Review Intelligence

Collect reviews from supported, keyless sources and turn them into a clearly-labelled analysis — while keeping the source reviews and the derived analysis strictly separate. The Actor never rewrites a review, and never presents its analysis as if it were something a customer actually wrote.

What does this Actor do?

1. Collect reviews from:
   • Apple App Store — the keyless iTunes customer-reviews RSS JSON feed, optionally across several countries.
   • Generic JSON-LD — any page URL that embeds schema.org Review markup (reviewUrls).
2. Label every output record with an explicit recordType:
   • review — one record per collected source review, with its verbatim fields.
   • product — app metadata from the iTunes lookup endpoint (when appId is given).
   • analysis — one derived analysis record.
3. Analyse (optional) — derive topics, complaints, feature requests and an overall sentiment tendency from the collected review text. Every figure carries its sample size and the analysisMethod used, so the analysis can never be mistaken for the reviews themselves.

Why the source/analysis split matters

Review content and conclusions about review content are different things. This Actor keeps them in separate records so you can always trace an analysis figure back to the reviews it was computed from — and never accidentally treat a derived sentiment label as a real customer quote.

Input

Field	Type	Description
appId	string	Apple App Store numeric app ID (optional if reviewUrls given).
countries	array	App Store country codes (default ["us"]).
maxPagesPerCountry	integer	Review pages to fetch per country (default 5).
reviewUrls	array	Page URLs that embed schema.org Review JSON-LD (optional).
analyze	boolean	Produce the labelled analysis record (default true).
proxyConfiguration	object	Proxy settings (iTunes RSS is keyless).

Input example

{
  "appId": "284882215",
  "countries": ["us", "gb"],
  "maxPagesPerCountry": 5,
  "reviewUrls": [],
  "analyze": true
}

Output

A dataset of labelled records. Source reviews first, then an optional product record, then a single analysis record.

Source review (recordType: "review")

{
  "recordType": "review",
  "rating": 1.0,
  "title": "Crashes constantly",
  "text": "App keeps crashing and is buggy…",
  "author": "Bob",
  "date": "2026-01-02T00:00:00-07:00",
  "url": "https://itunes.apple.com/us/reviews/id2",
  "source": "apple_app_store",
  "productName": "Cool App",
  "country": "us",
  "appId": "12345",
  "appVersion": "2.0"
}

Derived analysis (recordType: "analysis")

{
  "recordType": "analysis",
  "note": "Derived analysis of the collected reviews — NOT source review text…",
  "analysisMethod": "lexicon+keyword-frequency",
  "sampleSize": 3,
  "ratingSummary": { "sampleSize": 3, "mean": 3.33, "median": 4.0 },
  "positiveTopics": [ { "keyword": "reliable", "reviewsMentioning": 2 } ],
  "negativeTopics": [ { "keyword": "crashing", "reviewsMentioning": 2 } ],
  "commonKeywords": [ … ],
  "complaints": [ { "category": "reliability", "reviewsMentioning": 1, "pctOfReviews": 33.3 } ],
  "featureRequests": [ "…wish they would add dark mode" ],
  "overallSentiment": { "label": "negative", "score": -0.2, "analysisMethod": "lexicon" }
}

You can download the dataset in JSON, CSV, Excel or HTML.