Snooze! Investigating the User-Defined Deferral of Mobile Notifications

Snooze! Investigating the User-Deﬁned Deferral

of Mobile Notiﬁcations

Dominik Weber

, Alexandra Voit

, Jonas Auda

, Stefan Schneegass

, Niels Henze

VIS, University of Stuttgart

Stuttgart, Germany

{ﬁrstname.lastname}@vis.uni-stuttgart.de

University of Duisburg-Essen

Essen, Germany

{ﬁrstname.lastname}@uni-due.de

ABSTRACT

Notiﬁcations on mobile devices are a prominent source of

interruptions. Previous work suggests using opportune mo-

ments to deliver notiﬁcations to reduce negative effects. In

this paper, we instead explore the manual deferral of noti-

ﬁcations. We developed an Android app that allows users

to “snooze” mobile notiﬁcations for a user-deﬁned amount

of time or to a user-deﬁned point in time. Using this app,

we conducted a year-long in-the-wild study with 295 active

users. To complement the ﬁndings, we recruited 16 further

participants who used the app for one week and subsequently

interviewed them. In both studies, snoozing was mainly used

to defer notiﬁcations related to people and events. The rea-

sons for deferral were manifold, from not being able to attend

notiﬁcations immediately to not wanting to. Daily routines

played an important role in the deferral of notiﬁcations. Most

notiﬁcations were deferred to the same day or next morning,

and a deferral of more than two days was an exception. Based

on our ﬁndings, we derive design implications that can inform

the design of future smart notiﬁcation systems.

ACM Classiﬁcation Keywords

H.5.m. Information Interfaces and Presentation (e.g. HCI):

Miscellaneous

Author Keywords

Mobile notiﬁcations; interruptions; mobile HCI; notiﬁcation

management; deferral; in-the-wild; in-situ.

INTRODUCTION

Notiﬁcations are a key feature of current smartphones. They

are used to proactively inform users about incoming messages,

emails as well as social network updates and events [30]. Noti-

ﬁcations use visual cues, auditory signals, and haptic feedback

to gain the user’s attention [14]. An increasing number of mo-

bile applications issue notiﬁcations for various purposes [34].

As a result, users of mobile devices currently receive a large

Permission to make digital or hard copies of all or part of this work for personal or

classroom use is granted without fee provided that copies are not made or distributed

for proﬁt or commercial advantage and that copies bear this notice and the full citation

on the ﬁrst page. Copyrights for components of this work owned by others than the

author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or

republish, to post on servers or to redistribute to lists, requires prior speciﬁc permission

and/or a fee. Request permissions from [email protected].

MobileHCI ’18, September 3–6, 2018, Barcelona, Spain

ISBN 978-1-4503-5898-9/18/09. . . $15.00

DOI: https://doi.org/10.1145/3229434.3229436

(a) Duration Method (b) Point-in-Time Method

Figure 1. The available snooze options depend on the selected method.

Left: Duration. Right: Point-in-time (depending on the time of the day).

number of notiﬁcations on a daily basis [30, 34]. The recipi-

ents of notiﬁcations, especially on mobile devices which ac-

company its users throughout the day, are coping with a mostly

uncontrolled stream of proactively delivered information. Suit-

ably controlling this stream is a major research challenge that

is under heavy investigation nowadays.

Notiﬁcations can induce negative effects. They can be distract-

ing, might cause negative emotions, or are just not important

for their recipients [26, 30, 32, 34]. One strand of research

aims to reduce distraction by ﬁnding more suitable moments

for notiﬁcation delivery. With SMS and instant messaging (IM)

services generating numerous notiﬁcations, further research is

needed to develop more appropriate notiﬁcation management

services [4]. Previous work found that delaying the delivery

of notiﬁcations to opportune moments helps to reduce distrac-

tion [10, 22, 23, 24, 26, 36]. Contextual information about

the user can be used to determine suitable moments to deliver

notiﬁcations. These moments promise to alleviate negative

effects on the recipient of notiﬁcations.

Despite the increasing interest in ﬁnding opportune moments

for delivering notiﬁcations, little is known about the type of

notiﬁcation that should be delayed and how long users like the

delay to be. Therefore, we developed the Android application

NHistory that enables users to snooze notiﬁcations similar to

snoozing an alarm clock. In contrast to previous approaches,

the app lets the recipient of a notiﬁcation specify the moment

when the notiﬁcation should be redelivered. Notiﬁcations can

be snoozed for a speciﬁc duration or to a later point-in-time,

and NHistory reissues them after the speciﬁed period. Users

can also access a history of all notiﬁcations they have received,

enabling them to go through the notiﬁcations and snooze de-

sired notiﬁcations to more appropriate moments. By snoozing

notiﬁcations to more suitable times, users can reduce the num-

ber of pending notiﬁcations in the notiﬁcation drawer. While

this is indeed a positive aspect of the app, we want to note

that this does not necessarily mitigate adverse effects on the

notiﬁcations’ recipient. In some cases, it might be even worse

than an uncontrolled notiﬁcation delivery, because a snoozed

and redelivered notiﬁcation may induce a second interruption.

However, observing how a recipient of a notiﬁcation post-

pones its arrival enables us to gain insights into the types of

notiﬁcations users want to defer and how long users would

like the deferral to be.

In this paper, we report a year-long in-the-wild study with 295

active users. The goal of the study was to understand which

notiﬁcations users want to defer and the periods of time users

want to defer them. We focused on deriving insights into how

users postpone notiﬁcations rather than eliminating negative

effects that might inﬂuence the notiﬁcation recipient directly.

We collected usage data including applications that issued no-

tiﬁcations, how long notiﬁcations were snoozed for, and when

they were re-triggered. To complement the collected quantita-

tive data, we conducted a second study with 16 participants for

one week and subsequently interviewed them. In both studies,

snoozing was mainly used to defer notiﬁcations related to peo-

ple and events. Reasons for the deferral of notiﬁcations were

manifold and aligned with daily routines. Most notiﬁcations

were deferred to the same day or the next morning, indicating

an upper bound for the deferral of notiﬁcations. Based on our

ﬁndings, we derive design implications that can inform the

design of future smart notiﬁcation systems.

RELATED WORK

Notiﬁcations received on mobile devices are used to inform the

user about incoming messages, emails, social network updates,

and events [30, 39]. Using notiﬁcations, these devices compete

for the attention of their user [12, 19]. Besides the informative

value of notiﬁcations, the delivery can cause negative effects

such as interruption of the user. In the following, we present

previous work on notiﬁcations and interruptions as well as

opportune moments for presenting notiﬁcations to users.

Notiﬁcations and Interruptions

The number of generated notiﬁcations is constantly increas-

ing [30]. Smartphone users themselves often do not realize

how many notiﬁcations they receive [40]. Sahami et al. found

that a large number of notiﬁcations is issued by messaging

applications [34, 38]. On the one hand, users value notiﬁ-

cations issued by such applications. On the other hand, not

all notiﬁcations that users receive are considered important.

Church and Oliveira compared SMS to instant messaging ap-

plications like WhatsApp [4]. Their study revealed several

concerns regarding WhatsApp messages and notiﬁcations, e.g.,

coping with too many messages or interruptions, and the fear

of missing business-related messages if notiﬁcation modalities

are switched off. Understanding how users handle messaging

notiﬁcations might help to build messaging services which do

not overload users by issuing too many notiﬁcations.

A notiﬁcation recipient might beneﬁt from valuable informa-

tion that he or she receives in a proactive manner [7, 19].

However, previous studies show that notiﬁcations can cause

negative effects to its recipient. Ill-timed notiﬁcations can also

distract or interrupt the recipient [16, 30]. Further negative

effects include decreased productivity and slower and more

error-prone performance [1, 3, 11, 28, 35].

Merely disabling notiﬁcations is not a suitable solution to alle-

viate corresponding adverse effects on the recipient of a notiﬁ-

cation [32, 33]. The desire to meet social expectations or the

fear to appear rude by not replying to messages on time are rea-

sons for keeping notiﬁcations enabled. Further, notiﬁcations

act as reminders for important events. Stawarz et al. investi-

gated efﬁcient reminders for taking medicine [37]. Time-based

reminders offer only a small beneﬁt to users because often the

reminder cannot be postponed. Immediate action is required

by the user so that taking medicine is not forgotten. Stawarz et

al. propose that users should be able to defer notiﬁcations is-

sued by reminders if they are not able to respond immediately.

