App Store Review Intelligence

Collect public Apple App Store reviews and turn them into lightweight product-intelligence signals: negative/neutral/positive rating classes, keyword-group tags, app metadata, and an app-level summary.

What it does

• Accepts Apple numeric app IDs or App Store URLs.
• Reads public Apple review RSS pages for a selected country storefront.
• Emits one dataset row per review with rating, title, text, version, author name, URL, and matched keyword groups.
• Writes a SUMMARY key-value-store record with per-app rating distribution, signal counts, latest review timestamp, and total review counts.

Good use cases

• Product teams tracking recent App Store complaints and feature requests.
• Agencies comparing competitors before app-store optimization or retention work.
• Founders validating whether users complain about pricing, crashes, missing features, or onboarding.
• Review-intelligence pipelines that need clean rows without credentials or proxies.

Input example

{
  "appIds": ["284882215"],
  "country": "us",
  "maxReviewsPerApp": 100,
  "pagesPerApp": 2,
  "keywordGroups": {
    "pricing": ["price", "subscription", "expensive", "free", "pay"],
    "bugs": ["bug", "crash", "broken", "freeze", "error"],
    "features": ["feature", "wish", "please add", "missing", "need"]
  }
}

You can also pass appUrls, for example:

{
  "appUrls": ["https://apps.apple.com/us/app/facebook/id284882215"],
  "country": "us",
  "maxReviewsPerApp": 50
}

Output fields

Dataset rows include:

• appId, appName, developer, country
• reviewId, rating, ratingClass, version
• title, content, author, updatedAt
• matchedGroups, matchedKeywords
• reviewUrl, appUrl, collectedAt

The SUMMARY record includes per-app review counts, rating distributions, signal counts, and run parameters.

Notes and limits

• Uses public Apple iTunes/App Store RSS and lookup endpoints only; no login, proxies, or private APIs are required.
• Apple RSS typically returns up to 50 reviews per page. Use pagesPerApp and maxReviewsPerApp to control cost and runtime.
• Reviews are storefront-specific; set country to the market you care about.
• Keyword matching is simple substring tagging intended for fast triage, not a substitute for human review or a full NLP sentiment model.