google play store case study

Navigation Menu

Search code, repositories, users, issues, pull requests..., provide feedback.

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly.

To see all available qualifiers, see our documentation .

• Notifications

Data analysis of apps on the play store.

therealyash/Google-Play-Store-Case-Study

Folders and files, repository files navigation, google-play-store-case-study.

This repo contains a case study of the applications on the Google Play Store, where we handle data, compare, the attributes like size, price, ratings of the app and visualize to gain valuable insights.

The tasks performed on the data set are as follows:

• Data Cleaning
• Data Manipulation
• Data Analysis
• Data Communication with Information Visualization

The analysis focuses on the breadth and present the data with aesthetics, instead of focusing on a single attribute in depth. This will give us an overall understanding of the applications listend on Play Store.

• Jupyter Notebook 100.0%

Numbers, Facts and Trends Shaping Your World

Read our research on:

Full Topic List

Regions & Countries

• Publications
• Our Methods
• Short Reads
• Tools & Resources

Read Our Research On:

• Apps Permissions in the Google Play Store
• Chapter 2: An Analysis of Apps in the Google Play Store

Table of Contents

• Chapter 1: The Majority of Smartphone Owners Download Apps
• Chapter 3: An Analysis of Android App Permissions
• About This Report

Pew Research Center surveys and independent analysts have found that roughly one-third of all American adults – and about half of smartphone owners – have an Android smartphone. All apps for Android phones are housed in the Google Play Store, and each app has its own web page where information about the app is available. Each web page includes basic descriptive information such as what type of app it is; how much, if anything, the app costs to download; the app’s content rating; and what information or other permissions the app requires from users.

To collect information on these apps, Pew Research Center scraped each app’s page in the Google Play Store using the Chrome browser and a custom browser extension. The extension copied all of the information on each page and stored it in a database for later analysis. The data collection was conducted from June to September 2014, and captured information of about 1,041,336 unique apps (see the Methods section for a detailed description of this process).

Pew Research Center elected to study the Android ecosystem for two specific reasons. First, information about apps in the Google Play Store is freely available on the Google Play Store website and therefore easily aggregated. Second, the Google Play Store publishes information about app permissions along with the other metadata, and the issue of permissions is a key focus of this research.

Other platforms, such as iOS (Apple’s mobile operating system), are more challenging for outside parties to access, and they handle the process of informing users about app behavior differently and in ways that are considerably more complicated to analyze. For these reasons, only apps within the Android ecosystem in mid-2014 are included in this analysis. It should be treated as a study of one app ecosystem and how it handles informing users about how their information is being collected and used.

Moreover, the situation in the Google Play Store has changed somewhat from the time these data were collected and the permissions regime that exists for those who now want to install apps on their Android phone. As noted above, the Android operating system was changed this summer when Google announced a new feature in the next version of the Android operating system. This new feature would allow users to turn off certain permissions on an app-by-app basis and to see all of the apps permissions in a single place (sometimes referred to as a “permissions dashboard”). While this change will be significant when it is fully available, it is still built upon the existing structure of permissions.

At the time of the data collection, the Google Play Store broke apps down into 41 general categories. Education apps were the most common individual category, comprising 8% of the total number of apps available for download. 7

Overall, eight categories of apps (Education, Entertainment, Personalization, Tools, Lifestyle, Books and Reference, Business, and Travel & Local) comprised more than half of the apps available for download (53.58% in total).

Music apps were the least prevalent category, comprising just 668 apps — or 0.06% of the more than 1 million total apps in the Store. When collecting this app data, Pew Research Center used the categories in the Google Play Store and conducted no additional categorization of the apps in the dataset.

The majority of apps in the Google Play Store (82%) were free to download at the time of the data collection. Most, but not all, apps that are free to download were supported by advertising. On average, free apps ask for two more permissions than paid apps (Six permissions vs. four permissions.)

