SimpleSentinel Parental Control App Review Study, November 2026: data files
Study page: https://simplesentinel.com/parental-control-app-reviews-study   Press: https://simplesentinel.com/press

corpus_summary.json     6,731 public U.S. App Store reviews, ten apps, pulled 2026-11-04; average rating, one-or-two-star count, reviews by year.
per_app.csv             per app: reviews in sample, average rating, share one-or-two star, share five star, oldest and newest review year.
per_theme.csv           per theme: reviews matching, share of 6,731, average rating, share one-or-two star.
per_app_per_theme.csv   the theme counts broken out by app.
keyword_patterns.txt    the exact patterns behind each theme.
theme_counts.py         the script. Run it on the review file and it reproduces per_theme.csv and per_app_per_theme.csv.
hand_check.json         the blind hand check: a reader labeled 200 reviews (used to tune the patterns) and a further 120 held-out reviews
                        before seeing the classifier's output; per theme and pooled precision, recall and agreement for both samples.

How to read the theme counts: the patterns favor precision over recall (held-out sample: pooled precision 70 percent, recall 54 percent,
agreement 95 percent across 720 review-by-theme decisions), so each theme count is a floor, not an estimate of the true share. Themes
were assigned by the script, not by a person reading every review; a person read the 320 sampled reviews and the central quotes of
every cluster. A review may count toward more than one theme. Nothing was excluded. The review text belongs to its authors and Apple;
we publish counts and short excerpts, not the corpus.