A Python library for intelligent HTML segmentation and ROI extraction. It builds a DOM tree from raw HTML and extracts content-rich regions for efficient web scraping and analysis.
Project description
betterhtmlchunking
A Python library for intelligently chunking HTML documents into structured, size-limited segments based on DOM tree analysis.
Our Discord Server:
Overview
This library processes HTML content to split it into semantically coherent chunks while respecting specified size constraints. It analyzes the DOM structure to identify optimal split points, preserving contextual information and document hierarchy.
DeepWiki Docs: https://deepwiki.com/carlosplanchon/betterhtmlchunking
Key Features
- Custom DOM tree representation.
- Configurable chunk size limits (counting by text or HTML length).
- Intelligent region-of-interest detection.
- Dual output formats: HTML and plain text chunks.
- Preservation of structure relationships.
- Customizable tag filtering.
Installation
pip install betterhtmlchunking
Dependencies
- Python 3.12+
- attrs
- treelib
- beautifulsoup4
- parsel-text
- lxml
- attrs-strict
Usage
Basic Example
from betterhtmlchunking import DomRepresentation
from betterhtmlchunking.main import ReprLengthComparisionBy
from betterhtmlchunking.main import tag_list_to_filter_out
html_content = """
<html>
<body>
<div id="content">
<h1>Document Title</h1>
<p>First paragraph...</p>
<p>Second paragraph...</p>
</div>
</body>
</html>
"""
# Create document representation with 20 character chunks.
dom_repr = DomRepresentation(
MAX_NODE_REPR_LENGTH=20,
website_code=html_content,
repr_length_compared_by=ReprLengthComparisionBy.HTML_LENGTH
# tag_list_to_filter_out=["/head", "/header", "..."] # By default tag_list_to_filter_out is used.
)
dom_repr.start()
# Render HTML:
for idx in dom_repr.tree_regions_system.sorted_roi_by_pos_xpath:
print("*" * 50)
print(f"IDX: {idx}")
roi_html_render: str =\
dom_repr.render_system.get_roi_html_render_with_pos_xpath(
roi_idx=idx
)
print(roi_html_render)
# Render text:
for idx in dom_repr.tree_regions_system.sorted_roi_by_pos_xpath:
print("*" * 50)
print(f"IDX: {idx}")
roi_text_render: str =\
dom_repr.render_system.get_roi_text_render_with_pos_xpath(
roi_idx=idx
)
print(roi_text_render)
Render output (HTML):
**************************************************
IDX: 0
<h1>
Document Title
</h1>
**************************************************
IDX: 1
<p>
First paragraph...
</p>
**************************************************
IDX: 2
<p>
Second paragraph...
</p>
Render output (text):
**************************************************
IDX: 0
Document Title
**************************************************
IDX: 1
First paragraph...
**************************************************
IDX: 2
Second paragraph...
Configuration
Key Parameters
MAX_NODE_REPR_LENGTH: Maximum allowed length for each chunk (in characters)repr_length_compared_by: Length calculation method:- ReprLengthComparisionBy.HTML_LENGTH: HTML source length
- ReprLengthComparisionBy.TEXT_LENGTH: Rendered text length
website_code: Input HTML content
Advanced Features
# Access the DOM tree structure
tree = dom_repr.tree_representation.tree
# Get node metadata:
for node in tree.all_nodes():
if node.data is not None:
print(f"XPath: {node.identifier}")
print(f"Text length: {node.data.text_length}")
print(f"HTML length: {node.data.html_length}")
How It Works
-
DOM Parsing
- Builds a tree representation of the HTML document.
- Calculates metadata (text length, HTML length) for each node.
-
Region Detection
- Uses Breadth First Search (BFS) to traverse the DOM tree in a level-order fashion, ensuring that each node is processed systematically.
- Combines nodes until the specified size limit is reached.
- Preserves parent-child relationships to maintain contextual integrity.
-
Chunk Generation
- Creates HTML chunks with original markup.
- Generates parallel text-only chunks.
- Maintains chunk order based on document structure.
Comparison to popular Chunking Techniques
The actual practice (Feb. 2025) is to use plain-text or token-based chunking strategies, primarily aimed at keeping prompts within certain token limits for large language models. This approach is ideal for quick semantic retrieval or QA tasks on unstructured text.
By contrast, betterhtmlchunking preserves the HTML DOM structure, calculating chunk boundaries based on each node’s text or HTML length. This approach is especially useful when you want to:
- Retain or leverage the hierarchical relationships in the HTML (e.g., headings, nested divs)
- Filter out undesired tags or sections (like
<script>or<style>) - Pinpoint exactly where each chunk originated in the document (via positional XPaths)
You can even combine the two techniques if you need both structured extraction (via betterhtmlchunking) and LLM-friendly text chunking (via LangChain) for advanced tasks such as summarization, semantic search, or large-scale QA pipelines.
CLI
The package ships with a small command line interface built with Typer. You can pipe HTML to the tool and obtain a specific chunk as HTML:
cat input.html | betterhtmlchunking --max-length 32768 --chunk-index 0 > chunk.html
By default the command reads from stdin, processes chunks up to a maximum length of 32,768 characters, and prints the HTML corresponding to chunk index 0 to stdout.
Verbose mode
You can enable progress logging with --verbose. Logs are written to stderr so they don’t interfere with chunk output:
cat input.html | betterhtmlchunking --max-length 32768 --chunk-index 0 --verbose > chunk.html
Maximal Verbose Mode
For a detailed inspection of the DOM, nodes, ROIs, and chunk lengths, use --maximal-verbose. This logs:
- Total DOM nodes and their HTML/text lengths
- Each ROI (region of interest) with constituent node XPaths and lengths
- Final chunk HTML and text sizes
cat input.html | betterhtmlchunking --max-length 32768 --chunk-index 0 --maximal-verbose > chunk.html 2> logs.txt
chunk.htmlcontains the selected chunk HTML.logs.txtcaptures all detailed logging information.
This mode is useful for debugging, testing, or analyzing how the document is split into chunks.
License
MIT License
Contributing
Feel free to open issues or submit pull requests if you have suggestions or improvements.
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file betterhtmlchunking-0.9.4.tar.gz.
File metadata
- Download URL: betterhtmlchunking-0.9.4.tar.gz
- Upload date:
- Size: 18.2 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.14.2
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
11f2a1ff085d55421bb3fe130f7d1076d723a2b210a64d222ee93aa7c9f53c81
|
|
| MD5 |
3fc41261263d944f4aa8e8e18fcd5a09
|
|
| BLAKE2b-256 |
1a4bd24f9b401c8325a6c79e3512867e58743a8f72cc5726a71900ab9a8940da
|
File details
Details for the file betterhtmlchunking-0.9.4-py3-none-any.whl.
File metadata
- Download URL: betterhtmlchunking-0.9.4-py3-none-any.whl
- Upload date:
- Size: 14.0 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.14.2
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
4cdb707e2ac1b39a575ee0f1aa21170ffeb7998cf96f7829457b0812fc55bd65
|
|
| MD5 |
737cd18eeb31a38142d0f906b3603d3f
|
|
| BLAKE2b-256 |
76f8b59a6c91c1a4582207d95ce27fc35d4f1416b2de9edac0b98e5183b5c92a
|
