-16.
6. Brandt, J., Dontcheva, M., Weskamp, M., and Klemmer, S. R. (2010). Example-centric programming: integrating web search into the development environment. CHI, 513-522.
7. Buxton, W. (1995). Integrating the periphery and context: A new taxonomy of telematics. GI, 239-246.
8. Cadiz, J. J., Venolia, G., Jancke, G., and Gupta, A. (2002). Designing and deploying an information awareness interface. ACM CSCW '02, 314-323.
9. Dixon, M. and Fogarty, J. (2010). Prefab: implementing advanced behaviors using pixel-based reverse engineering of interface structure. ACM CHI,1525-1534.
10. Dourish, P. and Bly, S. (1992). Portholes: supporting awareness in a distributed work group. CHI, 541-547.
11. Eagan, J. R. and Stasko, J. T. (2008). The buzz: supporting user tailorability in awareness applications. ACM CHI. 1729-1738.
12. Farkas, D. K. (1993). The role of balloon help. Journal of Computer Documentation. 17(2):3-19.
13. Fitzmaurice, G., Matejka, J., Mordatch, I., Khan, A., and Kurtenbach, G. (2008). Safe 3D navigation. I3D, 7-15.
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2759
14. Fosso, F. and Porta, M. (2009). A vision-based attentive user interface with (semi-)automatic parameter calibration. Proceedings CompSysTech '09, vol. 433. 1-6
15. Goodall, S. D. (1992). Online help: a part of documentation. Systems Documentation. 169-174.
16. Grabler, F., Agrawala, M., Li, W., Dontcheva, M. and Igarashi, T. (2009). Generating photo manipulation tutorials by demonstration. ACM SIGGRAPH. 66.
17. Grandhi, S. and Jones, Q. (2010). Technology-mediated interruption management. Int. J. Hum.-Comput. Stud. 68, 5 (May. 2010), 288-306.
18. Gray, W. D. and Boehm-Davis, D. A. (2000). Milliseconds Matter: An Introduction to Microstrategies and to Their Use in Describing and Predicting Interactive Behavior. Journal of Experimental Psychology: Applied, 6 (4), 322-335.
19. Grossman, T., Matejka, J., and Fitzmaurice, G. (2010). Chronicle: Capture, Exploration, and Playback of Document Workflow Histories. ACM UIST’10.
20. Grossman, T. and Fitzmaurice, G. (2010). ToolClips: an investigation of contextual video assistance for functionality understanding. ACM CHI '10. 1515-1524.
21. Grossman, T., Fitzmaurice, G., and Attar, R. (2009). A survey of software learnability: metrics, methodologies and guidelines. ACM CHI, 649-658.
22. Hameed, S., Ferris, T., Jayaraman, S. and Sarter, N. (2009). Using Informative Peripheral Visual and Tactile Cues to Support Task and Interruption Management. Human Factors, 51 (2). 126-135.
23. Harrison, S. M. (1995). A comparison of still, animated, or nonillustrated on-line help with written or spoken instructions in a graphical user interface. CHI. 82-89.
24. Hazlewood, W. R., Connelly, K., Makice, K., and Lim, Y. (2008). Exploring evaluation methods for ambient information systems.CHI Ext. Abstracts, 2973-2978.
25. Henzinger, M., Chang, B., Milch, B., and Brin, S. (2005). Query-Free News Search. ACM WWW, 101-126.
26. Horvitz, E. (1999). Principles of mixed-initiative user interfaces. ACM SIGCHI. 159-166.
27. Jackson, S. (2001). Editing computer hardware pro-cedures for multimedia presentation. SIGDOC. 68-72.
28. Karr-Wisniewski, P. and Lu, Y. (2010). When more is too much: Operationalizing technology overload and exploring its impact on knowledge worker productivity. Comput. Hum. Behav. 26, 5 (Sep. 2010), 1061-1072.
29. Kelleher, C. and Pausch, R. (2005). Stencils-based tutorials: design and evaluation. ACM CHI. 541-550.
30. Knabe, K. (1995). Apple guide: a case study in user-aided design of online help. ACM CHI. 286-287.
31. Liu, T.-Y., Huang, Y.-C. and Chen, W.-C. (2001). A novel algorithm for real-time full screen capture system. Multimedia Signal Processing. 395-400.
32. Maekawa, T., Yanagisawa, Y., Sakurai, Y., Kishino, Y., Kamei, K., and Okadome, T. (2009). Web searching for daily living. ACM SIGIR'09, 27-34.
33. Mankoff, J., Dey, A. K., Hsieh, G., Kientz, J., Lederer, S., and Ames, M. (2003). Heuristic evaluation of ambient displays. In ACM CHI '03, 169-176.
34. Matejka, J., Li, W., Grossman, T., and Fitzmaurice, G. (2009). CommunityCommands: command recommend-ations for software applications. UIST 193-202.
35. Matthews, T. (2006). Designing and evaluating glanceable peripheral displays. DIS '06. ACM, 343-345.
36. McFarlane, D. C. and Latorella, K. A. (2002). The scope and importance of human interruption in human-computer interaction design. Hum.-Comput. Interact. 17, 1 (Mar. 2002), 1-61.
37. Novick, D. G. and Ward, K. (2006). Why don't people read the manual? ACM international Conference on Design of Communication. 11-18
38. Palmiter, S. and Elkerton, J. (1993). Animated demonstrations for learning procedural computer-based tasks. Human.-Computer Interaction. 8(3):193-216.
39. Plaisant, C. and Shneiderman, B. (2005). Show Me! Guidelines for Producing Recorded Demonstrations. Visual Languages and Human-Centric Comp. 171-178.
40. Plaue, C. and Stasko, J. (2007). Animation in a peripheral display: distraction, appeal, and information conveyance in varying display configurations. GI, 135-142.
41. Pousman, Z. and Stasko, J. (2006). A taxonomy of ambient information systems: four patterns of design. In ACM AVI '06, 67-74.
42. Sawyer, R.K. (Ed.), The Cambridge Handbook of the Learning Sciences. Cambridge University Press, 2006.
43. Shen, X., Moere, A. V., Eades, P., and Hong, S. (2008). The long-term evaluation of Fisherman in a partial-attention environment. In ACM Beyond Time and Errors: Novel Evaluation Methods For information Visualization, BELIV '08, 1-6.
44. Shneiderman, B. (1983). Direct Manipulation: A Step Beyond Programming Languages. Computer. 16(8):57-69.
45. Twidale, M. B. (2005). Over the Shoulder Learning: Supporting Brief Informal Learning. CSCW, 505-547.
46. Vogel, D. and Balakrishnan, R. (2004). Interactive public ambient displays: transitioning from implicit to explicit, public to personal, interaction with multiple users. ACM UIST 2004, 137-146.
47. Yeh, T., Chang, T., and Miller, R. C. (2009). Sikuli: using GUI screenshots for search and automation. ACM UIST. 183-192.
48. Zeleznik, R. C., Bragdon, A., Liu, C.-C. and Forsberg, A. (2008). Lineogrammer: creating diagrams by drawing. ACM UIST. 161-170.
CHI 2011 • Session: Shortcuts Commands & Expertise May 7–12, 2011 • Vancouver, BC, Canada
2760