The Google Play Store contained more than 1 million apps, but the overwhelming majority of these apps had been installed by only a small number of users. Close to half (47%) of all apps available had been installed fewer than 500 times, and more than 90% had been installed fewer than 50,000 times. On the other end of the spectrum, a relatively small number of apps had been installed by vast numbers of users. 8

Indeed, four apps were downloaded between 1 billion and 5 billion times as of September 2014 — Google Maps, Gmail, Google Play Services and YouTube. It is important to note that all four of these apps, however, are required downloads for all Android devices. Put another way, users did not necessarily choose to download these apps, they came preloaded on their device (or at least users were prompted to download them automatically when setting up their device for the first time).

In the next group of most downloaded apps, there are two that are not required by Google. Seven apps — Facebook, Google Play Books, Google+, Google Search, Google Text-to-Speech, Google Street View and WhatsApp — had been installed between 500 million and 1 billion times as of September 2014. Of those seven, Facebook and WhatsApp (a messaging app owned by Facebook) are not Google apps and are not required installs on Android devices. 9 Google did relax its pre-installed app requirement somewhat in August of 2015 when the firm removed Google+ and Google Play Books from the list of required apps. 10 The main reasons apps stay updated are to keep up with Google updating the Android operating system and to deliver new features to their users. With half the apps not updating at all in 2014, it is clear that many apps are not trying to keep up with this process or are simply not being used.

• The category of “games” has 17 subcategories displayed here. In the Google Play Store “Games” is a super category in which users can see all games combined together or browse by game category; here the subcategories of “Games” are displayed individually. “Games” reported as a single category would be the largest at around 11% of the Google Play Store. Eight apps did not have category information. ↩
• 7,293 apps did not have install information. ↩
• Google Search is referred to as “Google” in the current version of the app referenced here. ↩
• Hildenbrand, Jerry. “Your new phone will have less Google bloatware, and that’s awesome.” Androidcentral.com. August 19, 2015. ↩

Sign up for our weekly newsletter

Fresh data delivery Saturday mornings

Sign up for The Briefing

Weekly updates on the world of news & information

• Emerging Technology
• Online Privacy & Security
• Privacy Rights

Measuring News Consumption in a Digital Era

Rising share of lawmakers – but few republicans – are using the term latinx on social media, sharp decline in remittances expected in 2020 amid covid-19 lockdowns in top sending nations, what is machine learning, and how does it work, for global legislators on twitter, an engaged minority creates outsize share of content, most popular, report materials.

• Interactive : Google Play Store Apps Permissions

1615 L St. NW, Suite 800 Washington, DC 20036 USA (+1) 202-419-4300 | Main (+1) 202-857-8562 | Fax (+1) 202-419-4372 |  Media Inquiries

Research Topics

• Age & Generations
• Coronavirus (COVID-19)
• Economy & Work
• Family & Relationships
• Gender & LGBTQ
• Immigration & Migration
• International Affairs
• Internet & Technology
• Methodological Research
• News Habits & Media
• Non-U.S. Governments
• Other Topics
• Politics & Policy
• Race & Ethnicity
• Email Newsletters

ABOUT PEW RESEARCH CENTER  Pew Research Center is a nonpartisan fact tank that informs the public about the issues, attitudes and trends shaping the world. It conducts public opinion polling, demographic research, media content analysis and other empirical social science research. Pew Research Center does not take policy positions. It is a subsidiary of  The Pew Charitable Trusts .

Copyright 2024 Pew Research Center

Terms & Conditions

Privacy Policy

Cookie Settings

Reprints, Permissions & Use Policy

• Google Play Marcomms
• Data Overload
• Facet Wealth Case Study
• Modern Data Platform Case Study
• OAC/Snowflake
• The Architect Room
• Elastic Delivery
• Analytics Engine
• The Avalanche Initiative
• The Glimpse Initiative
• Implementations
• ETL Converter
• Red Pill Library

Red Pill Analytics builds a serverless digital marketing analytics platform for Google Play Marcomms in the Google Cloud Platform

With more than one billion users, Android has the largest installed base of any operating system. A contributor to that success is Dom Elliott, prior Head of Developer Marketing Communications (Marcomms) and current Product Manager at Google Play, who spent his time improving engagement with Android developers. By examining data from services including Youtube, Twitter, and an internal email marketing platform, his team could see the efficacy of marketing campaigns targeting game and app developers on the Google Play marketplace.

