Determining and Expressing Interruptibility

1
Determining and Expressing Interruptibility

• Lavar Askew

2
Interruptibility

• Interrupt is the temporary stopping of a task to
  give attention to another task.
• Interruptibility reflects how likely an interrupt
  will affect the completion of the users primary
  task.
• If a users interruptibility is high then it is
  likely that the user may be open to an
  interruption.

3
How Do Interruptions Affect Task Performance

• Users perform interrupted task 5 - 40 slower
  non-interrupted task.
• Surprisingly, there was no correlation between
  the quality of completion of the interrupted
  tasks versus non-interrupted tasks.

4
Opportune Times for Interruption

• Towards the end of the primary task rather than
  the beginning.
• During a temporary break in the execution of the
  primary task.
• Outside of social interaction.
• User defined.

5
Studies

• Where are we in terms of developing
  interruptibility systems?
• Look at a couple of studies which will provide
  some context.

6
Grapevine
7
Grapevine Details

• System developed by IBM to provide information to
  potential communicators to leverage contextual
  information in making decisions about when to
  initiate contact and via which channel. Too
  Much Information, Jim Christensen et al.
• A users computer, mobile device, telephone and
  motion detectors were used to provide the
  contextual information.

8
What was learned from Grapevine Study?

• Do not expect users to do anything extra to
  provide context.
• There must be a way to protect users privacy.
• A substantial semantic gap exists between the
  information that low-level sensors and programs
  can detect and the high-level ability and
  willingness of a person to communicate with
  someone else.

9
James Fogartys work

• Predicting Human Interruptibility with Sensors
• Three key elements to this study
• The human subjects whose actions were record in
  an office setting.
• The human estimators who studied the recordings
  of the human subjects.
• Human coders used to manually simulate sensors.

10
Human Subjects

• Subjects were prompted for interruptibility
  self-reports at random, but controlled, intervals
  averaging two prompts per hour.
• Subjects were asked to rate your current
  interruptibility on a five-point scale, with 1
  corresponding to Highly Interruptible and 5 to
  Highly Non-Interruptible.
• Subjects were present for 672 of these prompts.

11
Human Estimators

• 40 estimator subjects were shown portions of
  the records collected from the video subjects.
• Estimator subjects were given 15 30 seconds
  before the video subjects indicated their
  interruptibility to determine on a scale from 1
  to 5 the interruptibility of the video subject.

12
Results of Human Estimators

• In deciding whether the human subject was
  interruptible on a scale from 1 to 5 the
  estimators had an overall accuracy of 30.7
• Accuracy when off by 1 65.8
• Accuracy when deciding between highly
  non-interruptible and all other choices 76.9
• Chance (always choosing highly
  non-interruptible) 70.6

13
Simulated Sensors

• Wizard of Oz Technique using humans to
  simulate sensor behavior.
• 24 events were identified by coders. These
  events were chosen because they were believed to
  be highly related to interruptibility and
  physical sensors could easily be built to capture
  these events.

14
Sensor Types

• Occupant Related
• Occupant presence.
• Speaking, writing, sitting, standing, or on the
  phone
• Touch of or interaction with desk, table, file
  cabinet, food, drink, keyboard, mouse, monitor,
  and papers.
• Guest Related
• Number of guests present.
• For each guest sitting, standing, talking or
  touching

15
Sensor Type (Cont.)

• Environment
• Time of the day (hour only).
• Aggregate
• Anybody talk (occupant and guest talking).

16
Derivations Applied to Sensors

• Imm whether the event occurred in the 15 second
  interval containing the self-report sample.
• All-N whether event occurred in every 15 second
  interval during N seconds prior to the sample.
• Any-N whether event occurred in any 15 second
  interval during N seconds prior to the sample.
• Count-N - the number of times the event occurred
  during intervals in N seconds prior to the sample.

17
Derivations (cont.)

• Change-N - the number of consecutive intervals
  for which the event occurred in one and did not
  occur in the other during N seconds prior to the
  sample.
• Net-N - the difference in the sensor between the
  first interval in N seconds prior to the sample
  and the sensor in the interval containing the
  sample.

18
Feature Set Defined

• The combination of sensor types and derivations
  define our feature set.
• Must construct interruptibility model based on
  this feature set.
• Which features are the most effective at
  predicting interruptibility?

19
Wrapper-Based Feature Selection Strategy

• Start with empty set of features.
• Add features to a model to determine which ones
  most improve the accuracy of the model.
• Remove those which do not.
• Perform this cycle until there is no change that
  results in improvement.
• Prevents overfitting of data.
• Can be slow because this strategy requires
  repeated application of a machine learning
  technique to learn which features are most
  important.

20
Wrapper-Based Feature Selection Using Decision
Trees

• Used 90 of the data for training and 10 for
  testing.
• Training resulted in the ten features chosen to
  represent the model.
• Model was 82.4 accurate in distinguishing
  between the human subject being Highly
  Non-Interruptible and all others.
• Human Estimators76.9, Chance 70.6

1 Any Talk (Imm)
2 Telephone (Any-30)
3 Time of Day (Hour Only)
4 Desk (Change-120)
5 Monitor (Any-300)
6 Occupant Talk (Net-120)
7 Writing (Count-30)
8 Writing (Count-60)
9 Papers (Count-300)
10 Mouse (All-120)
21
Wrapper-Based Feature Selection (Cont.)
Human Estimator Simulated Sensors
Overall Accuracy 30.7 51.5
Accuracy Within 1 65.8 75.1
22
Whats Been Learned from Fogarty et al. Work

• Determining interruptibility based on passive
  sensors can perform as well or better than humans
  without user explicitly indicating their
  interruptibility or interacting with calendars.
• Users are most likely Highly Non-Interruptible
  when engaged in a task or a social situation.
• In deciding the degree of interruptibility
  estimates of 3 or 4 could be used to initiate
  a negotiated interruption with an ambient
  information display.
• estimates of 1 or 2 could be used to decide
  to initiate with a more direct method.

23
Addressing Issues Raised in Grapevine Study

• No extra context from users - no user input
  necessary
• Privacy - low-level sensors do not transmit or
  record data
• Short comings of low-level sensors in determining
  a high-level concept such as interruptibility -
  use of wrapper-based feature selection and
  decision trees to infer the degree
  interruptibility from low-level sensor data.

24
Interruption Management Systems which Incorporate
Whats Been Learned
25
AuraOrb
26
AuraOrb Details

• AuraOrb uses social awareness cues, such as eye
  contact to detect user interest in an initially
  ambient light notification.
• Once detected, it displays a text message with a
  notification heading visible from 360 degrees.
• Touching the orb causes the associated message
  to be displayed on the users computer screen.
• When user interest is lost, AuraOrb
  automatically reverts back to its idle state.

27
TunnelVision
28
TunnelVision Details

• User presses F10 causing the operating system to
  focus on a single application.
• Operating system temporarily suspends desktop
  notifications while user is in focus.
• Bluetooth-enabled desktop connects to users
  Bluetooth-enabled cell phone disabling the ringer
  and notifying all other Bluetooth devices that
  user is in a non-interruptible state.
• If user goes outside of the 30-foot radius of
  desktop then the cell phones ringer is
  re-enabled.

29
Conclusion

• Defined interruptibility.
• Revealed how interruptions affect task
  performance.
• Provided some socially acceptable opportunities
  for interruption.
• Considered the concerns of those providing the
  context for interruptibility systems.
• Provided some techniques and studies for
  designing interruptibility systems.
• Gave examples of context-aware systems which can
  infer interruptibility.

30
Questions