Parsing the IPC patent classification tree

data engineering
open data
patents
Python
A self-contained Python notebook that downloads WIPO’s official IPC XML schema, reconstructs full hierarchical descriptions for all 79,000+ classification symbols, and uploads the result to Google Sheets.
Published

March 1, 2025

Patent data analysis often begins with the same tedious problem: the International Patent Classification (IPC) system has over 79,000 technology symbols, but their descriptions in raw databases are relative — a leaf node only tells you what it is relative to its parent. To understand what A01B0003280000 actually means, you have to walk up the tree through A01B0003240000, A01B0003000000, A01B, A01, and A, concatenating descriptions as you go.

This notebook solves that once, for all symbols, and pushes the result to a Google Sheet that anyone on a team can use as a lookup table.

What it does

Input: the official IPC master XML file, downloaded on-the-fly directly from WIPO’s public server — no local copy needed.

Pipeline:

  1. Fetch and unzip the schema in memory (no disk writes)
  2. Walk the XML tree, filtering out meta-nodes (i, g, n, t kinds)
  3. Build a parent–child map across all 79,473 symbols
  4. For each symbol, recursively collect ancestor descriptions and concatenate them into a single human-readable string
  5. Convert WIPO’s 14-character binary symbol format (SCCUMMMMGGGGGG) to the standard PATSTAT format (e.g. A01B 1/02) so the output joins cleanly with patent databases
  6. Write a CSV (ipc_descriptions.csv) and upload two sheets to Google Sheets:
    • Schema — every symbol with its full hierarchical description (79,473 rows)
    • Children — every symbol with the flat list of all its descendants

Output example:

WIPO symbol PATSTAT symbol Full description
A01B0003280000 A01B 3/28 Agriculture; Soil working in agriculture or forestry; Soil-shifting machines; Arrangements for converting hand-operated into power-operated ploughs; Ploughs; Rotary tillers; Equipments with working tools; Details

Why it’s useful

Most patent analytics tools expose IPC codes as short strings (A01B 3/28) with no human-readable label attached. Joining against this table adds the full description in one step — useful for labelling charts, filtering by technology area, or explaining results to a non-expert audience.

The Google Sheets output means the whole team has access to an always-current lookup table without any local setup.

Notebook

The full notebook is embedded below and can be downloaded here. To use it yourself, replace YOUR_SHEET_ID in the upload cells with the ID of your own Google Sheet (the string between /d/ and /edit in the sheet URL).