dom-content-extraction

Crates.io Downloads (recent) GitHub License

A Rust library for extracting main content from web pages using text density analysis. This is an implementation of the Content Extraction via Text Density (CETD) algorithm described in the paper by Fei Sun, Dandan Song and Lejian Liao: Content Extraction via Text Density.

See specs/overview.md for detailed architecture and internals.

What Problem Does This Solve?

Web pages often contain a lot of peripheral content like navigation menus, advertisements, footers, and sidebars. This makes it challenging to extract just the main content programmatically. This library helps solve this problem by:

Analyzing the text density patterns in HTML documents
Identifying content-rich sections versus navigational/peripheral elements
Extracting the main content while filtering out noise
Handling various HTML layouts and structures

Key Features

Build a density tree representing text distribution in the HTML document
Calculate composite text density using multiple metrics
Extract main content blocks based on density patterns
Unicode Support
Support for nested HTML structures
Efficient processing of large documents
Error handling for malformed HTML
Markdown output (optional feature) - Extract content as structured markdown
Article extraction — get_article/DensityTree::extract_article returns main-article text with sidebar and ticker content excluded, using the same anchor-and-walk-up strategy as the markdown path

Unicode Support

DOM Content Extraction includes Unicode support for handling multilingual content:

Proper character counting using Unicode grapheme clusters
Unicode normalization (NFC) for consistent text representation
Support for various writing systems including Latin, Cyrillic, and CJK scripts
Accurate text density calculations across different languages

This ensures accurate content extraction from web pages in any language, with proper handling of:

Combining characters (like accents in European languages)
Bidirectional text
Complex script rendering
Multi-code-point graphemes (like emojis)

Usage

MSRV is 1.85 due to 2024 edition. Living on the edge!

Basic usage example:

use scraper::Html;
use dom_content_extraction::get_content;

fn main() {
    let html = r#"<!DOCTYPE html><html><body>
        <nav>Home | About</nav>
        <main>
            <article>
                <h1>Main Article</h1>
                <p>This is the primary content that contains enough text to maintain proper density metrics. The paragraph needs sufficient length to establish text-to-link ratio.</p>
                <p>Second paragraph adds more textual density to ensure the content extraction algorithm works correctly.</p>
                <a href="\#">Related link</a>
            </article>
        </main>
        <footer>Copyright 2024</footer>
    </body></html>"#;

    let document = Html::parse_document(html);
    let content = get_content(&document).unwrap();
    println!("{}", content);
}

Installation

Add it it with:

cargo add dom-content-extraction

or add to you Cargo.toml

dom-content-extraction = "0.4"

Optional Features

To enable markdown output support:

dom-content-extraction = { version = "0.4", features = ["markdown"] }

Documentation

Read the docs!

dom-content-extraction documentation

Library Usage with Markdown

use dom_content_extraction::{DensityTree, extract_content_as_markdown, scraper::Html};

let html = "<html><body><article><h1>Title</h1><p>Content</p></article></body></html>";
let document = Html::parse_document(html);
let mut dtree = DensityTree::from_document(&document)?;
dtree.calculate_density_sum()?;

// Extract as markdown
let markdown = extract_content_as_markdown(&dtree, &document)?;
println!("{}", markdown);
# Ok::<(), dom_content_extraction::DomExtractionError>(())

Article Extraction (ticker/sidebar-free)

get_content uses a contiguous-block heuristic that can sweep in dense sidebar/ticker content appearing before the article body in the DOM. For clean article text, use get_article, which anchors at the densest subtree and walks up to its enclosing container:

use dom_content_extraction::{get_article, scraper::Html};

let document = Html::parse_document(html);
// Excludes "Latest News" tickers, sidebars, and other high-density noise
// that precedes the article body in the DOM.
let article = get_article(&document).unwrap();

Run examples

Check examples.

This one will extract content from generated "lorem ipsum" page

cargo run --example check -- lorem-ipsum

This one prints node with highest density:

cargo run --example check -- test4

Extract content as markdown from lorem ipsum (requires markdown feature):

cargo run --example check -- lorem-ipsum-markdown

Run extraction over every page in html/pages.zip (text default, or --article for ticker-clean output, or --markdown):

cargo run --example check_pages -- --article

There is scoring example i'm trying to implement scoring. You will need to download GoldenStandard and finalrun-input datasets from:

https://sigwac.org.uk/cleaneval/

and unpack archives into data/ directory.

cargo run --example ce_score

As far as i see there is problem opening some files:

Error processing file 730: Failed to read file: "data/finalrun-input/730.html"

Caused by:
    stream did not contain valid UTF-8

But overall extraction works pretty well:

Overall Performance:
  Files processed: 653
  Average Precision: 0.88
  Average Recall: 0.83
  Average F1 Score: 0.78
  Average Sorensen-Dice: 0.79
Total processing time: 11.32s
Average time per file: 17.34ms

Read documentation on docs.rs

CLI Tool

For command-line usage (URL fetching, file processing, encoding detection), see pageinfo-rs.

dom-content-extraction 0.4.4