Crime by Crowd is Not Crime any More: Predictive Analysis of Riot and Radicalization Oriented Text of Social Media

In plain words

Riots rarely come from nowhere — they simmer in public posts, hashtags and calls to gather, often days before anything happens on the street. This chapter builds a pipeline that reads that simmer and estimates the probability of a riot forming.

A dataset of riot- and radicalization-oriented text was assembled from social media and microblogging sites using the Apriori algorithm and Association Rule Learning — data-mining techniques that surface tell-tale co-occurring patterns. After preprocessing, classification models were trained on it to score new text for riot orientation.

How it works

1. Mine the patternsApriori and association-rule mining pull frequently co-occurring riot-oriented language from public posts.
2. Build the datasetThe mined text is cleaned and labelled into a training corpus.
3. Train classifiersClassification algorithms learn to recognise riot- and radicalization-oriented text.
4. Score the feedNew posts are scored continuously; rising scores act as an early-warning signal of possible unrest.

Why it matters

Civil unrest costs lives and livelihoods, and the window for de-escalation is before the crowd forms. A model that flags rising riot-oriented chatter gives administrators and communities that window — turning hindsight into foresight.