BREAKING
Bellingcat AI Flags Civilian-Harm Posts
0
+
geolocated incidents
0
total training items
0
features used
Positive vs negative samples
Harm posts
5848
Non-harm posts
48545
XGBoost wins model bake-off
XGBoost
Selected
●
Best on all metrics
●
Semantic similarity signals
Random Forest
Compared
●
Crying-face top feature
●
Baseline tree model
Crying-face emoji is a top signal
Open notebook for the OSINT community
AI NEWS BLITZ
Bellingcat has open-sourced a machine learning model to detect civilian-harm posts on Telegram.