Google Play Marcomms uses multiple digital channels to engage with developers, and each platform has it’s own siloed analytics capabilities. At least once a month, Dom would login to each platform and run a series of reports, dumping the data into Google Sheets, and manually compile and aggrege for final delivery into a Google Slides deck used to communicate engagement to his team’s management. It was a lot of wasted hours; the team needed a real-time analytics platform capable of showing general trends across all the different channels. Although Dom and his team worked for Google, they were not engineers and did not have access to development resources or other internal Google assets: they only had access to the Google Cloud Platform. For this reason, the final solution needed to be serverless, maintenance-free, and run entirely in GCP.

An iterative process, GCP event streaming, and data visualization tools

To solve the Marcomms requirement, Google partnered with Red Pill Analytics, led by founder Stewart Bryson. The project team utilized an agile and iterative delivery methodology enabled by the collaborative features in Google Data Studio, and serverless event streaming data developed using Pub/Sub and Cloud Functions. With no infrastructure to provision, the project team delivered the data processing and analytics of a new digital marketing source every week, with the entire project lasting only three months.

In the end, GCP event streaming and data visualization services addressed the client’s low maintenance requirements. Stewart and his team focused on serverless solutions that could scale and continue running once the implementation was complete. The team used Google’s BigQuery Data Transfer Service to load data from Google services, and a combination of Pub/Sub and Cloud Functions to access non-Google services with public APIs. Organizing all the digital channel information in BigQuery made it easily accessible from Data Studio, providing a single view of engagement for all of Google Play Marketing Communications.

Using analytical insights to improve efficiency

The analytics platform built by Red Pill Analytics allows for both a holistic and targeted understanding of Google Play users. Besides having deeper and more integrated analytics, the extensive time saving offered by the tool allows the Google Play team to focus on their core responsibility: enabling a more productive Android developer community.

One More Thing

In addition to Marcomms’ requirements for the project, Red Pill decided to add in sentiment analysis , which allows them to track attitudes and opinions on various channels. Dom found this feature to be the most surprising and beneficial feature of the entire project and particularly useful for LinkedIn and Twitter. Within a few hours of the latest Android release, Google knows whether the feedback is generally positive or negative and can respond accordingly.  Out of all of the features from this project, sentiment analysis was unexpected but provided the largest value to the Google Play Marcomms team.

Related Links

Check out the Google Cloud Platform Podcast featuring Stewart Bryson and Dom Elliott where they discuss this project.

• Google Natural Language API
• Google Cloud Pub/Sub
• Google Cloud Functions
• Google BigQuery
• Google Data Studio
• RPA GCP Partner Page

WHAT DO THEY THINK?

• Give the Google Play Marcomms team access to streaming digital marketing data in an accessible format they can easily understand and use to make well-informed decisions
• Build an easy-to-use and self-sustaining data and analytics platform for all of Google Play’s digital marketing channels
• Build in the Google Cloud Platform (GCP)

THE APPROACH

• Use Red Pill’s agile and iterative Capacity Analytics service to define and capture numerous social media and digital marketing sources and land them in GCP
• Move from discovery to dashboard for each distinct marketing source in roughly one week’s time
• Deliver an event-driven and serverless ETL process to store raw source data in a data lake in Google Cloud Storage, and curate and aggregate analytics data in Google BigQuery
• Utilize the interactive and collaborative capabilities of Data Studio to keep the customer engaged throughout the process
• Use only GCP services that are truly serverless, ensuring the solution is scalable and maintenance-free post-implementation

THE RESULTS

• An analytics platform measuring user engagement with the Google Play platform
• Sentiment analysis allows the team to track attitudes and opinions over time
• No more wasted time pulling data from each platform and creating reports for internal presentations

THE PRODUCTS

• Google Cloud Platform: App Engine, BigQuery, Google Cloud Storage, Data Transfer Service, Data Studio
• Google Products: Google Analytics, YouTube, Firebase, Google Sheets
• APIs: Twitter, Google+, LinkedIn

About Red Pill Analytics

Latest tweets, latest posts.

