My Phone and Me: Understanding Peoples Receptivity to Mobile - PowerPoint PPT Presentation

My Phone and Me: Understanding People’s Receptivity to Mobile Notifications Abhinav Mehrotra Veljko Pejovic Jo Vermeulen Robert Hendley Mirco Musolesi

My Phone and Me: Understanding People’s Receptivity to Mobile Notifications Abhinav Mehrotra Veljko Pejovic Jo Vermeulen Robert Hendley Mirco Musolesi

My Phone and Me: Understanding People’s Receptivity to Mobile Notifications Abhinav Mehrotra Veljko Pejovic Jo Vermeulen Robert Hendley Mirco Musolesi

Anticipatory Mobile Computing [Veljko Pejovic and Mirco Musolesi. Anticipatory Mobile Computing: A Survey of the State of the Art and Research Challenges. In ACM Computing Surveys. Volume 47. Issue 3. ACM Press. April 2015.]

Anticipatory Mobile Computing Problem : sending the right information at the right time without annoying the user through notifications, maximizing user’s receptivity.

Anticipatory Mobile Computing Applications : ads, marketing, but also positive behaviour intervention.

Notifications are Beneficial Notifications provide an effortless way to be aware of newly available information in real-time. Communication Online Social Networks System, Tools and Others

Issues with Notifications On arriving at inopportune moments, notifications adversely affect: Ongoing Task Emotional State

Understanding Interruptibility: State-of-the-art Users’ receptivity to a notification is determined by: • their subjective experience in the notification; • the type of application that triggers it; • its time criticality and social pressure. Limitation Cognitive context has not been considered.

Inferring Interruptibility: State-of-the-art Different approaches for predicting interruptibility by using: • context data • notification content Limitation Cognitive context has not been considered.

Bridging the Gap We present the first study to collect objective and subjective data about real-world mobile notifications. We investigate users’ interaction with mobile notifications in different physical and cognitive contexts.

“My Phone and Me” App

Dataset Participants: 20 Minimum questionnaires per participant: 14 Size of the notification sample: 10372 Questionnaire responses: 474

Results

Understanding Response Time Click (c1) Decision Seen Time Time Dismiss (c2) (a) (b) Response Time = Seen Time + Decision Time

Understanding Response Time Click (c1) Decision Seen Time Time Dismiss (c2) (a) (b) Seen Time = time from the notification arrival until the notification was seen by the user

Understanding Response Time Click (c1) Decision Seen Time Time Dismiss (c2) (a) (b) Decision Time = time from the moment a user saw a notification until the time they acted upon it (click or dismiss)

Impact on Seen Time Alert Modality Task Type Task Complexity Task Completion Level Relationship with Sender

The Impact of Alert Modality on Seen Time F = 26.41, p < 0.001 8 Seen Time (mins) 6 4 2 0 Silent Vibrate Sound Sound with vibrate Note 1 : we ignored notifications that arrived when the user was already engaged with the phone. Note 2 : similar observations are reported in Pielot et al. MobileHCI’14 .

The Impact of Ongoing Task Type on Seen Time F = 2.963, p = 0.013 10 8 Seen Time (mins) 6 4 2 0 Note : the information users provided about the ongoing task was manually classified by two coders into six categories.

The Impact of Ongoing Task Complexity on Seen Time We encoded the reported task complexity values as ordinal numbers. Strongly disagree =1 Somewhat disagree =2 Neutral =3 Somewhat agree =4 Strongly agree =5 Spearman’s rank correlation coefficient = -0.18 [p < 0.005]. User’s attentiveness increases (reducing the seen time) with the increase in the complexity of an ongoing task.

Impact on Decision Time Alert Modality Task Type Task Complexity Task Completion Level Relationship with Sender

The Impact of Sender-Recipient Relationship on Decision Time F = 2.429, p = 0.009 14 12 Seen Time (seconds) 10 8 6 4 2 0

Why Do Notifications Become Disruptive? Alert Modality Task Type Task Complexity Task Completion Level Relationship with Sender

The Role of Ongoing Task Type F = 13.03, p < 0.001 Work Highest Traveling Leisure Communicating and Maintenance Lowest Idle

The Role of Ongoing Task Completion Level F = 19.43, p = 0.013 Highest Middle of a task Finishing a task Starting a task or Idle Lowest

The Role of Sender F = 3.987, p = 0.013 Highest Sender is not a person and subordinate at work Colleagues and service providers Other senders Extended family Lowest

Frequent and Not Frequent Senders • Quite interestingly, the decision time is higher for the notifications from less frequently contacted senders. • Conjecture: maybe the content less familiar to users?

The Role of Ongoing Task Complexity Spearman’s rank correlation coefficient = 0.477 [p < 0.001]. Perceived disruption increases with the increase in the complexity of an ongoing task.

Understanding the Acceptance of Notifications - How did you handle the notification when you first saw it? - Select all factors that made you decide to click/dismiss the notification.

Why do Users Accept (click) Notifications? Accept (Disruptive Option Accept notifications) Sender is important 31.546 25.926 The content is important 27.129 33.333 The content is urgent 14.511 20.370 The content is useful 31.546 35.185 I was waiting for this notification 15.773 11.111 20.189 16.667 The action demanded by the sender does not require a lot of effort At this moment, I was free 37.224 18.519

Why Do Users Dismiss Notifications? Option Dismiss Sender is not important 19.565 The content is not important 40.580 The content is not urgent 43.478 The content is not useful 38.406 3.623 The action demanded by the sender does require a lot of effort I was busy 19.565

Does Personality Matter? Linear regression model with the five personality traits as independent variables and reported disruption, seen time and decision time of notifications as dependent variables. • R 2 = 0.737 [for reported disruption] • R 2 =0.9007 [for seen time] • R 2 =0.9035 [for decision time]

Implications • It is a good idea to defer non-useful notifications at busy moments. • It is possible to improve notification presentation by displaying useful content. • We demonstrate that we can build a personality- dependent interruptibility model.

Questions? Mirco Musolesi University College London E: m.musolesi@ucl.ac.uk W: http://www.ucl.ac.uk/~musolesm T: @mircomusolesi

Questionnaire I Did you notice the alert (e.g., vibration, sound, flashing LED) for this notification when it first arrived? How did you handle the notification when you first saw it? Factors that made you to decide to click/dismiss the notification. What best describes your relationships to the sender .

Questionnaire II Please describe what the notification was about. Please describe what activity you were involved with when you received the notification. Activity performed when the notification arrived. Complexity of the activity performed when the notification arrived. Level of the disruption of the notification.