User Complaints of Wearable Apps Examination:A Case Study on Android Wear
Abstract—Wearable
apps are becoming increasingly popularin recent years. However, to date, very
few studies examinedthe issues that wearable apps face. Prior studies showed
thatuser reviews contain a plethora of insights that can be usedto understand
quality issues and help developers build betterquality mobile apps. Therefore,
in this paper, we mine userreviews in order to understand the user complaints
of wearableapps. We manually sample and categorize 589 reviews from 6Android
wearable apps. Our findings indicate that the mostfrequent complaints are
related to functional errors, lack offunctionality, and cost. Learn More Our results are
useful to the wearabledeveloper community since they highlight the issues that
usersface and care most about.Keywords-Wear Apps; Users’ Reviews; Google Play
Store;Empirical Studies
STUDY DESIGN
The goal of our study is to determine the most frequent
andnegatively impacting user complaints of wearable apps. To doso, we mine the
Google Play Store for the reviews of wearapps. In the following sections we
describe our data selectionand collection, as well as detail our review
classificationmethodology.A. Data Collection and SelectionFor the purpose of
our study, we select a number of wearableapps that have user reviews. First, we
obtained the availableAndroid Wear apps on Google Play Store by collecting
theiridentifiers from two alternative app markets:Android WearCenter[1] and
GoKo [4]. The two aforementioned sourceshave been used in prior work focusing
on wearable apps [6].Then, we mined the wearable apps using a data scrapperthat
we developed.
The scrapper collected various informationabout each wear apps,
including: the user review’s text, itsrating, the developer’s reply to the
review, if any, and the apps’overall rating. To enhance performance of the
scrapper, it wasdeployed on a cluster of machines in order to distribute
therequests sent.In total, we mined the data for 4,722 wear apps thatare
developed by 2,732 unique developers. The 4,722 appshad 1,284,349 user reviews,
which we mined. Since we areinterested in user complaints, and similar to the
prior study byKhalidet al.[17], we select the low-rated reviews (i.e., 1 and2
star reviews) since they are most likely to contain the usercomplaints. Since
we need a reasonable amount of reviews toperform our analysis, we only considered
apps that had morethan 100 low-rated reviews.
After performing these steps,
werandomly select 6 apps that have 791 reviews.Since this is the first study to
examine user complaints forwear apps, we opt to perform our analysis of the
user com-plaints manually. Given that this manual classification is a timeand
resource intensive task, we selected a random statisticallyrepresentative
sample of complaints from each application.The sample sizes were selected to
attain a 5% confidenceinterval and a 95% confidence level in the population
beingsampled. This random sampling process resulted in 589 totalreviews varying
from 115 to 154 reviews per app.
The listof the studied wear apps, their
overall rating on Google Playstore, the number of total low rated reviews and
the numberof examined reviews is shown in Table I.B. Manual Classification of
User ReviewsOnce we obtained all of the reviews, we categorize themin order to
come up with the different types of complaints.To do so, we used a Grounded
Theory type of technique [9],[27], where we took a random sample of reviews
from all theselected apps and did a simple manual classification of them.This
step was done mainly to come up with an initial set ofcategories that the
reviews can be grouped into. In the end ofthis step, we came up with 15
different categories, which wecall complaint types.Once we came up with the
initial 15 complaint types, weproceeded to categorize the sampled user reviews
(in total589 reviews). To facilitate the categorization of the reviews,we built
a web-based tool that enabled the categorization ofthe review - presenting the
review details and the respectivedeveloper reply, if a developer posted a reply
to the review.The tool also had the option to add a new category in case
areview belonged to a category that was not listed. Every reviewwas tagged with
all suitable categories, i.e., one review can1549797
Comments
Post a Comment