• How to easily get metrics per dbt Model in Snowflake
• Data Monitoring: using dbt and SnowFlake’s time-travel to track your KPIs
• dbt tests: False negatives when testing between project layers
• Congratulations to our Oracle ACE Director, Deiby Gómez!
• Snowpark’s Main Features and its Support for Python

Google Play Store Data Analysis

Google Play Store is a place where an Android device user can download apps for his/her requirement onto the device.There are thousands of apps available in Google Play Store.

In this project, I am trying to analysis the data available in the dataset obtained from https://www.kaggle.com/lava18/google-play-store-apps . This dataset contains details about the Google Play Apps,Category to which it belongs to,Ratings etc.I am using the Pandas,Numpy and Matplotlib libraries for the analysis.

The course Data Analysis with Python: Zero to Pandas was very good.The lectures was very easy to understand.I learned the basics of Python,application of Pandas and Numpy Library, Data Visualization using Matplotlib from this course.The case study gave us a practical exposure to all that we learned in the earlier lectures.

How to run the code

This is an executable Jupyter notebook hosted on Jovian.ml , a platform for sharing data science projects. You can run and experiment with the code in a couple of ways: using free online resources (recommended) or on your own computer .

Option 1: Running using free online resources (1-click, recommended)

The easiest way to start executing this notebook is to click the "Run" button at the top of this page, and select "Run on Binder". This will run the notebook on mybinder.org , a free online service for running Jupyter notebooks. You can also select "Run on Colab" or "Run on Kaggle".

Option 2: Running on your computer locally

Install Conda by following these instructions . Add Conda binaries to your system PATH , so you can use the conda command on your terminal.

Create a Conda environment and install the required libraries by running these commands on the terminal:

• Press the "Clone" button above to copy the command for downloading the notebook, and run it on the terminal. This will create a new directory and download the notebook. The command will look something like this:
• Enter the newly created directory using cd directory-name and start the Jupyter notebook.

You can now access Jupyter's web interface by clicking the link that shows up on the terminal or by visiting http://localhost:8888 on your browser. Click on the notebook file (it has a .ipynb extension) to open it.

Downloading the Dataset

In this project, I am trying to analysis the data available in the dataset obtained from https://www.kaggle.com/lava18/google-play-store-apps . This dataset contains details about the Google Play Apps,Category to which it belongs to,Ratings etc.I downloaded the dataset using the opendatasets Python library and extracted it using the os Python library.

Let's begin by downloading the data, and listing the files within the dataset.

A Multi-Modal Neural Embeddings Approach for Detecting Mobile Counterfeit Apps: A Case Study on Google Play Store

Ieee account.

• Change Username/Password
• Update Address

Purchase Details

• Payment Options
• Order History
• View Purchased Documents

Profile Information

• Communications Preferences
• Profession and Education
• Technical Interests
• US & Canada: +1 800 678 4333
• Worldwide: +1 732 981 0060
• Contact & Support
• About IEEE Xplore
• Accessibility
• Terms of Use
• Nondiscrimination Policy
• Privacy & Opting Out of Cookies

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. © Copyright 2024 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

Tung M Phung's Blog

Data cleaning case study: google play store dataset.

This post attempts to give readers a practical example of how to clean a dataset. The data we wrangle with today is named Google Play Store Apps, which is a simply-formatted CSV-table with each row representing an application.

Dataset Name : Google Play Store Apps Dataset Source : Kaggle Task : Data cleaning Language : Python

Column description

Overall, there are 13 columns:

• App: Application name.
• Category: Category the app belongs to.
• Rating: Overall user rating of the app (as when scraped).
• Reviews: Number of user reviews for the app (as when scraped).
• Size: Size of the app (as when scraped).
• Installs: Number of user downloads/installs for the app (as when scraped).
• Type: Paid or Free.
• Price: Price of the app (as when scraped).
• Content Rating: Age group the app is targeted at – Children / Mature 21+ / Adult.
• Genres: An app can belong to multiple genres (apart from its main category). For eg, a musical family game will belong to Music, Game, Family genres.
• Last Updated: Date when the app was last updated on Play Store (as when scraped).
• Current Ver: Current version of the app available on Play Store (as when scraped).
• Android Ver: Min required Android version (as when scraped).

