# Changelog of Article Content Extractor — Clean Text, Markdown & Metadata (`haketa/article-content-extractor`) Actor

- **URL**: https://apify.com/haketa/article-content-extractor/changelog.md
- **Full Actor documentation**: https://apify.com/haketa/article-content-extractor.md

## Changelog

### 0.1.5 — 2026-11-01

- **Fixed a metadata value bug**: JSON-LD `author` and `inLanguage` values that are nested objects or arrays of objects (e.g. `[{ "@type": "Person", "name": "..." }]` or `[{ "@type": "Language", "alternateName": "en-US" }]`) are now read correctly instead of stringifying to `[object Object]`. A new `ldText()` helper resolves `name` / `alternateName` / `@value` from any JSON-LD shape, and `clean()` now guards against any raw object reaching a text field.
- **More accurate `author`, `lang`, `keywords`, `siteName` and `title`** on publishers that use structured (Schema.org) author/language objects — these now return the real human-readable value where the page exposes one.
- **Verified against a fresh 500-URL run** across 31 distinct publishers (Wikipedia, BBC, The Guardian, TechCrunch, WIRED, Ars Technica, The Verge, MDN, Stack Overflow, GitHub, Netflix Tech Blog, Medium and more). Result: 500/500 records, 0 duplicate URLs, and among successfully fetched pages — `title` 100%, `wordCount` 100%, `lang` 100%, `text`/`markdown` 99.8%, `excerpt` 98.7%, `author` 98.1%, `siteName` 94.8%, `publishedDate` 89.8%. Zero `[object Object]` values after the fix.
- **README refreshed** to the latest output: corrected the Output example and field-reference table to match real extracted records, added honest coverage notes for data-dependent fields, and expanded the FAQ, tips and pagination/scale sections.

### 0.1.0 — 2026-08-25

- Initial release: paste any list of article, blog or news URLs and get back one clean record per URL — title, author, publish/modified dates, excerpt, site name, language, keywords, main image, word count and reading time, plus clean plain text and LLM-ready Markdown.
- Main-content extraction (reader-mode style) strips navigation, ads, cookie banners and related-post boilerplate.
- Metadata pulled from JSON-LD, Open Graph and standard meta tags with layered fallbacks.
- Parallel processing with configurable concurrency, automatic retries with proxy-session rotation, and per-URL `error` records so a run never hangs.
