Drag the types of action you consider extreme
Action types
Data & Methodology
The dataset behind this project covers political actions carried out by Israeli citizens in public spaces, in Israel and the occupied territories, from the beginning of 2023: demonstrations, roadblocks, clashes, assaults, property damage, land seizure, and similar actions. Remarks, posts, and statements unaccompanied by action on the ground are not included.
Event descriptions and dates are taken from ACLED, an international research organization that documents protest and political violence based on reports from media outlets and local sources, and from HaMivtzar's settler violence log; an event appearing in both sources is counted once. Group assignments, action classifications, and crowd-size estimates were derived from the descriptions as part of this project. The descriptions were edited and shortened, and ACLED descriptions were translated into Hebrew, preserving the facts and the context in which the events took place. The English descriptions shown here were translated from the edited Hebrew versions.
Each event was assigned to the group that carried out its central action, not to those harmed or to the security forces that responded. When the group carried out several actions, the event was classified by the most severe one, using a severity ranking of ten action types built for the project with reference to existing severity indices. Assignment, classification, and translation were carried out with OpenAI language models, following instructions written for the project and refined over many rounds of review and manual correction.
The division into political groups is intended to enable comparison between the patterns of action of different camps. It does not imply that all members of a group act in the same way or hold the same views. The dataset also reflects what was reported, not everything that happened, and classifying political events is not always clear-cut.
Full methodology: sources, selection rules, classification, translation, duplicate detection, and the complete model instructions.
Research, design, and development: Eyal Raz (eyalraz.com)
Mentorship and guidance: Mushon Zer-Aviv
Journalistic advisor: Oren Persico
Thanks to Rakefet Canaan, the Shenkar School of Visual Communication, Shenkar - Engineering. Design. Art.
Thanks to the HaMivtzar team · Anat Saragusti, Miki Rosenthal, Amir Sperling, Ronit Antler and others
Database: ACLED, HaMivtzar