Readable Web Extractor Comparison

How do various readable website extractor libraries (ie. libraries that provide a feature like Reader View in Safari) perform?

This repo exists to provide a way to compare many libraries at once across many pages at once.

Currently the following libraries are implemented:

• mozilla/readability
• cleanview
• metascraper
• @postlight/mercury-parser
• TODO - clean-mark (377 stars)
• TODO - ascrape-js (13 stars)

Results

The latest output from running the comparisons on a set of 16 random pages selected from Hacker News in June 2020 is available on the gh-pages branch (direct link to report).

Based on these comparisons @awendland is intending to use the mozilla/readability project.

Example Report

Usage

Make sure to run yarn to ensure all dependencies are installed. Each command should include --help documentation and produce explanatory output during execution.

Fetching Test Pages

Create a newline delimited list of URLs to fetch and store them in a text file such as test_urls.txt.

Use the fetch-test-pages script to retrieve and save them into a folder such as test_pages/ for report processing.

yarn scripts:run ./scripts/fetch-test-pages.ts --listOfUrls test_urls.txt --outDir test_pages/ --parallelism 30

They will be saved as JSON files containing information such as the source URL and the HTML contents of the page.

Generating Comparison Report

Once test pages have been retrieved a report can be generated. The following command would be used to generate a report named report.html from test pages saved in test_pages/.

yarn scripts:run ./scripts/generate-report.ts --testPages 'test_pages/*.json' --reportFile report.html

Contributing

Adding New Libraries for Comparison

Adding a new library to the comparison involves several steps:

1. Add the library (and any associated @types/ package) as a project dependency

   yarn add LIBRARY_NAME --exact
   

2. Authoring an adapter for the library in scripts/lib/adapters/adapter-LIBRARY_NAME.ts which conforms to the following type (detailed in scripts/lib/types.ts):

   type Adapter = {
     metadata: AdapterMetadata
     extract(params: ExtractParams): Promise<ExtractedInfo | null>
   }
   

3. Registering the adapter in scripts/lib/adapters/index.ts

4. Generating a report to make sure that it works