Skip to main content

DocxToMarkdownPipeline

What it is​

A document conversion pipeline that reads a DOCX file and produces Markdown by extracting paragraphs from word/document.xml and mapping basic Word paragraph styles (headings, lists) to Markdown syntax.

Public API​

  • DocxToMarkdownPipelineConfiguration

    • Extends ConvertToMarkdownBasePipelineConfiguration.
    • mime_type: Defaults to application/vnd.openxmlformats-officedocument.wordprocessingml.document.

  • DocxToMarkdownPipelineParameters

    • Extends ConvertToMarkdownBasePipelineParameters.
    • processor_iri: Defaults to http://ontology.naas.ai/abi/document/DocxToMarkdownProcessor.

  • DocxToMarkdownPipeline

    • Extends ConvertToMarkdownBasePipeline.
    • convert_to_markdown(file: File) -> str: Converts a DOCX File to a Markdown string.

Configuration/Dependencies​

  • Depends on:
    • naas_abi_marketplace.domains.document.pipelines.ConvertToMarkdownBasePipeline (base classes/config/params)
    • naas_abi_marketplace...File for reading file bytes (file.read() must return bytes)
    • Standard libs: zipfile, xml.etree.ElementTree, re, io
    • pydantic.Field and typing.Annotated for configuration metadata

Usage​

from naas_abi_marketplace.domains.document.pipelines.DocxToMarkdownPipeline import (
    DocxToMarkdownPipeline,
)
from naas_abi_marketplace.domains.document.ontologies.classes.ontology_naas_ai.abi.document.File import (
    File,
)

pipeline = DocxToMarkdownPipeline()
docx_file: File = ...  # must support .read() -> bytes for a DOCX
markdown = pipeline.convert_to_markdown(docx_file)
print(markdown)

Caveats​

  • Only processes word/document.xml; invalid DOCX content or missing word/document.xml raises ValueError("Invalid DOCX content: missing word/document.xml").
  • Markdown conversion is minimal:
    • Headings: only paragraph styles matching Heading1…Heading6.
    • Lists: any paragraph with numbering properties (w:numPr) becomes - ... (no ordered list numbering).
    • Inline formatting (bold/italic/links/images/tables) is not handled; text is extracted from w:t nodes only.
  • Whitespace normalization collapses runs of spaces/tabs to a single space and trims ends; explicit w:tab becomes four spaces and w:br/w:cr become newline characters (which may be normalized).