To reduce the negative effects of the increasing number of

notiﬁcations, previous research focused on developing models,

rules, and systems that manage notiﬁcation delivery. However,

users might not accept a notiﬁcation management systems that

removes important notiﬁcations [19].

Opportune Moments

To identify moments in which notiﬁcations can be best pre-

sented to the user, researchers exploited contextual informa-

tion, notiﬁcation content, and personal traits. Corno et al.

introduced a smart notiﬁcation system that uses machine learn-

ing to manage incoming notiﬁcations [6]. These algorithms

use information about the context of the user and the user itself

such as location and user’s activity. Several approaches used

sensor data to develop context-aware notiﬁcation systems [12,

15]. Kern and Schiele introduced a model that makes use of

body-worn sensor data. The model supports notiﬁcation clas-

siﬁcation according to the user’s current context to determine

his or her interruptibility. Horvitz et al. introduced BusyBody,

a system running on a desktop computer that predicts the cost

of an interruption on a user [13]. It analyzes desktop events

and the user’s context (e.g., if he or she is speaking or not) to

train a model. Mehrotra et al. used the context of a notiﬁcation

recipient in combination with the content of the notiﬁcation

to realize a non-disruptive notiﬁcation mechanism [18]. No-

tiﬁcations were grouped into different categories according

to the application that issued the notiﬁcation as well as the

relationship between sender and receiver. They showed that

their machine learning based predictors could outperform user-

deﬁned rules for notiﬁcation delivery on smartphones. Further,

Mehrotra et al. proposed an interruptibility management solu-

tion for mobile notiﬁcations [17]. Their system extracts rules

to handle notiﬁcations automatically based on how the user is

interacting with his or her device.

Yuan et al. proposed a model which can determine the inter-

ruptibility of its user [41]. The model determines if the user

is available to react to an incoming notiﬁcation using sensory

data and additional information such as the mood of the user.

They found that including personality traits is important for

predicting interruptibility. Siewiorek et al. proposed SenSay,

a context-aware mobile phone [36]. It uses sensors to acquire

contextual information about its user. Using this information

the device can switch to the uninterruptible state. In this state,

unwanted interruptions will not be delivered to the user.

Pielot et al. investigated if mobile phone usage patterns can

be utilized to train a machine learning model that is able to

detect boredom [31]. They suggest that bored people may

be more appreciative of incoming notiﬁcations. Dingler et

al. investigated if detected boredom can be used to engage

a user in micro-learning sessions through notiﬁcations [9].

Indeed people search for stimulation during boredom, but

they found that a mentally demanding task is not suitable. To

voluntary engage users to interact with recommended content

Pielot et al. used a machine learning approach to determine

opportune moments for notiﬁcation delivery [29]. Higher user

engagement was observed if notiﬁcations were issued at these

moments. Further, they suggest observing past interest in

notiﬁcation content to reduce future interruptions.

To further mitigate negative effects caused by interruptions

previous research investigated the suitability of breakpoints

for notiﬁcation delivery. A breakpoint promises to be more

suitable for notiﬁcation delivery because they occur between

two consecutive activities [21]. Recipients of interruptions

during these moments may not be as strongly inﬂuenced by

interruption related adverse effects as they would be while per-

forming a speciﬁc activity or task. Fischer et al. reported that

people attended to notiﬁcations quicker at the end of a mobile

interaction (i.e., calling a contact or reading SMS) compared

to notiﬁcations received at random times [10]. Different tools

were suggested to cope with the number of incoming notiﬁca-

tions. Ho and Intille used data from accelerometers to automat-

ically detect transitions between physical activities [12]. They

suggested that these transitions might be suitable for reducing

negative effects caused by interruption from mobile devices.

Okoshi et al. proposed Attelia, a service that identiﬁes break-

points for notiﬁcation delivery [22, 23, 24, 25]. Attelia runs

on the user’s smartphone and can detect breakpoints of the

user’s activity on his or her mobile device based on running

applications and machine learning techniques. Further, it can

detect physical breakpoints through smartwatches. Investi-

gating breakpoints in the wild, Okoshi et al. conducted a

large-scale study with their breakpoint detection system in-

cluded into a popular Android application. They reduced the

response time to the delayed notiﬁcations signiﬁcantly. They

also observed a continuously increasing number of clicks as

well as a higher level of user engagement. Park et al. proposed

the breakpoint-based Social Context-Aware smartphone Notiﬁ-

cation system (SCAN) [26]. SCAN can identify breakpoints to

which it defers incoming notiﬁcations. Park et al. reported that

SCAN could help the participants to better focus on their social

interaction. Pejovic and Musolesi proposed InterruptMe, a

library for interruption management for Android [27]. With

the use of contextual information about the user, the library

determines whether he or she is interruptible. InterruptMe

can inform other apps when an opportune moment for an

interruption occurs.

Figure 2. Snoozing a notiﬁcation via the notiﬁcation drawer. 1. User dis-

misses the notiﬁcation. 2. NHistory brieﬂy shows a “snooze notiﬁcation”.

3. Tapping on the snooze notiﬁcation opens the snooze options. 4. After

the selected duration, the notiﬁcation is re-triggered.

In summary, previous work showed that delaying the delivery

of notiﬁcations to opportune moments can partially reduce no-

tiﬁcations’ negative effects. A body of work focused on auto-

matically ﬁnding opportune moments for notiﬁcation delivery.

However, little is known about which types of notiﬁcations

and to what times users would defer notiﬁcations manually.

SYSTEM

In contrast to previous work, we investigate the manual defer-

ral of notiﬁcations to study reasons for deferring notiﬁcations,

selected moments, and types of notiﬁcations to better under-

stand why and how users defer notiﬁcations. Therefore, we

developed an Android application that enables users to defer

notiﬁcations manually. In the following, we ﬁrst summarize

Android’s notiﬁcation mechanisms and afterward describe

how NHistory extends the Android system.

Notiﬁcations in Android

Notiﬁcations play a central role in the Android mobile oper-

ating system. They consist, at the minimum, of a small icon

and two lines of text. Notiﬁcations are accessible from the

notiﬁcation drawer; a list of active notiﬁcations that can be

accessed by swiping down from the top of the screen. An-

droid’s status bar displays icons to indicate active notiﬁcations

in the notiﬁcation drawer. Newer versions of Android dis-

play notiﬁcations on the lock screen as well. Notiﬁcations are

ephemeral. Users can dismiss single or all active notiﬁcations

at once. Dismissing a notiﬁcation removes it from the system,

an action that cannot be undone. An active notiﬁcation might

be seen as a “nagging reminder” to take action while dismiss-

ing a notiﬁcation might cause the user to forget about taking

said action. Users have to decide if they want to respond to

the notiﬁcation (e.g., reply to an instant message), dismiss the

notiﬁcation, or just ignore it until a later point in time [5].

NHistory

We developed NHistory to investigate the user-deﬁned deferral

of mobile notiﬁcations. The app provides a timeline of all ac-

tive and dismissed notiﬁcations, sorted by the time of creation.

Further, the app allows users to dismiss notiﬁcations and auto-

matically re-triggers them after a user-deﬁned duration or at a

user-deﬁned point-in-time. This behavior of temporarily mut-

ing is known from alarm clocks and generally referred to as

(a) Starred Notiﬁcations (b) Notiﬁcation History (c) Snoozed Notiﬁcations (d) App Blacklist

Figure 3. Screenshots of the NHistory app. From left to right: List of starred notiﬁcations, notiﬁcation history showing active and past notiﬁcations

(main view), list of snoozed notiﬁcations, and blacklist to exclude notiﬁcations from speciﬁc apps from the history.

“snoozing” an alarm. To enable this functionality, the app uses

Android’s Notiﬁcation Listener API [8]. When ﬁrst opening

NHistory, users are informed about its data collection. To use

the app, users have to agree to the data collection explicitly;

declining closes the app. After accepting the data collection,

the user is prompted to grant NHistory the permission to access

notiﬁcations. Then, the app’s main view is shown (Figure 3b).

The history view shows a list of active and past notiﬁcations.

