Crime prediction using Twitter sentiment and weather

Abstract

Social networking services have the hidden potential to reveal valuable insights when statistical analysis is applied to their unstructured data. As shown by previous research, GPS-tagged Twitter data enables the prediction of future crimes in a major city, Chicago, Illinois, of the United States. However, existing crime prediction models that incorporate data from Twitter have limitations in describing criminal incidents due to the absence of sentiment polarity and weather factors. The addition of sentiment analysis and weather predictors to such models would deliver significant insight about how crime. Our aim is to predict the time and location in which a specific type of crime will occur. Our approach is based on sentiment analysis by applying lexicon-based methods and understanding of categorized weather data, combined with kernel density estimation based on historical crime incidents and prediction via linear modeling. By testing our model's ability to predict future crime on each area of the city, we observed that the model surpassed the benchmark model, which predicts crime incidents using kernel density estimation.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 137 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.