Google Play Store Apps (Kaggle - lava18)

One crawl, ten thousand apps, every storefront attribute that mattered in 2019. Scraped in February 2019 by Lavanya Gupta and published on Kaggle, the main file holds 10,841 app rows with 13 columns covering category, rating, review count, download size, install bracket, free/paid type, price, content rating, genres, last-update date, current version and minimum Android version. A companion file adds 64,295 translated reviews for 1,074 of those apps, each pre-scored for sentiment.

It is the most-voted tabular Play Store dataset on Kaggle - 5,147 upvotes, 2.06M views, 299,857 downloads and 1,220 associated notebooks - which is why it functions as the de facto benchmark: when a paper claims a better rating-prediction model or a cleaner category analysis, this is usually the table under it. Because nothing has changed since version 6 landed on 2019-02-03, every number anyone has ever published against it stays reproducible. Treat it as a fixed baseline for methodological work; current market sizing belongs to a live feed. Datadory delivers it as a production dataset alongside the rest of the application software catalog.

Flat rows, no nesting, no envelope to unwrap. Two rows straight out of the two files:

App:            Photo Editor & Candy Camera & Grid & ScrapBook
Category:       ART_AND_DESIGN
Rating:         4.1
Reviews:        159
Size:           19M
Installs:       10,000+
Type:           Free
Price:          0
Content Rating: Everyone
Genres:         Art & Design
Last Updated:   January 7, 2018
Android Ver:    4.0.3 and up

And one row from the review-sentiment file, joined back to its app by name:

App                     : 10 Best Foods for You
Translated_Review       : "I like eat delicious food. That's I'm cooking food myself, case \"10 Best Foods\" helps lot, also \"Best Before (Shelf Life)\""
Sentiment               : Positive
Sentiment_Polarity      : 1.0
Sentiment_Subjectivity  : 0.5333333333333333

The first row is the whole value proposition in miniature: an art-and-design app at 4.1 stars from 159 reviews, "10,000+" installs, free, priced at 0, rated Everyone, last touched January 7, 2018. Thirteen facts, one line, 10,840 more lines behind it.

Thirteen documented columns in the app file, verified against the source card with examples pulled from the data itself. Installs arrive as brackets such as 10,000+, Size keeps its unit suffix (19M) or the literal Varies with device, and Genres can carry semicolon-separated sub-genres like Art & Design;Pretend Play. The dictionary below is the complete schema.

Geography - Google Play Store listings in English; the author never documented which storefront country the crawler used, so treat locale as inferred rather than confirmed metadata.

Temporal - a static snapshot scraped in February 2019, last published as version 6 on 2019-02-03. Each row preserves its own Last Updated value at scrape time - "January 7, 2018" in the sample above - so historical distributions stay intact inside the freeze.

Granularity - one row per app in the main file; one row per translated review in the sentiment file. 75,136 combined rows across both, small enough that any laptop loads the lot into memory.

API, files, or your warehouse. Daily, weekly, or hourly.

• Rating-prediction models - the canonical regression target: 13 tidy predictors against a 1-to-5 star score, with 10,841 labeled rows that make results comparable to every published notebook built on the same table.
• Pricing and monetization studies - Type splits free from paid and Price rides along in USD, so paid-app pricing distributions by category fall straight out of a group-by.
• Sentiment analysis benchmarks - 64,295 translated reviews with precomputed polarity (-1 to 1) and subjectivity (0 to 1) give NLP work an instant ground truth without labeling effort.
• Category and genre analysis - Category and semicolon-delimited Genres support market-structure reads: how many apps chase ART_AND_DESIGN versus FAMILY, and how ratings cluster once sub-genres are separated.

Data scientists and ML engineers get the reference teaching corpus - small enough to iterate on, famous enough to benchmark against. Journalists, academics and students get a citable, stable snapshot whose numbers do not move between draft and publication. Market researchers get category counts and rating distributions from a period when the Play Store's composition was already mature. Developers building data products get a flat CSV-shaped schema that prototypes an app-search or analytics demo in minutes.

Three things worth knowing upfront. First, this is a frozen 2019 baseline: the store has changed substantially since, so use it for methodology, teaching and historical comparison rather than current market sizing - live coverage needs a refreshed feed. Second, known data-quality quirks require a cleaning pass: one row carries Category 1.9, some ratings are missing, and duplicate app names exist, so plan deduplication and type coercion in the pipeline. Third, the author documents neither the storefront country nor the scrape methodology, and the sentiment scores come pre-computed by the uploader rather than from a documented model.