Users can tap on notiﬁcations to expand them. “Ongoing” no-

tiﬁcations, like downloads, are ignored by NHistory. Clicking

on the star icon next to a notiﬁcation allows users to bookmark

them. Starred notiﬁcations are shown to the left of the history

(Figure 3a) and can be removed by clicking on the star icon

again. To the right of the history, notiﬁcations that are cur-

rently snoozed are listed (Figure 3c). In the settings, users can

set a limit of how many notiﬁcations should be stored in the

history or exclude speciﬁc apps from appearing in the history

using a blacklist (Figure 3d). Users can access NHistory by ei-

ther clicking on its icon in the launcher or through an optional

persistent notiﬁcation in the notiﬁcation drawer.

Snoozing Notiﬁcations

NHistory offers multiple ways to snooze notiﬁcations. In the

app itself, users can long-press on a notiﬁcation or expand a

notiﬁcation and subsequently click on a snooze button. Active

notiﬁcations are automatically dismissed from the notiﬁca-

tion drawer when snoozing them. While the Android system

does not allow modifying the notiﬁcation drawer, we also im-

plemented an option to snooze notiﬁcations from the drawer

directly. NHistory detects when a user dismisses a notiﬁcation

from the notiﬁcation drawer. The app then creates a tempo-

rary notiﬁcation in its place, that is shown for ﬁve seconds.

Clicking this temporary “snooze notiﬁcation” allows the user

to trigger the snooze action for the dismissed notiﬁcation (see

Figure 2). The user is then prompted to deﬁne how long or un-

til when the notiﬁcation should be snoozed. We implemented

the two “snooze” methods duration and point-in-time with

corresponding options. Both methods provide a set of eight

options. The duration method features ﬁxed options, while

the point-in-time options depend on the time of the day (see

Figure 1). Both methods feature a custom duration or custom

point-in-time as the ﬁrst option. On the ﬁrst start of the app,

one of the snooze methods is randomly assigned. Users can

change the assigned method in the settings.

Data Collection

NHistory collects information about the device, notiﬁcation

meta-data, and interaction with the app. Upon installation

of NHistory, a random identiﬁer is created. All requests to

our server use a secure connection and include the identiﬁer.

Device information is sent periodically, including device type,

Android version, and system language. The app records all

major user interaction including starring and snoozing notiﬁca-

tions, and blacklisting apps from the history. Further, package

names and timestamps of all notiﬁcations created on the device

are recorded. We did not record any text or information that

could be used to identify users. All recorded data is queued

for sending until a Wi-Fi connection is established.

IN-THE-WILD STUDY

To gain insights into the types of notiﬁcations users want to

defer, and for how long, we conducted an in-the-wild study.

Method and Participants

We released NHistory on the Google Play Store as a free down-

load. People across the globe were able to download and use

the application. The app automatically reported anonymized

usage and notiﬁcation meta-data back to us. Users who down-

loaded the app were informed about the data collection on

the ﬁrst start of the app. All users had to consent to the data

collection to use the app.

Between January 12, 2017, and January 12, 2018, NHistory

was installed on

1, 555

devices. According to the Google Play

Store statistics, the app was downloaded from

countries,

with most downloads originating from India (

43.08%

), the

United States (

12.43%

), and Germany (

4.25%

). The most

popular device languages were variants of English (

81.28%

followed by German (

3.67%

). Android versions 6.0 (

33.89%

7.0 (

27.65%

), and 7.1 (

16.98%

) contributed the most to the

user base. Only

users installed the app on Android tablets;

the remaining

1, 542

installs were on smartphones. Thirty

users rated the app, resulting in an average rating of 4.37 stars

(1=worst; 5=best).

Results

Of the

1, 555

users who downloaded the app, over half (

876

)

agreed to the terms of the study on the ﬁrst start. It is important

to note that users tend to try out free apps and quickly uninstall

them if they do not fulﬁll their expectations. Indeed, a number

of users uninstalled the app right after the setup.

581

users

used the app for less than two days. Thus, we excluded them

from our analysis. The remaining

295

users used the app

between 2 and 360 days (M=46; SD=68; Md=15 days).

In-The-Wild Study Controlled Study

Category Normalized% Total Events Users Apps Normalized% Total Events Users Apps

Calendar/Reminder 18.68 430 41 26 4.47 5 3 3

Email 8.19 111 30 11 13.66 21 5 5

Game 2.95 22 6 7 3.33 1 1 1

Health/Fitness .25 15 5 5 2.76 3 2 2

Media 3.08 26 12 9 .35 1 1 1

News 1.82 8 5 6 4.39 4 2 3

Phone 8.45 102 22 10 4.44 2 2 1

Shopping/Finance 3.75 26 16 13 2.22 1 1 1

SMS/IM 28.23 251 64 30 49.15 48 13 6

Social 9.20 73 28 20 4.11 9 3 4

System 6.56 40 24 12 8.03 5 4 2

Tool 8.83 87 27 36 3.07 9 2 4

Table 1. Normalized distribution of the initial snooze events for all users and all categories. Left: Values for the in-the-wild study (1,191 events).

Right: Values for the controlled study (109 events). In both studies, notiﬁcations of the category SMS/IM were proportionally snoozed most often.

Notiﬁcations and Apps

In total, we logged

20, 345, 277

notiﬁcations from

3, 667

apps.

Users had on average

apps notifying them (SD=

). We

recorded on average

1, 960

notiﬁcation events per user per

day (SD=

3, 942

; Md=

832

). At ﬁrst glance, these numbers

seem unusually high. The reason for these numbers is that

in Android updating an existing notiﬁcation is realized by

replacing the original notiﬁcation. Some apps continuously

update notiﬁcations in the background, e.g., to display loca-

tion updates or battery statistics. Although these background

updates might happen every other second, they are often not

noticeable to the user because they happen silently. Since

we did not record the text of notiﬁcations, we were unable

to ﬁlter these updates. The apps that created the most no-

tiﬁcations were Google Maps (

2, 049, 889

notiﬁcations;

210

users), Power Clean (

1, 665, 971

notiﬁcations;

users), and

WhatsApp (1, 325, 423 notiﬁcations; 165 users).

Most notiﬁcations were created between 8pm and 10pm, peak-

ing at 9pm. At 4am the least number of notiﬁcations were

created. Regarding the days of the week, the notiﬁcations were

evenly distributed over weekdays and the weekend.

Notiﬁcation History Blacklist

109

295

users made use of the option to exclude apps

from the history. These users excluded between

and

435

apps (M=

; SD=

; Md=

), totaling in

2, 760

different apps.

Some users added apps to the blacklist that did not yet post

notiﬁcations. The apps excluded by most users were the An-

droid OS (

users), the Android System UI (

users), and

the Google Play Store (42 users).

Starred Notiﬁcations

Only

295

users starred notiﬁcations. Users starred be-

tween

and

notiﬁcations (M=

; SD=

; Md=

) for at least

one hour, totaling in

159

star events from

apps. Notiﬁ-

cations from apps that were starred by more than one user

include Google Calendar (

events;

users), Facebook (

events; 5 users), and Google Keep (6 events; 4 users).

Snoozed Notiﬁcations

We recorded

2, 648

snooze events from

151

users and

219

apps. Since users interested in the snooze functionality would

likely try it after installing the app, we excluded snooze events

Snoozing of

Compared to Health/Fitness SMS/IM

T diff. p T diff. p

Calendar/Reminder 5.00

< .001* 40.61 = .002

Email 27.00

< .001* 51.50

< .001*

Game 3.20 = .002 39.50

< .001*

Health/Fitness — — 9.50

< .001*

Media 2.67 = .001 38.00

< .001*

News 2.50 = .002 25.50

< .001*

Phone 17.50

< .001* 64.00

< .001*

Shopping/Finance 3.50 = .001 65.00

< .001*

SMS/IM 9.50

< .001* — —

Social 7.00

< .001* 73.50

< .001*

System 20.00

< .001* 61.00

< .001*

Tool 7.00

< .001* 34.47 = .001

Table 2. Results of the post-hoc analysis with Wilcoxon signed-rank

tests of the normalized snooze categories (in-the-wild study). p-values

are Holm-Bonferroni adjusted. Signiﬁcant differences are marked with

“*”, signiﬁcant less snoozed categories with “

”, and signiﬁcantly more

snoozed categories with “

”.

from the day when NHistory was installed, resulting in

2, 390

snooze events from

129

users and

185

apps.

users were

randomly assigned the point-in-time (pit) method during in-

stallation, and

users were assigned the duration method.

users stayed with their assigned method (

duration,

pit),

users switched to the other method (

from duration to pit,

from pit to duration), and

users switched but returned

to their assigned method (7 duration, 11 pit).

