Francesco/mappa-mundi

Mappa Mundi

53,868 historical events from 3000 BC to today, each placed on the globe and ranked by how much the rest of Wikipedia points at it. The top of the list is World War II, World War I, the COVID-19 pandemic and the American Civil War. It's the data behind Mappa Mundi, an atlas of history you drag through time, and it's built by the pipeline in its repo (scripts/).

Two tables, same rows in the same order, joined on qid:

file	what	rows
events.arrow	the facts: when, where, what kind, how important	53,868
leads.arrow	the English Wikipedia text: lead, description, thumbnail	53,868

Both are Feather v2 (Arrow IPC file), ZSTD-compressed, one record batch, which pandas, polars and pyarrow read natively. After the unzip, numeric columns load with no parsing: a reader maps them straight onto the bytes.

events.arrow

Rows are sorted by score, most important first.

column	type
qid	string	Wikidata id, e.g. Q362
label	string	English label
article	string	English Wikipedia title, e.g. World_War_II
category	dictionary<string>	battle, war, politics, epidemic, disaster, exploration, culture, founding, landmark, history
start, end	float32	historical years, negative = BC (no year 0). end = start for a point in time
lon, lat	float32	where it happened, WGS84 degrees
position_source	dictionary<string>	which rule placed it, see below
score	float32	importance, see below
inlinks	float32	English Wikipedia articles linking to it
sitelinks	float32	Wikipedia language editions with an article on it

Position, first match wins (position_source):

1. wikidata: the event's own Wikidata coordinates.
2. wikipedia: its English article's {{coord}}. Many battles have one even when Wikidata doesn't, naval battles especially.
3. sub-events: the median of its located sub-events, when it has at least two. A war is where its battles were.
4. location: its first-listed Wikidata location. For a naval battle that is often a sea, which is the right answer.

Events none of these can place are left out, and so are articles with fewer than 10 in-links.

Score = robust in-degree × category weight.

• In-degree counts incoming links (CirrusSearch incoming_links). It's capped at 60 × sitelinks^1.5, because links arriving through navbox templates inflate some articles 20,000×, and real prominence also shows up across languages.
• The category weight damps places (founding 0.15, landmark and culture 0.3), whose links measure the modern city rather than its founding.

leads.arrow

column	type
qid	string	joins events.arrow
description	string?	Wikipedia's one-line short description
extract	string?	the lead section as plain text, paragraphs separated by blank lines
lead_html	string?	the same lead as minimal HTML: <p>, <b>, <i> and <a href="/wiki/…"> only, footnotes and coordinates stripped
thumbnail	string?	URL of the article's lead image on Wikimedia, 640px
fetched_at	string?	when the lead was fetched (UTC)

Null means Wikipedia had nothing for that field.

Load it

import pandas as pd
events = pd.read_feather("hf://datasets/Francesco/mappa-mundi/events.arrow")
leads = pd.read_feather("hf://datasets/Francesco/mappa-mundi/leads.arrow")

import polars as pl
events = pl.read_ipc("hf://datasets/Francesco/mappa-mundi/events.arrow")

from datasets import load_dataset
events = load_dataset("Francesco/mappa-mundi", "events", split="train")

import { tableFromIPC } from '@uwdata/flechette'
const url = 'https://huggingface.co/datasets/Francesco/mappa-mundi/resolve/main/events.arrow'
const events = tableFromIPC(await (await fetch(url)).arrayBuffer())
events.getChild('lon').toArray() // a Float32Array over the downloaded bytes, no copy

Sources and license

• Wikidata (CC0): the events, their classes, dates, coordinates, locations and sitelinks, queried through QLever.
• English Wikipedia: in-links, coordinates, descriptions and leads, through the Action API. The text in leads.arrow is CC BY-SA 4.0, so the dataset as a whole is CC BY-SA 4.0: credit Wikipedia and share alike.
• Thumbnails are links to Wikimedia Commons, not copies. Each image keeps its own license, shown on its Commons page.