(copied from the data source .)

Data Cleaning

Load and take an overview.

In general, to have an overview of the data frame, I would print out the following information:

• data shape: the number of instances and features.
• several data rows: to have a sense of the values each data point may contain.
• data types of the columns.
• common statistics of the data frame (using .describe() method)
• Missing-value status (using .isna() method).
• The number of unique values for each column.

Remove duplicates

The above shows that we have 10841 rows, however, only 9660 of their App names are unique. This raises a question: Do the duplicated names refer to the same app or not?

If the Play Store restricts the app’s names that different apps must have different names, those duplicates in the dataset are duplicated data points and should be handled so that only 1 of the duplicates remains.

However, it turns out that Google does allow apps with exactly the same name, except for the names that were declared as trademarks, in which case, the names will be protected by the laws. However, we have no clue if any names from these applications have been registered as a trademark, so, to maintain data integrity, we assume that every name in this dataset is duplicatable (i.e. many apps may have the same name).

In other words, we should not delete a row just because its app name is identical to another row. Nevertheless, it is unrealistic to think that 2 different apps may be the same in every property, from the name, rating, reviews to size, etc. So, we remove the duplicates that coincide with another one in all the listed features.

Column: Last Updated

This should obviously be a Date Time column, yet, pandas recognizes it as of object type (the default data type). This implies there is an issue with this column, let’s fix it.

A simple effort to cast this column to datetime gets the following error:

There seem to be some rows with ill-formatted values. The below code will show those rows out.

The problem is clear: in row 10472, there is a missing for the value at the Category. To solve this, we shift all the values of this row to the right:

On a side note, shifting the values of this row to right also solves the problem you may have noticed at the beginning of this work, that the maximum value of the Rating is perceived to be 19.0.

Now, how about the category of this app? Well, I do a search on the Play Store and take it that this app belongs to the LIFESTYLE group. Hoping that this app’s category hasn’t changed from 11 Feb 2018, we fill it in:

Ok, so we can convert the Last Updated column into datetime without any errors.

Column: Rating

The rating is already in the form of a float-like string, we only need to make a call to convert it.

Column: Reviews

Similarly, the Reviews column is also ready to be cast to the Integer type.

Column: Size

Amongst all the apps in this dataset, none has its size reach 1GB. All the values we have to parse either end with ‘M’ (Megabyte), ‘k’ (kilobyte) or equal to ‘Varies with device’. The below script verifies that all of our 10358 rows fall into 1 of these 3 options.

When converting these values into float, we must have them in the same unit, either MB or kB is fine. In this notebook, I choose to use the Megabyte unit, thus each app’s size that was originally in kB will be divided by 1024.

The last problem is: How to handle the ‘ Varies with device ‘ value. Technically, this is not so much different from NaN as we have no information about the actual size of the application. However, the take is we know that these apps have different versions available at once.

To conclude, we set the rows with ‘ Varies with device ‘ NaN and make a new dummy column named Variable Size to better distinguish them.

Column: Installs

The number of installs is shown by buckets. Let’s see which buckets are there:

The smallest values are ‘0’, ‘0+’, and ‘1+’. This is quite surprising to me that both ‘0’ and ‘0+’ exist even though we need only one of them. After a bit of searching on the internet, I couldn’t find the official source of Play Store about these bucket counts. However, as stated by most other sources, ‘x+’ has the lower bound as x+1. That is, ’50+’ means 51-100, ‘100+’ means 101-500, and so on. Thus, it makes sense to deduce that ‘0’ means 0, ‘0+’ means 1, ‘1+’ means 2-5, etc. (Please correct me if I’m wrong.)

Follow the above rule, we convert this column to Integer type:

Column: Type

Apart from Free and Paid, this column also has 1 NaN, which is shown below:

Notice that this is also the only row with 0 Install, which makes me feel there might be something wrong with this record.

Further investigation shows that this game seems to appear on Play Store later than the recorded time in the dataset. Wikipedia says it was first on Play Store from 2018-12-04, which is after the Last Updated value 2018-06-28.