Looking closer at the source of the

2, 390

snooze events

showed that

1, 191

individual notiﬁcations were snoozed.

728

notiﬁcations were snoozed once, and

463

notiﬁcations

were snoozed multiple times, with a single notiﬁcation be-

ing snoozed

times by a user. We will now report on the

1, 191

initial snooze events, followed by the re-snooze events.

The

1, 191

initial snooze events by

129

users break down to

402

snooze events per user (M=

; SD=

; Md=

). The

apps that were snoozed the most were the Google Calendar

(

340

events;

users), WhatsApp (

events;

users), and

Outlook (58 events; 4 users).

To abstract from single apps, two researchers independently

categorized the

185

apps. The resulting

categories are based

on Google Play Store listings and prior literature [34] (see Ta-

ble 1). Disagreements were discussed until an agreement was

reached. The left part of Table 1 shows the normalized distribu-

Figure 4. Normalized distribution of selected options for all 1,191 initial

snooze events, both methods, and all users (in-the-wild study). Duration:

797 events from 75 users. Point-in-time: 394 events from 64 users.

tion of snooze events based on the categories. The categories

with the most snooze events were SMS/IM (

28.23%

), Calen-

dar/Reminder (18.68%), and Social (

9.20%

). We conducted

a Friedman test to compare the normalized data of snoozed

notiﬁcations from different categories. The results show that

the category has an inﬂuence on the users’ snoozing behavior

(χ

(11) = 189.96, p < .001).

We conducted a post-hoc analysis with Wilcoxon signed-

rank tests and applied a Holm-Bonferroni correction (see

Table 2). The results show that notiﬁcations from the cat-

egory Health/Fitness were signiﬁcantly less often snoozed

than from the categories SMS/IM, Calendar/Reminder, Social,

Tool, Phone, Email, and System. Further, notiﬁcations from

the category SMS/IM were signiﬁcantly more often snoozed

than all other categories except Calendar/Reminder and Tool.

The usage of the duration and point-in-time methods was

equally distributed. Normalized over all users,

54%

of the

1, 191

initial snooze events were executed using the dura-

tion method and

46%

using point-in-time. We normalized

the selected options for all users and both methods (see Fig-

ure 4). The most popular options for the duration method were

“5 minutes” (

29.32%

), “1 hour” (

16.88%

), and “30 minutes”

(

14.99%

). The custom duration option was mainly used to ﬁll

the gaps between the pre-deﬁned options, with most durations

being between 10-15 minutes and 2 hours. The median custom

duration was 1 hour and the maximum duration 5 days. For

the point-in-time method, the most popular options were “next

full hour” (

24.42%

), picking a custom time (

17.50%

), and

“today evening” (

11.87%

). For custom points-in-time, users

mostly selected times dividable by 15 minutes, with a median

snooze time of 5.62 hours, and a maximum point-in-time of

4.67 days in the future.

Figure 5 shows the posted, snoozed, and re-triggered times of

the initial snooze events. Most notiﬁcations that were snoozed

were posted in the morning and early noon. Users quickly

Figure 5. Plots showing the distribution of the 1,191 initial snooze events

from the in-the-wild study. The line charts show the time distribu-

tion when the notiﬁcations were posted (top), snoozed (middle), and re-

triggered (bottom). The bar charts show the median reaction time (top)

and median snooze time (middle).

attended the notiﬁcations, resulting in low reaction times

(Md=

35min

). Consequently, the distribution of the snooze

events follows the posted events, peaking in the morning and

dropping over the course of the day. During the day we ob-

served notiﬁcations being snoozed for shorter time spans. In

the evening and until after midnight the time spans notiﬁca-

tions were snoozed for increased by several hours, as users

snoozed the notiﬁcations until the next morning. Most snoozed

notiﬁcations were re-triggered during the day, between 8am

and 10pm. We can see peaks at 9am, 12pm, 3pm, 6pm, and

9pm. 9am being the highest peak can be explained by notiﬁca-

tions that were snoozed for a short time in the early morning

and notiﬁcations from the previous day. These times are likely

inﬂuenced by the pre-deﬁned options of NHistory but might

also correlate with before and after work hours, and lunch

breaks. We want to highlight that

79.26%

of the notiﬁcations

were posted, snoozed and re-triggered on the same day. Only

16.54%

were re-triggered on the following day and

4.20%

over multiple days. This results in an overall mean deferral

time of

689

minutes (SD=

1, 543

; Md=

274

), between a notiﬁ-

cation being posted and eventually re-triggered.

As mentioned earlier,

463

notiﬁcations were snoozed more

than once (Min=

; Max=

; Md=

). These notiﬁcations were

mainly from the categories Calendar/Reminder (

55.94%

SMS/IM (

15.77%

), and Email (

10.15%

). Looking at the times

when these notiﬁcations were initially posted and ﬁnally re-

triggered, we still found that most were on the same day

(

69.61%

), some on the following day (

20.49%

), and more

than two days being an exception (9.91%).

Summary

Temporarily snoozing was more popular than permanently

starring notiﬁcations, indicating the ephemeral nature of no-

tiﬁcations. We assume that users only snoozed particularly

important notiﬁcations that they were unable to attend di-

rectly. Users mainly snoozed notiﬁcations from the categories

SMS/IM and Calendar/Reminder. They were fast to snooze

notiﬁcations, with overall low reaction times. The number of

snooze events spiked in the morning and declined over the

course of the day. Most notiﬁcations were re-triggered in the

morning, at noon and in the evening. These times likely corre-

late with before and after work hours, and lunch breaks. We

found that users snoozed notiﬁcations mainly to the same day.

Only few notiﬁcations were snoozed to the following day and

even less to the day after.

CONTROLLED STUDY

The in-the-wild study provided insights into the types of notiﬁ-

cations and times users are interested in deferring notiﬁcations.

However, we know little about the users’ motivation. To com-

plement the quantitative results, we conducted a second in-situ

study with a smaller set of participants and subsequently inter-

viewed them [20].

Method

In this more controlled study, we invited participants to use

NHistory for one week. During this study, we used the same

version of NHistory as for the in-the-wild study and collected

the same data. Additionally, we conducted semi-structured

interviews at the end of the controlled study to gain further in-

sights. In the interviews, we asked the participants to estimate

how many notiﬁcations they receive on a daily basis, how the

number of notiﬁcations affects them, and how they deal with

interruptions. Further, we asked them about their opinions on

NHistory and the provided functionality.

We individually invited the participants to our lab. All par-

ticipants signed a consent form, informing them about the

procedure of the study and the data collection. Further, they

were informed that they are allowed to withdraw their study

participation at any time. We then asked them to ﬁll out a

questionnaire about demographic data. Afterwards, we intro-

duced them to NHistory. We installed the app on their personal

Android smartphones and walked them through the different

features. We then asked them to use the app as they see ﬁt for

one week. We explicitly told them that they do not have to use

it at all if they ﬁnd it unnecessary. After seven full days of us-

age, we again invited the participants to our lab and conducted

interviews with a duration of approximately

minutes each.

Participants received EUR 15 for their participation.

Participants

We invited

participants to use NHistory for one week. One

participant withdrew the study participation; thus, we excluded

this participant from our evaluation. The remaining

partic-

ipants (

female,

male) were

years old (M=

26.50

;

SD=

4.07

). Four of them were PhD students, eight students,

two software engineers, and two teachers. All but the teachers

had a technical background.

Results

The device language of the participants’ smartphones was set

to German in

cases, and to English for the other four cases.

Most participants used a device running Android 6.0 (

) and

5.0 (

). Android 5.1, 7.0, and 7.1 were used once. After the

day of installation, all participants used NHistory for seven

full days to cover every day of the week.

Notiﬁcations and Apps

During the seven days of the study, the

participants received

between

489

and

67, 332

notiﬁcations. The participants had

between

and

apps notifying them. In total, we logged

102, 386

notiﬁcations from

147

apps. The apps that created

the most notiﬁcations were GPS Status & Toolbox (

62, 597

notiﬁcations;

participant), WhatsApp (

12, 586

notiﬁcations;

participants), and the Android OS (

6, 079

notiﬁcations;

participants). Notiﬁcation creation peaked between 6pm and

8pm. At 2am the least number of notiﬁcations were created.

Participants received twice the amount of notiﬁcations on the

weekend compared to the rest of the week.

