Violence Detection

The violence classifier is a specialized text classifier that analyses media posts that communicate violent actions, whether about specific threats of physical violence or accounts of violence against others.


TypeSpeedPartner Type
Post ProcessingInstantDatastreamer Internal

Example Use Cases

The violence classifier is designed to uncover instances where people communicate violent threats or reports of violence against others. Used in conjunction with keywords, named entities, and location information, users can filter where threats might be associated with this other pertinent information. The classifier is meant to identify explicit instances of violent language and not about the subject of violence per se.

Compatible Data Scoures

As a stream-integrated classifier, it is run on ingestion for specific sources.

Applicable Data SourcesCompatible?
wsl_InstagramYes, English only
wsl_blogsYes, English only
wsl_forumsYes, English only
data365_twitter_keywordsYes, English only


Recipe Available

View the below recipe to see it in action, and easily view how to integrate it into your own data pipeline.


The violence classifier outputs 3 different labels relating to violence: ‘No violence’, ‘Threat of physical violence’, and ‘Reported violence’, as well as a confidence value for each prediction.

"enrichment": {
    	"reported_violence": {
      "label": "No violence",
      "confidence": 0.9926758