Investigating user profiles for mobile application categories and providing recommendations to software engineering

Abstract

Mobile apps are a type of software applications developed for use on mobile
devices such as smartphones and tablets. Once developed, an app is made
available for users via an application distribution platform, commonly known as an
app store. (Lim, Bentley, Kanakam, Ishikawa & Honiden, 2015). Mobile apps in an
app store are presented via mobile application categories. List of app categories may
change from app store to app store. A mobile app can be in more than one category.
For instance: A photo editing and sharing mobile app can be in both social
networking and photo & video app category. Generally a mobile application’s
success is measured by its number of downloads. App stores and mobile app
developers want more users to download their apps. In order to do this, they prepare
a marketing strategy and promote their apps through different platforms. (E.g: Social
Media, Search Ads, TV Advertising, etc). To make this app promotion strategy more
effective they need to identify best user profiles for their app category. In other
words, they need to identify user profiles which have a high percentage of
downloading apps from that app category.
My thesis work is about using data mining to identify these user profiles for
each mobile app category. In order to apply data mining techniques I needed to find
an open source dataset about user profiles and their preferences about mobile apps.
Luckily, I found a research about this topic including a survey and an open source
dataset. Dataset is filled with users’ answers to survey questions. This academic
work is performed from researchers of University of College London and National
Institute of Informatics, Japan. These researchers published an academic paper
which is about mobile app user behaviour based on country differences. (Lim,
Bentley, Kanakam, Ishikawa & Honiden, 2015).
Throughout data mining and master thesis process I followed CRISP-DM
Methodology. CRISP-DM methodology is a structured way of planning a data mining
project It consists six different stages: Business understanding, Data understanding,
Data preparation, Modeling, Evaluation and Deployment. (Vorhies, 2016) I followed
methodology in its exact order. However, I wrote only one deployment section which
covers all app categories.