Notiﬁcation History Blacklist

Eight of the

participants made use of the option to exclude

speciﬁc apps from the history. These eight participants ex-

cluded between

and

apps, totaling in

different apps.

The apps excluded by most participants were Google Maps

(3 participants) and the Android OS (2 participants).

Starred Notiﬁcations

Participants rarely used the option to star notiﬁcations. Only

three of

participants used the feature at least once, resulting

in ﬁve star events. This again shows that participants were not

interested in permanently bookmarking notiﬁcations.

Snoozed Notiﬁcations

We recorded

116

snooze events from

participants and

apps. One participant did not snooze notiﬁcations. We as-

signed the duration and point-in-time (pit) methods evenly to

the participants. Seven participants stayed with their assigned

method (

duration,

pit),

switched to the other method (

duration to pit,

pit to duration), and

switched but returned

to their assigned method (2 duration, 1 pit).

In total,

109

individual notiﬁcations were snoozed.

105

were

snoozed once and

notiﬁcations were snoozed multiple times.

We will now report the

109

initial snooze events. The

109

initial snooze events by

participants break down to

snooze events per participant (Md=

). The apps that were

snoozed the most were WhatsApp (

events;

participants),

Blue Mail (

events;

participant), Facebook Messenger (

events;

participants), and Facebook (

events;

participants).

We again categorized all apps and conducted a Friedman test

to compare the normalized data of snoozed notiﬁcations from

different application categories. The results show that the

application category has an inﬂuence on the users’ snooz-

ing behavior (

(11) = 50.06, p < .001

). We conducted a

post-hoc analysis with Holm-Bonferroni corrected Wilcoxon

signed-rank tests. Our post-hoc analysis found no signiﬁcant

differences between the categories. As shown on the right

side of Table 1, most snoozed notiﬁcations were from the

categories SMS/IM (49.15%) and Email (13.66%).

We again saw a similar usage of the duration and point-in-time

methods, and the provided options. However, in contrast to

the in-the-wild study, the snooze events were more evenly

distributed over the day. Peaks for snooze events can be seen

at 10am, 1pm, and 5pm. Most notiﬁcations were re-triggered

at 12pm, 3pm, 6pm, and 9pm. Compared to the in-the-wild

study, 9am was an unpopular option to re-trigger notiﬁcations.

We again found that most notiﬁcations were snoozed and re-

triggered on the same day (

79.82%

), some on the following

day (15.60%), and only few on more than two days (4.59%).

Interviews

After the participants used NHistory for a week, we invited

them back to our lab and conducted semi-structured interviews.

We used open coding for the analysis of the interviews. Three

researchers coded the answers individually. Disagreements

were discussed until an agreement was reached.

Handling Notiﬁcations and Interruptions

Participants estimated that they receive

150

notiﬁcations

from

apps per day. One participant stated that he per-

ceives the number of notiﬁcations he has to deal with as

“low”, eight participants perceived them as “between okay

and high” and seven participants as “too high”. Six partici-

pants felt “never or rarely” being interrupted by notiﬁcations,

seven “sometimes”, and three “often”. Participants stated that

they usually react immediately when they notice a notiﬁcation.

They stated that they attend communication-related notiﬁca-

tions as soon as possible and other types of notiﬁcations if they

have time. Silencing the phone was mentioned as a method to

cope with the number of notiﬁcations participants receive on a

daily basis. One participant stated that he sometimes places

the phone out of reach in addition to silencing it. To cope

with annoying notiﬁcations, participants mentioned simply

ignoring them, dismissing them immediately, and revoking

the permission to show notiﬁcations. One participant reported

uninstalling apps because of annoying notiﬁcations.

Notiﬁcation History and Blacklist

Five participants stated that they did not use the notiﬁcation

history feature because they saw no need for it. Six other

participants stated that they used the history to read notiﬁca-

tions. P4 explained that the history enables him to read longer

messages than the notiﬁcation drawer. He further remarked

that another beneﬁt of reading notiﬁcations in the history is

that the corresponding messages are not marked as “read”.

Thus, chat partners do not expect him to reply immediately.

Participants used the history to snooze notiﬁcations and to

reﬂect on notiﬁcations they received during the day. P10 used

the history to remember important notiﬁcations.

Nine participants stated that they saw no need to blacklist apps

in the notiﬁcation history. Three other participants said that

they blacklisted apps which create many unimportant notiﬁca-

tions. Examples of this kind of notiﬁcations were notiﬁcations

from music apps (P16), GPS tracking apps (P4), system no-

tiﬁcations about available Wi-Fi networks (P12), and timers

(P16). These examples match the ﬁndings of the quantitative

results. This indicates that these kinds of notiﬁcations are only

of relevance for a limited time and participants see no point in

revisiting them. Additionally, P3 excluded an app because the

app generates notiﬁcations with sensitive data.

Starring Notiﬁcations

Thirteen participants found starring notiﬁcations unnecessary.

P10 explained that notiﬁcations are temporary and should not

be saved persistently; instead, he snoozes them if necessary.

In contrast, P4 stated that he starred a notiﬁcation which he

described as “cool and memorable”. P3 stated that he starred

notiﬁcations to remember important information and tasks

because it allows quick access to the notiﬁcation.

Snoozing Notiﬁcations

Participants explained that they snoozed notiﬁcations at work,

university, or while studying. Further contexts for snoozing no-

tiﬁcations were being on-the-go, during sport, gaming, driving,

and because they were tired.

The participants provided examples when snoozing notiﬁca-

tions was beneﬁcial. For instance, P2 explained that she likes

snoozing notiﬁcations because she does not have to deal with

the notiﬁcation itself anymore if the system can remind her.

Furthermore, participants found that snoozing notiﬁcations

supports their attention management. P16 explained that he is

less distracted if he snoozes interesting notiﬁcations to a more

appropriate time. Additionally, P4 explained that snoozing

notiﬁcations helps to keep the status bar clean.

Participants also told us what they disliked. Snoozing notiﬁ-

cations manually was sometimes regarded as unnecessary, as

it takes the same amount of time as, for example, answering

a short message. The “snooze notiﬁcation” that appeared af-

ter dismissing a notiﬁcation was described as annoying. P3

explained that most of the time she does not want to snooze

a notiﬁcation when dismissing it. Another participant com-

plained about the short period the “snooze notiﬁcation” was

shown (

5sec

) because he was sometimes too slow to snooze

the notiﬁcation directly in the notiﬁcation center and had to

open NHistory to snooze the notiﬁcation from the history in-

stead (P4). P2 reported that she was sometimes not able to ﬁnd

a notiﬁcation in the history to snooze it. Participants disliked

that they have to unlock their phones before they can snooze

notiﬁcations from the lock screen. P11 raised concerns regard-

ing snoozing notiﬁcations because he feels a social pressure

to answer messages quickly and when a message is snoozed

the sender has to wait for a reply. Another participant was

concerned about being overwhelmed when snoozing too many

notiﬁcations during the day to the same time slot. She sug-

gested creating an overview of all received notiﬁcations in the

evening instead.

Reasons for Snoozing Notiﬁcations

Participants mainly snoozed notiﬁcations to create reminders.

For example, participants mentioned that they snoozed

SMS/IM and Email notiﬁcations because they were at work

and did not want to deal with personal notiﬁcations. P12 ex-

plained that she concentrated on work and wanted to receive

a reminder afterwards. This was a common theme in the in-

terviews. Participants snoozed SMS/IM notiﬁcations because

they wanted to avoid switching their current context. Snooz-

ing SMS/IM and Email notiﬁcations was often mentioned in

regard of not forgetting a task. Another reason for snooz-

ing notiﬁcations was that participants were sometimes not in

the mood to deal with the notiﬁcations when they received

them. For instance, P11 explained that he received “20-30”

instant messages from a group chat and wanted to read them

later. Participants also had to do other things ﬁrst before being

able to react to notiﬁcations. P13 explained that he received

a message about a meeting with friends and had to ask his

wife before he could accept the meeting. Further, participants

snoozed notiﬁcations on-the-go to deal with them at home. For

instance, P11 mentioned that he received a Social notiﬁcation,

which he wanted to read at home on his desktop computer.

P13 explained that he snoozed an app update notiﬁcation be-

cause he wanted to install the updates at home using a Wi-Fi

connection. P12 snoozed a Game notiﬁcation not to miss an

in-game reward.