Because of this ambiguity, I decide to remove this row from the data.

This column, Type, is then replaced by ‘Is Free’ – a dummy variable.

Column: Price

The price of an app seems to follow this rule: ‘0’ if it is free, else a dollar sign followed by a floating-point number. Let’s check if there are any exceptions:

There is no exception. Great. We can go straight to the conversion code:

Note that although this Price column does cover the information of the Is Free column, I decide to keep both of them for now to emphasize the difference between free and paid apps. This use of dummy variables is elaborated in the post about when should we add a dummy variable .

Later, in case we are concerned with multicollinearity or similar issues, we would remove Is Free if necessary.

Column: Content Rating

Let’s take a look at all the content rating tags:

These tags seem to follow the standard of America .

We have 2 problems with this column: How should Unrated be understood and How to handle this ordinal variable.

To have a clearer view of the situation, let’s check the distribution of the values:

There are only 2 unrated apps in the dataset. As the number is so small, we have the option to remove them if we think doing so will not bias our data mining process.

For now, I decide to keep these rows to not affect the completeness of the dataset. ‘Unrated’ will be converted to NaN in the following step.

A detailed tutorial on how to treat ordinal variables is given in another post. In short, for practical purposes, ordinal variables are recommended to be converted to interval in most cases. We will stick with this for now.

Note that aside from converting tags to the corresponding ages, we also have the option to convert them to ranking, i.e. ‘Everyone’ is mapped to 0, ‘Everyone 10+’ to 1, ‘Teen’ to 2, etc.

Column: Android Ver

Look at the value counts:

and the missing values:

we have these observations:

Note that we by removing ‘Android Ver’, we actually lose some information about the minor Android version. It depends on our purposes on mining this dataset that we decide to extract the minor version or not (in most case, the answer seems to be No).

Column: Current Ver

Some information can be extracted from this column are:

It is also worth mentioning that many values in this field do not follow the standard. I list some of those values in the table below. We will treat those as NaNs.

Column: Category and Genres

We will convert these 2 columns to numerical with one-hot encoding. It is lucky that there are no NaNs here. However, note that while an app may only belong to 1 category, there might be multiple genres associate with it.

Handling the Category is simple:

It involves a bit more work for the Genres:

Everything is done. Here are the first 5 rows of the resulting data frame after being cleaned.

What we have done are:

The Jupyter Notebook containing full code is given here .

Leave a Reply Cancel reply

Help | Advanced Search

Computer Science > Cryptography and Security

Title: a multi-modal neural embeddings approach for detecting mobile counterfeit apps: a case study on google play store.

Abstract: Counterfeit apps impersonate existing popular apps in attempts to misguide users to install them for various reasons such as collecting personal information or spreading malware. Many counterfeits can be identified once installed, however even a tech-savvy user may struggle to detect them before installation. To this end, this paper proposes to leverage the recent advances in deep learning methods to create image and text embeddings so that counterfeit apps can be efficiently identified when they are submitted for publication. We show that a novel approach of combining content embeddings and style embeddings outperforms the baseline methods for image similarity such as SIFT, SURF, and various image hashing methods. We first evaluate the performance of the proposed method on two well-known datasets for evaluating image similarity methods and show that content, style, and combined embeddings increase precision@k and recall@k by 10%-15% and 12%-25%, respectively when retrieving five nearest neighbours. Second, specifically for the app counterfeit detection problem, combined content and style embeddings achieve 12% and 14% increase in precision@k and recall@k, respectively compared to the baseline methods. Third, we present an analysis of approximately 1.2 million apps from Google Play Store and identify a set of potential counterfeits for top-10,000 popular apps. Under a conservative assumption, we were able to find 2,040 potential counterfeits that contain malware in a set of 49,608 apps that showed high similarity to one of the top-10,000 popular apps in Google Play Store. We also find 1,565 potential counterfeits asking for at least five additional dangerous permissions than the original app and 1,407 potential counterfeits having at least five extra third party advertisement libraries.

Submission history

Access paper:.

• Other Formats

References & Citations

• Google Scholar
• Semantic Scholar

DBLP - CS Bibliography

Bibtex formatted citation.

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

• Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .