Language usage on Twitter predicts crime rates

Abstract

Social networks 1 produce enormous quantity of data. Twitter, a microblogging network, consists of over 230 million active users posting over 500 million tweets every day. We propose to analyze public data from Twitter to predict crime rates. Crime rates have increased in the past recent years. Although crime stoppers are utilizing various technics to reduce crime rates, none of the previous approaches targeted utilizing the language usage (offensive vs. non-offensive) in Tweets as a source of information to predict crime rates. In this paper, we hypothesize that analyzing the language usage in tweets is a valid measure to predict crime rates in cities. Tweets were collected for a period of 3 months in the Houston and New York City by locking the collection by geographic longitude and latitude. Further, tweets regarding crime events in the two cities were collected for verification of the validity of the prediction algorithm. We utilized Support Vector Machine (SVM) classifier to create a model of prediction of crime rates based on tweets. Finally, we report the validity of prediction algorithm in predicting crime rates in cities.