Duration vs. Point-in-Time

Six participants liked the duration method to snooze notiﬁca-

tions, as it enables them to estimate when they will be able

to deal with the notiﬁcations. P11 stated that she likes the

method because it allows her to decide how long to snooze

based on the current situation. Participants found the duration

method better for short periods of time. P3 especially liked

it for snoozing notiﬁcations to the same day. However, other

participants stated that the duration method requires more cog-

nitive effort. P2 explained that she thinks in times of the day

and therefore she would have to calculate the time distance

herself. Further, P7 mentioned he does not know if he has time

to deal with the notiﬁcation in a particular distance in time.

Eight participants stated that the point-in-time method sup-

ports their daily routines. For example, P7 explained that he

usually knows when he can deal with notiﬁcations, e.g., in the

lunch break or after work. Additionally, another participant

explained that he does not have to think about how long it

will take until he can deal with the notiﬁcation, e.g., in the

evening. Participants mentioned that they prefer the point-

in-time method for longer distances in time, e.g., more than

an hour (P10) or a week (P3). P9 stated that the pre-deﬁned

time slots are not always useful. Further, two participants

found that entering a custom time is difﬁcult when being busy.

P11 mentioned that the point-in-time method requires a higher

cognitive effort because he has to calculate the time when he

wants to receive the notiﬁcation.

Usefulness of NHistory

Eleven participants stated that the app helps them to deal

with notiﬁcations, four found it somewhat useful, and one

participant found it to be not useful at all. Participants liked

that the app enables them to have a “clean” notiﬁcation drawer,

and they found the notiﬁcation history helpful. P4 especially

liked that he can dismiss notiﬁcations without losing them.

P3 liked that she can reﬂect on received notiﬁcations and was

interested in how many notiﬁcations she receives.

Concerns and Suggestions

Participants disliked the “snooze notiﬁcation” that was shown

when dismissing a notiﬁcation. They suggested that, instead,

the functionality should be added to the notiﬁcation itself us-

ing a gesture or long-press action. Two participants would

like to have the duration and point-in-time options available at

the same time, and two other participants would like to cus-

tomize the options. Participants disliked how multiple instant

messages are grouped in notiﬁcations and snoozing a single

conversation is sometimes not possible. P1 and P5 suggested

that after snoozing a notiﬁcation, all following notiﬁcations

from the same app should be snoozed as well. Further, par-

ticipants would like to snooze all notiﬁcations from all apps

for a speciﬁc duration or a speciﬁc point-in-time. P15 sug-

gested automatic rules to snooze notiﬁcations. Participants

also suggested using location-based triggers instead of time-

based ones. P10 wished that snoozing a notiﬁcation would

create a corresponding calendar event which then could be

synchronized with other devices. Regarding the notiﬁcation

history, P4 suggested that “spammy” notiﬁcations should be

automatically detected and excluded. P8 would like to see

statistics about received notiﬁcations and P14 suggested an

end-of-day summary of all received notiﬁcations.

DISCUSSION AND LIMITATIONS

We observed comparable usage patterns in both studies. Tem-

porarily snoozing notiﬁcations was favored compared to per-

manently bookmarking them. Most snoozed notiﬁcations were

of the SMS/IM category, followed by Calendar/Reminder (in-

the-wild) and Email (controlled).

In both studies, we observed peaks in re-triggered notiﬁcations

at 12pm, 3pm, 6pm, and 9pm. Additionally, in the in-the-wild

study 9am was preferred. These times are likely inﬂuenced by

the pre-deﬁned options of NHistory but might also correlate

with before and after work hours, and lunch breaks. In the

controlled study, almost a quarter of the snoozed notiﬁcations

were re-triggered at 6pm, likely correlating with after work

hours. This was further strengthened in the interviews with

many participants stating that they snoozed personal notiﬁca-

tions at work or while studying. We observed that during the

day notiﬁcations were typically deferred for short amounts

of time. However, deferring notiﬁcations from the morning

or noon to the evening, or from the evening until the next

morning were also common use-cases. Another similarity of

both studies was that the deferral of notiﬁcations was mostly

limited to the same day. Few notiﬁcations were snoozed until

the next day and more than two days was an exception. Still,

in both studies, we observed only a small fraction of received

notiﬁcations being snoozed. We assume these notiﬁcations

to be of high relevance to the users but, at the same time,

out-of-context and with a low urgency.

A limitation of the in-the-wild study is that we did not collect

demographic data from the users. While the Google Play

Store statistics indicate a diverse set of users, we have little

background information. We assume that the active users

were, to a certain degree, tech-savvy, as they found, installed,

and conﬁgured NHistory without detailed instructions. In the

controlled study, most participants had a technical background

and were, therefore, tech-savvy as well. A second limitation

is that we limited the data collection due to privacy concerns.

This resulted in little knowledge about the notiﬁcations apart

from which app issued them at what time. Future studies

should consider context data and the notiﬁcations’ content.

While we conducted the studies, the latest version of Android

was announced (March 21, 2017) and eventually released (Au-

gust 21, 2017). In Android 8.0 (“Oreo”) snoozing notiﬁcations

was implemented natively. Swiping a notiﬁcation from left

to right unveils a “clock” icon. Tapping on the icon allows

the user to snooze the notiﬁcation for 15 minutes, 30 minutes,

1 hour, or 2 hours. This interaction is similar to the sugges-

tions we received in the interviews. Comparing the provided

options with NHistory and our ﬁndings, we notice the lack of

long-term options for use-cases such as snoozing notiﬁcations

until after work or the next morning. Since we concluded

the controlled study on March 22, 2017, it is unlikely that

the announcement of Android 8.0 inﬂuenced the participants.

However, we assume that some users of the in-the-wild study

read about the feature and subsequently searched and down-

loaded NHistory from the Google Play Store.

DESIGN IMPLICATIONS

From the ﬁndings of the in-the-wild and controlled studies, we

derived design implications for future notiﬁcation systems.

Consider Context and Daily Routines

Participants in the interviews reported multiple reasons for de-

ferring notiﬁcations. For instance, when the user is focused on

another task (work, studying), the notiﬁcation is out-of-context

(personal notiﬁcations at work), when the user is unable to

attend the notiﬁcation (on the go, driving, sports), or simply

when the user is not in the mood to take action on the notiﬁ-

cation. Users already apply various strategies in these cases,

such as muting the phone, putting it away and revoking notiﬁ-

cation permissions. Future notiﬁcation systems should follow

and build on these strategies.

We observed peaks in deferred notiﬁcations to, what we as-

sume to be, before and after work hours, and lunch breaks.

Especially in the controlled study, many of the deferred notiﬁ-

cations were re-triggered in the evening. Future notiﬁcation

systems should consider the user’s context and daily routines.

For example, personal notiﬁcations at work could be auto-

matically be detected and subsequently deferred to after-work

hours. This could be combined with the creation of auto-

matic summaries. Although we offered multiple pre-deﬁned

options to snooze notiﬁcations, users made effective use of

custom durations and points-in-time. Further, participants of

the controlled study suggested personalization of these options,

indicating that a one-ﬁts-all approach might not be practical.

Balance Importance and Social Expectations

Notiﬁcations from the SMS/IM category were snoozed most of-

ten in both studies, highlighting their importance to users. Fur-

ther important categories include Calendar/Reminder, Email,

and Social; reiterating that “notiﬁcations are for messaging”

and “important notiﬁcations are about people and events” [34].

Participants stated in the interviews that they often attend

communication-related notiﬁcations immediately. Especially

for instant messages, there are social expectations to reply

quickly [4]. Future notiﬁcation systems should carefully as-

sess the importance of a notiﬁcation to decide whether or not

it should be deferred and for how long.

Mind the Ephemeral Nature of Notiﬁcations

Current mobile operating systems do not display how many

notiﬁcations users receive on a daily basis. We received posi-

tive feedback regarding the notiﬁcation history, as it enables

users to reﬂect on notiﬁcations. The history further allows

users to “safely” dismiss notiﬁcations because they can always

look them up afterward, reducing the number of pending noti-

ﬁcations in the notiﬁcation drawer. Still, participants regarded

notiﬁcations as temporary. Notiﬁcations were mostly deferred

to the same day or next morning. Even for notiﬁcations that

were snoozed multiple times, we observed that deferring a noti-

ﬁcation for more than two days was an exception. Developers

of future notiﬁcations systems that defer notiﬁcations can use

this as an upper bound. Within this time span, we observed a

high variance regarding the deferral duration, depending on

the time of the day and the users’ daily routines.

