How Newsmap classifies events
Every dot on Newsmap carries a set of labels — a disorder type, an event type, an interaction code, and a handful of other tags. Those labels are what make the map filterable and searchable, but they are also the part most likely to be misunderstood. This article explains where the taxonomy comes from, what each layer means, and why we chose a structured scheme instead of free text.
Why a taxonomy at all
A raw feed of incident reports is almost impossible to reason about at scale. Two reports might describe the same kind of event — say, a protest that turned violent — using completely different words. If you want to answer a question like “how much armed conflict happened in this region this week,” you need a controlled vocabulary: a fixed set of categories that every event is mapped onto, regardless of how the original report was phrased. That is what a taxonomy gives you.
The ACLED framework
Rather than invent our own categories, Newsmap follows the structure popularised by ACLED (the Armed Conflict Location & Event Data Project), which has spent years refining a vocabulary for political violence and disorder. Using an established framework means our labels line up with how researchers and journalists already think about these events, and it forces a useful discipline: an event has to fit somewhere in the scheme, which surfaces ambiguous cases instead of hiding them.
The layers of a label
Each event is described along several axes, from broad to specific:
- Disorder type — the widest bucket. It separates, for example, organised political violence from public demonstrations from strategic developments. This is the layer most people filter on first.
- Event type — the specific action being reported: a protest, an airstrike, an arrest, a clash, an explosion. This is what the map uses for most of its colour coding.
- Sub-event type — a finer distinction within an event type, where the source gives enough detail to support one.
- Interaction — who was involved and how. ACLED encodes the actor groups (state forces, rebel groups, civilians, and so on) and their interaction numerically. It is the most technical field and the one we surface last.
- Entity — named groups, organisations, or individuals mentioned as key actors, extracted from the report text.
How a report becomes a labelled event
Incoming reports run through an automated pipeline. A report is first translated into English where necessary, then summarised, geolocated to a place and country, and finally classified against the taxonomy above. The classifier proposes the most likely label at each layer; where it is genuinely unsure, the event still appears, but its lower-confidence fields may be sparse or marked as unavailable rather than guessed.
Geolocation deserves a note of its own. Many reports name a city or region explicitly, which gives a precise location. Others only imply a country, in which case the event is placed at a representative point and flagged with lower location confidence. Newsmap hides the lowest-confidence placements from the default view so the map does not imply precision it does not have.
What the labels are not
A classification is a description of what was reported, produced automatically. It is not a verdict on whether the event happened, who was at fault, or how it should be interpreted politically. Labels can be wrong — a sarcastic post misread as a literal claim, an ambiguous event type, a misattributed actor. That is exactly why every event links back to its original source, and why we pair the taxonomy with a separate veracity and severity signal. Treat the labels as a fast way to navigate and filter, and treat the source as the thing you actually cite.
Putting it to use
On the map, the taxonomy is what powers the filter panel: you can narrow to a single disorder type, combine event types, or isolate a named entity, then read the result spatially or as a feed. If you want the per-country rollups built from these same labels, see the country reports, and for a tour of the interface, read reading the map.