Rakuten Review VOC Collector

Collect complete, auditable review corpora from public Rakuten Ichiba product review pages for Voice of Customer (VOC), market research, and downstream text analysis.

Use this Actor when you already know which products matter and need the review corpus with evidence that every expected page was collected.

Why this Actor is different

Many Rakuten Actors are designed for product discovery, listing extraction, or a small sample of recent reviews. This Actor is deliberately narrower: it starts from a researcher's selected product list and collects the review corpus in a form that can be audited.

Need	Product/search Actors	Sample-oriented review Actors	This Actor
Find products by keyword/category	Primary use	Sometimes included	Not included
Collect a few recent reviews quickly	Indirect	Primary use	Supported with a page cap, but not the main purpose
Collect every available review page	Often not included	May stop at a review limit	Default behavior
Run a ranked batch of selected products	Varies	Often URL-by-URL	Yes; research IDs and ranks are preserved
Verify collection completeness	Usually row count only	Usually row count only	Page-level and run-level audit records
Prepare raw data for VOC coding/NLP	Requires reshaping	Often limited fields	Rich review, reviewer, SKU, item, and page metadata
AI-generated review summary	Sometimes included	Sometimes included	No; returns source data for your own analysis

Apify compute, storage, and data-transfer usage may incur charges according to the user's Apify plan and this Actor's Store pricing configuration.

When to use it

Choose this Actor when:

• You have a defined product universe, such as a top-N list, competitor set, or tracked SKU list.
• You need all available public reviews rather than a small preview.
• You must retain product rank and stable research keys across a multi-product crawl.
• You need to prove which review pages were collected and identify gaps before analysis.
• You want raw review-level evidence for taxonomy coding, sentiment analysis, topic modeling, or Excel/BI reporting.

Choose another Actor or workflow when:

• You still need to discover products by keyword, category, ranking, or seller.
• You only need a handful of recent reviews for a quick check.
• You want an AI summary instead of the underlying review corpus.
• You need non-public, authenticated, or account-specific Rakuten data.
• You need browser automation for interactions beyond public product review pages.

Input

products is required. Each product must provide either:

• review_url, or
• both shop_id and item_id.

Research fields such as product_rank, item_key, shop_code, and item_url are strongly recommended. They are carried into the output so the dataset can be joined back to the selected product universe.

{
  "products": [
    {
      "product_rank": 1,
      "item_key": "e-techpowershop:10000062",
      "item_id": "10000062",
      "shop_id": 429056,
      "shop_code": "e-techpowershop",
      "item_url": "https://item.rakuten.co.jp/e-techpowershop/set-d1018/",
      "review_url": "https://review.rakuten.co.jp/item/1/429056_10000062/1.1/"
    }
  ],
  "concurrency": 5,
  "max_pages_per_product": 0
}

Controls

Field	Meaning
concurrency	Parallel requests, from 1 to 10. Default: 5. Reduce it if Rakuten starts returning errors.
max_pages_per_product	0 collects all detected pages. A positive number caps pages per product for pilots or quick samples.

Rakuten currently exposes up to 30 reviews per review page. The Actor reads the displayed review count on page 1 and calculates the expected page count automatically.

Output

The default dataset intentionally contains three record types. Filter on record_type before analysis.

record_type	One row per	Purpose
review	Review	VOC text and metadata, including rating, title, body, dates, reviewer attributes when public, SKU information, helpful count, media, item metadata, and source URL.
page_audit	Review page	Coverage evidence: page number, expected pages, displayed review count, and reviews found on the page.
run_audit	Actor run	Run totals and a failures array for products or pages that could not be collected.

For a complete-corpus run, validate all three conditions before analysis:

1. run_audit.failures is empty.
2. Every expected page number has a corresponding page_audit row.
3. Unique review_key count is consistent with the collected page totals and the displayed product review count.

The review_key and encrypted_review_key fields support downstream deduplication. The Actor itself returns source rows and does not silently remove records.

Validation evidence

In an internal production QA run on 2026-08-20, 50 selected products achieved coverage of 312 / 312 expected review pages, producing 8,613 unique reviews after downstream deduplication. This is validation evidence for that run, not a guarantee that Rakuten page structure or availability will remain unchanged.

Limits and responsible use

• Collects public Rakuten Ichiba product review pages only.
• Does not search for products, rank products, or choose a market sample.
• Does not use residential proxies or bypass authenticated access controls.
• Rakuten page-structure changes can require parser maintenance.
• A run can finish while individual pages are listed in run_audit.failures; always inspect the audit record.
• Use a reasonable concurrency and follow applicable laws, Rakuten terms, and your organization's data-handling requirements.