CONCLUSION

In this paper, we investigated the user-deﬁned deferral of mo-

bile notiﬁcations. To reduce negative effects caused by inter-

ruptions from notiﬁcations, a body of related work investigated

using opportune moments for notiﬁcation delivery, often based

on breakpoints in the user’s activity. Our approach instead

allows users to manually “snooze” notiﬁcations. We focused

on deriving insights into how users postpone notiﬁcations

rather than eliminating negative effects directly, as snoozing a

notiﬁcation might even introduce a second interruption. We

developed the Android app NHistory that extends the Android

operating system in two ways. A notiﬁcation history enables

users to go back to previously dismissed notiﬁcations. Further,

the app allows users to snooze notiﬁcations for a user-deﬁned

duration or to a user-deﬁned point-in-time.

We explored how and why users make use of the provided func-

tionality in a year-long in-the-wild study with 295 active users

and a week-long controlled study with 16 participants. Notiﬁ-

cations of the categories SMS/IM, Calendar/Reminder, Social,

and Email, were snoozed most often. Even for notiﬁcations

that were snoozed multiple times, we observed that deferring

a notiﬁcation for more than two days was an exception. We

conducted interviews to gain insights into when and why peo-

ple ﬁnd deferring notiﬁcations useful. Participants mentioned

avoiding context switches, especially from attending personal

notiﬁcations during work. As a result, we observed a number

of notiﬁcations being snoozed to before and after work hours,

as well as lunch breaks. Participants raised concerns regarding

deferring communication-related notiﬁcations, due to social

expectations to respond as soon as possible. We conclude the

paper with design implications for future smart notiﬁcation

systems. These systems should consider different categories

of notiﬁcations, such as personal and work-related notiﬁca-

tions, the current context of the users, as well as their daily

routines. In future work, we plan to investigate snoozing notiﬁ-

cations with user-deﬁned rules [2], the generation of automatic

notiﬁcation summaries, and location-based triggers.

Acknowledgments:

This work is supported by the Ger-

man Ministry of Education and Research (BMBF) within

the DAAN project (13N13481) and by the DFG within the

SimTech Cluster of Excellence (EXC 310/2).

REFERENCES

1. Piotr D Adamczyk and Brian P Bailey. 2004. If not now,

when?: the effects of interruption at different moments

within task execution. In Proceedings of the SIGCHI

conference on Human factors in computing systems.

ACM, 271–278.

Jonas Auda, Dominik Weber, Alexandra Voit, and Stefan

Schneegass. 2018. Understanding User Preferences

Towards Rule-based Notiﬁcation Deferral. In Extended

Abstracts of the 2018 CHI Conference on Human Factors

in Computing Systems (CHI EA ’18). ACM, New York,

NY, USA, Article LBW584, 6 pages. DOI:

http://dx.doi.org/10.1145/3170427.3188688

3. Brian P Bailey, Joseph A Konstan, and John V Carlis.

2000. Measuring the effects of interruptions on task

performance in the user interface. In Systems, Man, and

Cybernetics, 2000 IEEE International Conference on,

Vol. 2. IEEE, 757–762.

4. Karen Church and Rodrigo de Oliveira. 2013. What’s up

with whatsapp?: comparing mobile instant messaging

behaviors with traditional SMS. 15th international

conference on Human-computer interaction with mobile

devices and services (MobileHCI’13) (2013), 352–361.

DOI:http://dx.doi.org/10.1145/2493190.2493225

5. Herbert H Clark. 1996. Using language. Cambridge

university press.

6. Fulvio Corno, Luigi De Russis, and Teodoro Montanaro.

2016. A context and user aware smart notiﬁcation system.

IEEE World Forum on Internet of Things, WF-IoT 2015 -

Proceedings (2016), 645–651. DOI:

http://dx.doi.org/10.1109/WF-IoT.2015.7389130

7. Mary Czerwinski, Edward Cutrell, and Eric Horvitz.

2000. Instant Messaging and Interruption: Inﬂuence of

Task Type on Performance. Proceedings of OZCHI 2000

(2000), 356–361. DOI:

http://dx.doi.org/10.1016/S1361-3723(02)01112-0

8. Android Developers. 2018. NotiﬁcationListenerService.

https://developer.android.com/reference/android/

service/notification/NotificationListenerService.html.

(2018). [Online; accessed 20-January-2018].

9. Tilman Dingler, Dominik Weber, Martin Pielot, Jennifer

Cooper, Chung-Cheng Chang, and Niels Henze. 2017.

Language Learning On-the-go: Opportune Moments and

Design of Mobile Microlearning Sessions. In

Proceedings of the 19th International Conference on

Human-Computer Interaction with Mobile Devices and

Services (MobileHCI ’17). ACM, New York, NY, USA,

Article 28, 12 pages. DOI:

http://dx.doi.org/10.1145/3098279.3098565

10. Joel E. Fischer, Chris Greenhalgh, and Steve Benford.

2011. Investigating episodes of mobile phone activity as

indicators of opportune moments to deliver notiﬁcations.

Proceedings of the 13th International Conference on

Human Computer Interaction with Mobile Devices and

Services - MobileHCI ’11 (2011), 181. DOI:

http://dx.doi.org/10.1145/2037373.2037402

11. Sandy J J Gould, Duncan P Brumby, and Anna L Cox.

2013. What does it mean for an interruption to be

relevant? An investigation of relevance as a memory

effect. Proceedings of the Human Factors and

Ergonomics Society Annual Meeting 57, 1 (2013),

149–153. DOI:

http://dx.doi.org/10.1177/1541931213571034

12. Joyce Ho and Stephen S Intille. 2005. Using

context-aware computing to reduce the perceived burden

of interruptions from mobile devices. Proceedings of the

SIGCHI conference on Human factors in computing

systems - CHI ’05 (2005), 909. DOI:

http://dx.doi.org/10.1145/1054972.1055100

13. Eric Horvitz, Paul Koch, and Johnson Apacible. 2004.

BusyBody: Creating and Fielding Personalized Models

of the Cost of Interruption. In Proceedings of the 2004

ACM Conference on Computer Supported Cooperative

Work (CSCW ’04). ACM, New York, NY, USA, 507–510.

DOI:http://dx.doi.org/10.1145/1031607.1031690

14. Shamsi T. Iqbal and Brian P. Bailey. 2010. Oasis. ACM

Transactions on Computer-Human Interaction 17, 4

(2010), 1–28. DOI:

http://dx.doi.org/10.1145/1879831.1879833

15.

N. Kern and B. Schiele. 2003. Context-aware notiﬁcation

for wearable computing. Seventh IEEE International

Symposium on Wearable Computers, 2003. Proceedings.

(2003), 223–230. DOI:

http://dx.doi.org/10.1109/ISWC.2003.1241415

16. Afra Mashhadi, Akhil Mathur, and Fahim Kawsar. 2014.

The Myth of Subtle Notiﬁcations. In Proceedings of the

2014 ACM International Joint Conference on Pervasive

and Ubiquitous Computing: Adjunct Publication

(UbiComp ’14 Adjunct). ACM, New York, NY, USA,

111–114. DOI:

http://dx.doi.org/10.1145/2638728.2638759

17.

Abhinav Mehrotra, Robert Hendley, and Mirco Musolesi.

2016. PrefMiner. Proceedings of the 2016 ACM

International Joint Conference on Pervasive and

Ubiquitous Computing - UbiComp ’16 October (2016),

1223–1234. DOI:

http://dx.doi.org/10.1145/2971648.2971747

18.

Abhinav Mehrotra, Mirco Musolesi, Robert Hendley, and

Veljko Pejovic. 2015. Designing content-driven

intelligent notiﬁcation mechanisms for mobile

applications. Proceedings of the 2015 ACM International

Joint Conference on Pervasive and Ubiquitous

Computing (UbiComp ’15) (2015), 813–824. DOI:

http://dx.doi.org/10.1145/2750858.2807544

19.

Abhinav Mehrotra, Veljko Pejovic, Jo Vermeulen, Robert

Hendley, and Mirco Musolesi. 2016. My Phone and Me:

Understanding People’s Receptivity to Mobile

Notiﬁcations. In Proceedings of the 2016 CHI

Conference on Human Factors in Computing Systems

(CHI ’16). ACM, New York, NY, USA, 1021–1032.

