# Changelog of Amazon Review Classifier — complaints & returns · $0.5/1k (`leoworks/amazon-review-classifier`) Actor

- **URL**: https://apify.com/leoworks/amazon-review-classifier/changelog.md
- **Full Actor documentation**: https://apify.com/leoworks/amazon-review-classifier.md

## Changelog

### 0.2.x (2026-10-18) — collect and classify in one run

- **New input "Products"** (ASINs or Amazon product URLs, up to 20): the Actor collects the reviews with an existing Amazon reviews scraper and classifies them in the same run. Default collector junglee/amazon-reviews-scraper (about $6 per 1,000 reviews, billed to your account by that Actor; Free Apify plan: 10 reviews per run), optional web_wanderer/amazon-reviews-extractor. Reviewer personal data is switched off, no Amazon login or cookies are used. New fields: Marketplace, Reviews to collect per product, Collect: order, Collector. The run log and `SUMMARY.collection` show the live cost estimate (about $6.50 per 1,000 reviews all-in with junglee). If the collector finds nothing the run stops with the reason and this Actor charges nothing.
- Existing inputs, output, event and price are unchanged. The README no longer repeats its introduction, and has a compliance and privacy section.

### 0.3 (beta) — review photo evidence

- **Judge review photos (beta, off by default, no extra charge)** — checks the photos buyers attached when the dataset has them (`reviewImages` and similar): `photoEvidence` = `supports_complaint_product`, `supports_complaint_packaging`, `damage_seen_text_silent`, `no_evidence` or `not_product_photo`, plus `photoStatus`, `photosJudged`, `photoLabels` and `photoUrl`. Reviews whose photo supports the complaint are flagged "act on it"; photo-backed product defects go to the top of `ACTION_QUEUE.csv`, packaging damage right after (new last columns `photoEvidence`, `photoUrl`). New dataset view "Photo evidence (beta)".
- Settings: Judge review photos (off / complaints / all), Photos judged per review (1–3, at most 2 with "all"), Photo size (512–1024), Photo field.
- Photos are downloaded only while they are judged and sent to TypeSafe (model `jev-1.14.0`); nothing is stored and no question is about people. With the option off, rows, summary and CSV columns are unchanged.

### 0.2 (2026-10-11)

- **Works with more scrapers** — field names are matched ignoring case, `_` and `-` (PascalCase and snake_case datasets no longer fail), ratings like "4.0 out of 5 stars" are read, and variants given as lists are joined.
- **What buyers praise** — 9 aspects (build quality, matches the listing, fit, performance, ease of use, looks, value, shipping, seller service). Check set: 91% correct.
- **Action queue** — `ACTION_QUEUE.csv` and the "act on it" flag: returns or refunds, safety complaints and "not as described".
- **Star vs. text mismatch** flag (no extra charge).
- **Readable rows** — flat columns `asin`, `variant`, `publishedAt`, `verified`, `vine`, `helpfulVotes`, `complaintTypes`, `praiseTypes`, `sentiment`, `returnIntent`, `action` next to the full `labels` object (unchanged). Three table views.
- **Summary by product as CSV** — `REPORT_PRODUCTS.csv`. REPORT adds return reasons, worst variants, verified share and verified-only complaint rate, Vine count, star distribution, monthly trend, last 90 days vs before and a low-sample flag.
- **Filters** (skipped reviews are not charged): newer than a date or period, at most N stars, verified purchases only, at most N reviews per product. Duplicate reviews are classified and charged once.
- **Field mapping**: star rating, review date and variant fields can be set by hand.
- Price unchanged.

### 0.1

- First version: Amazon review classification — complaint types (defect, not as described, size/fit, performance, usability, shipping, value, safety, seller service), sentiment, return intent, custom labels, per-product summary (REPORT). Field detection for the Amazon reviews scrapers on Apify Store.