Extracting and Understanding Call-to-actions
of Push-notifications

Beatriz Esteves, Kieran Fraser, Shridhar Kulkarni, Owen Conlan, VĂ­ctor RodrĂ­guez-Doncel

Paper Overview

Push-notifications are a communication tool leveraged by many apps to disseminate information, engage with their user base and provide a means of encouraging users to take particular actions. The nuanced intent behind a push is not always distinguishable to the end-user at moments of delivery. This work explores the text content of notifications pushed by a number of prominent apps in the marketplace over the period of 463 days. We present a new ontology that defines notification Call-to-action (CTA) labels in use today. This facilitates greater understanding behind a push and is a step towards standardisation for marketing teams. Subsequently, we then present results of a notification dataset annotated with our CTA labels and propose and evaluate a CTA text classification task, which could facilitate improved solutions for both users subscribed to, and marketers creating, push-notifications.

Pre-print APN documentation APN repo

Additional resources

  1. ORSD for the APN ontology
  2. Notifications assigned to CTA labels

ORSD for the APN ontology

Ontology for the Annotation of Push-Notifications
1. Purpose
The purpose of this ontology is to support the annotation and extraction of insights from push-notification data.
2. Scope
The scope of this ontology is limited to the definition of a set of parameters that can be used to classify push-notification data, in particular its text content, with the end goal of creating a service which pushes more empathethic content to push-notification subscribers. Specifically, the notification content will be used to categorize the notification in terms of: (i) its specific call-to-action, (ii) and respective marketing campaign type, (iii) its target audience, (iv) its time sensitiveness and (v) the presence of information that can be used to infer personal data from the subscribers.
3. Implementation Language
RDF
4. Intended End-Users
Organizations with Web or application services which include push-notifications in their communication flow with customers.
5. Intended Uses
Use 1. Creation of multi-label classification models to automatically annotate push-notifications.
Use 2. Generation of push-notification content based on annotations' model.
6. Ontology Requirements
a. Non-Functional Requirements
NFR 1. The ontology shall be published online with standard documentation.
b. Functional Requirements: Groups of Competency Questions
CQG1. Related to push-notification services CQG2. Related to push-notification categorization
CQ1. What is the content of the push-notification?
CQ2. Which service / application pushed the notification?
CQ3. What is the application type in relation to its content? 		CQ4. What is the call-to-action of the notification?
CQ5. Does the notification belong to a specific campaign type?
CQ6. How urgent is the notification?
CQ7. Does the notification have a clear target audience?
CQ8. Does the notification contain features which indicate personalization?

Notifications assigned to CTA labels

CTA label # Notifications
discover 2633
awareness 1394
purchase 1172
reminder/advice/tip 763
social 577
watch 356
play 310
feedback 278
claim 151
download 120
account management 99

Get in touch

Feel free to reach out to us regarding the research presented.