DOI:

http://dx.doi.org/10.1145/2858036.2858566

20. Alistair Morrison, Donald McMillan, Stuart Reeves,

Scott Sherwood, and Matthew Chalmers. 2012. A Hybrid

Mass Participation Approach to Mobile Software Trials.

In Proceedings of the SIGCHI Conference on Human

Factors in Computing Systems (CHI ’12). ACM, New

York, NY, USA, 1311–1320. DOI:

http://dx.doi.org/10.1145/2207676.2208588

21. Darren Newtson and Gretchen Engquist. 1976. The

perceptual organization of ongoing behavior. Journal of

Experimental Social Psychology 12, 5 (1976), 436–450.

DOI:http://dx.doi.org/10.1016/0022-1031(76)90076-7

22. Tadashi Okoshi, Julian Ramos, Hiroki Nozaki, Jin

Nakazawa, Anind K Dey, and Hideyuki Tokuda. 2015a.

Attelia : Reducing User ’ s Cognitive Load due to

Interruptive Notiﬁcations on Smart Phones. (2015),

96–104.

23. Tadashi Okoshi, Julian Ramos, Hiroki Nozaki, Jin

Nakazawa, Anind K. Dey, and Hideyuki Tokuda. 2015b.

Reducing Users’ Perceived Mental Effort due to

Interruptive Notiﬁcations in Multi-Device Mobile

Environments. Proceedings of the Joint International

Conference on Pervasive and Ubiquitous Computing and

the International Symposium on Wearable Computers

(Ubicomp/ISWC’15) (2015), 475–486. DOI:

http://dx.doi.org/10.1145/2750858.2807517

24. Tadashi Okoshi, Hideyuki Tokuda, and Jin Nakazawa.

2014. Attelia: Sensing User’s Attention Status on Smart

Phones. Proceedings of the 2014 ACM International Joint

Conference on Pervasive and Ubiquitous Computing

Adjunct Publication - UbiComp ’14 Adjunct (2014),

139–142. DOI:

http://dx.doi.org/10.1145/2638728.2638802

25. Tadashi Okoshi, Kota Tsubouchi, Masaya Taji, Takanori

Ichikawa, and Hideyuki Tokuda. 2017. Attention and

Engagement-Awareness in the Wild : A Large-Scale

Study with Adaptive Notiﬁcations. Proceedings of IEEE

International Conference on Pervasive Computing and

Communications 2017 (PerCom ’17) (2017).

26. Chunjong Park, Junsung Lim, Juho Kim, Sung-Ju Lee,

and Dongman Lee. 2017. Don’t Bother Me. I’m

Socializing!: A Breakpoint-Based Smartphone

Notiﬁcation System. In Proceedings of the 2017 ACM

Conference on Computer Supported Cooperative Work

and Social Computing (CSCW ’17). ACM, New York,

NY, USA, 541–554. DOI:

http://dx.doi.org/10.1145/2998181.2998189

27. Veljko Pejovic and United Kingdom. 2014. InterruptMe:

Designing Intelligent Prompting Mechanisms for

Pervasive Applications. (2014), 897–908.

28. Veljko Pejovic, Mirco Musolesi, and Abhinav Mehrotra.

2015. Investigating The Role of Task Engagement in

Mobile Interruptibility. In Proceedings of the 17th

International Conference on Human-Computer

Interaction with Mobile Devices and Services Adjunct

(MobileHCI ’15). ACM, New York, NY, USA,

1100–1105. DOI:

http://dx.doi.org/10.1145/2786567.2794336

29. Martin Pielot, Bruno Cardoso, Kleomenis Katevas, Joan

Serrà, Aleksandar Matic, and Nuria Oliver. 2017. Beyond

Interruptibility: Predicting Opportune Moments to

Engage Mobile Phone Users. Proc. ACM Interact. Mob.

Wearable Ubiquitous Technol. 1, 3, Article 91 (Sept.

2017), 25 pages.

DOI:http://dx.doi.org/10.1145/3130956

30. Martin Pielot, Karen Church, and Rodrigo de Oliveira.

2014. An In-Situ Study of Mobile Phone Notiﬁcations.

Proc. MobileHCI ’14 (2014), 233–242. DOI:

http://dx.doi.org/10.1145/2628363.2628364

31.

Martin Pielot, Tilman Dingler, Jose San Pedro, and Nuria

Oliver. 2015. When Attention is Not Scarce - Detecting

Boredom from Mobile Phone Usage. In Proceedings of

the 2015 ACM International Joint Conference on

Pervasive and Ubiquitous Computing (UbiComp ’15).

ACM, New York, NY, USA, 825–836. DOI:

http://dx.doi.org/10.1145/2750858.2804252

32. Martin Pielot and Luz Rello. 2015. The Do Not Disturb

Challenge: A Day Without Notiﬁcations. Extended

Abstracts of the ACM CHI’15 Conference on Human

Factors in Computing Systems 2 (2015), 1761–1766.

DOI:

http://dx.doi.org/10.1145/2702613.2732704

33. Martin Pielot and Luz Rello. 2017. Productive, Anxious,

Lonely: 24 Hours Without Push Notiﬁcations. In

Proceedings of the 19th International Conference on

Human-Computer Interaction with Mobile Devices and

Services (MobileHCI ’17). ACM, New York, NY, USA,

Article 11, 11 pages. DOI:

http://dx.doi.org/10.1145/3098279.3098526

34. Alireza Sahami Shirazi, Niels Henze, Tilman Dingler,

Martin Pielot, Dominik Weber, and Albrecht Schmidt.

2014. Large-scale assessment of mobile notiﬁcations.

Proceedings of the 32nd annual ACM conference on

Human factors in computing systems - CHI ’14 (2014),

3055–3064. DOI:

http://dx.doi.org/10.1145/2556288.2557189

35. Angela Sasse, Chris Johnson, and others. 1999.

Coordinating the interruption of people in

human-computer interaction. In Human-computer

interaction, INTERACT, Vol. 99. 295.

36. Daniel Siewiorek, Asim Smailagic, Junichi Furukawa,

Neema Moraveji, Kathryn Reiger, and Jeremy Shaffer.

2003. SenSay: A context-aware mobile phone. Seventh

IEEE International Symposium on Wearable Computers,

2003. Proceedings. (2003), 248–249. DOI:

http://dx.doi.org/10.1109/ISWC.2003.1241422

37. Katarzyna Stawarz, Anna L. Cox, and Ann Blandford.

2014. Don’t Forget Your Pill!: Designing Effective

Medication Reminder Apps That Support Users’ Daily

Routines. In Proceedings of the SIGCHI Conference on

Human Factors in Computing Systems (CHI ’14). ACM,

New York, NY, USA, 2269–2278. DOI:

http://dx.doi.org/10.1145/2556288.2557079

38. Dominik Weber, Alireza Sahami Shirazi, and Niels

Henze. 2015. Towards Smart Notiﬁcations Using

Research in the Large. In Proceedings of the 17th

International Conference on Human-Computer

Interaction with Mobile Devices and Services Adjunct

(MobileHCI ’15). ACM, New York, NY, USA,

1117–1122. DOI:

http://dx.doi.org/10.1145/2786567.2794334

39. Dominik Weber, Alexandra Voit, Philipp Kratzer, and

Niels Henze. 2016a. In-Situ Investigation of Notiﬁcations

in Multi-device Environments. In Proceedings of the

2016 ACM International Joint Conference on Pervasive

and Ubiquitous Computing (UbiComp ’16). ACM, New

York, NY, USA, 1259–1264. DOI:

http://dx.doi.org/10.1145/2971648.2971732

40. Dominik Weber, Alexandra Voit, Huy Viet Le, and Niels

Henze. 2016b. Notiﬁcation Dashboard: Enabling

Reﬂection on Mobile Notiﬁcations. In Proceedings of the

18th International Conference on Human-Computer

Interaction with Mobile Devices and Services Adjunct

(MobileHCI ’16). ACM, New York, NY, USA, 936–941.

DOI:http://dx.doi.org/10.1145/2957265.2962660

41. Fengpeng Yuan, Xianyi Gao, and Janne Lindqvist. 2017.

How Busy Are You?: Predicting the Interruptibility

Intensity of Mobile Users. In Proceedings of the 2017

CHI Conference on Human Factors in Computing

Systems (CHI ’17). ACM, New York, NY, USA,

5346–5360. DOI:

http://dx.doi.org/10.1145/3025453.3025946