Title: Instant Messaging and Interruption: Influence of Task Type on Performance
1Instant Messaging and Interruption Influence of
Task Type on Performance
- Mary Czerwinski
- Ed Cutrell
- Eric Horvitz
- Microsoft Research
2What Is Attention?
- There are many definitions of attention
- A function which selectively improves processing
for one item, location, or task at the expense of
others - Different modalities
3Some Motivation
- Miyata Norman (1986)
- Predicted interruptions after important actions
or between task execution and evaluation would be
less harmful when multitasking
4Background 1
- Notifications are most distracting when they bear
surface resemblance to the UI of the task at hand - Gillie Broadbent (1989) Kreifelt McCarthy
(1981) Rhodes, Benoit Payne (2000) - Auditory notifications can be more distracting
than visual notifications - Mollenhauer, et al. (1994)
5Background 2
- People habituate to notifications over time, and
training can help - Hess Detweiler (1994) Altmann Gray (2000)
- Interruptions can be useful!
- OConaill Frohlich (1995)
6McFarlane (1999)
- Examined 4 methods for instant messages
- immediate (requiring immediate user response)
- negotiated (user chooses when to attend)
- mediated (an intelligent agent might determine
when best to interrupt) - scheduled (interruptions come at prearranged time
intervals) interruption methods - Negotiated resulted in good performance
- users may postpone attending to interrupting
messages in these cases - Immediate is fast but users are less efficient
overall
7Previous Work 1
- Degree of disruption depends on what task a user
is doing and when the notification arrives - Relevant notifications are less disruptive than
irrelevant - Czerwinski, Cutrell Horvitz (2000)
8Previous Work 2
- Notifs during Web search task phases
- Planningdeciding what search terms to use
- Executionentering the search terms
- Evaluationsearch the list looking for target
- Disruption worst during
- Execution (chunking)
- Evaluation (? ? ?)
9(No Transcript)
10Day Trading StudyTask Type
- The time to respond to the notification depended
on the ongoing task. - Word and Excel task users more reluctant to
disengage.
11Day Trading StudyInterrupt Position
- Time to respond to the notification was least
when it arrived at the beginning of the task - Easiest phase at which to disengage in this study
12Day Trading StudyTask Type
- The time to respond to the notification depended
on the ongoing task. - Word and Excel task users more reluctant to
disengage.
13Messaging on the Web
- Some trials interrupted with MSNs Messenger
service (v. 2.0) - Relevant messages (design category)
- Irrelevant (factoid about target site)
- Interruptions occurred during one of the three
phases
14(No Transcript)
15(No Transcript)
16Web StudyTime to Switch
- Time to switch to the notification is slowest
when the execution phase is messaged (while a
user is typing). This replicates what we saw in
Day Trading study. - Execution phase slowest to disengage.
17Evaluation Stage Most Harmed
- The evaluation stage was reliably more disrupted
than otherswhen notification came in the
evaluation stage - Not slower to disengage, so slower to process
message and re-instate task context
18Web StudyEffect of Relevance
When notifications are irrelevant, total time on
task and time to resume task after interruption
are both slower
19Why is List Scanning So Susceptible to
Interruption?
- 2 hypotheses
- 1) Visual reorienting is hard to do People
just lose their place in the list Where was
I? - 2) Problem with conceptual reacquisition
Delay from accessing memory of goal What was
I doing?
20How do we find out?
- Sent IM messages to participants while they were
scanning lists of book titles - Two kinds of targets
- Verbatim title (easy visual scan)
- Gist of title (difficult semantic-based search)
- Can we help?
- Used a visual marker to save place in list
21Method
- 12 experienced Microsoft Office 2000 users, aged
25-54, participated in this study. 6 had some
experience with MSNs Messenger. - 64 sets of 80 book titles obtained from MS
library. Each set was an Excel spreadsheet.
Targets were distinctive within a set of 80
titles.
22List Search
- Target always visible at top
- Navigation with Cursor Up/Down or Page Up/Down
keys - Cursorgtmarker
- Search target either verbatim title or gist of
title
23Procedure
- In half of all trials, participants search task
was interrupted with an instant message asking
them a simple math problem. - Half of all trials had gist targets and half
had title targets - Navigation was blocked, with half of the
participants using Cursor Up/Down (Marked) first
and the other half using Page Up/Down (Unmarked).
24Results
- Only report time data accuracy was quite good
- Used log response times to normalize common
skewing variability of RT data
25Overall task times
- IMs slow down task times
- Searches for Gists are slower than for Titles
- No difference in navigation
26Task Times Minus IM Time
- Same pattern as overall task timeseffects not
due to device switching time - Marker only helped title search a
littlenavigational confound
27Summary
- IM is disruptive
- More disruptive for fast, stimulus-driven search
than for slower, semantic-based search tasks - Marker didnt seem to help, but was confounded by
navigation style - Reran study w/o confound, same result
28Memory Effects Notification
- Reran book title search study correcting for
navigation confounds - Removed title from top of page and added a
Remind Me button to list - Users could use button any time
- Recorded where and when users needed to be
reminded of search target
29Methods
- 16 Ss (9 female)
- Intermediate to advanced PC users
- All but 1 had tried IM before
- 2 (marker or not) x 2 (IM or not) within subjects
design - 64 search trials
30Results
31Attention-Based Principles of Notifications 1
- Unless you are absolutely sure the user wants to
know what youre telling them at that moment, be
careful of very salient notifications (this is
from previous work) - Autoarchive in Outlook
- Frequent audio alerts from Messenger
- Users trust is fragile. Once they perceive a
system is unreliable, it is very hard to win them
back (from ongoing work) - Be cautious repeating information it might be
outdated or irritating
32Attention-Based Principles of Notifications 2
- Make notifications situation-aware.
- Look for breakpoints and pauses in users
interactions. Weve identified a few Open or
Save as dialog boxes probably good places to
interrupt typing, selecting, and other direct
interactions probably bad - When possible, use smart monitoring.
- Monitor the user (what are they doing?)
- Content of interruption
- Obvious privacy issues, etc.
33Adaptive Systems and Interaction Group at MSR
- Complementary work on modeling and decision
making for alerts going on in our group - An information-theoretic perspective with
supportive infrastructure. - Work tends to rely on theories that consider
direct preference assessments about outcomes. - User studies will hopefully minimize the needs
for preference elicitation - Results of the work will be useful to the
cost-benefit modeling, decision making work.
34Future Work
- User models of distractibility
- Better cost/benefit user models of the value of
delaying information - Better UI for notifying and reminding user of
what they were doing before the notification - Field studies with teens
- Longitudinal studies of our beta Mobile Manager
software w/cell phones