Search

Predicting time and location of crime has been an important research topic that is potentially beneficial for governments and citizens alike. In this paper, we study crime prediction as a recommendation problem, using fine-grained open crime data. Given fine-grained spatial–temporal units, crime data would become very sparse. Modeling crime prediction as a recommendation problem, however, allows us to use methods in recommendation systems that deal with data sparsity. In addition to the problem formulation, we propose an extended version of matrix factorization, called contextually biased matrix factorization (CBMF) to solve the problem. Focusing on two major types of crimes in the city of San Francisco, we evaluate our approach against several baseline methods. The experimental results show that our method can outperform traditional crime prediction methods and is comparable with state-of-the-art recommendation methods. Specifically, our method captured over 90% of future thefts using only 50% man-hour, 5% more than the most effective traditional crime prediction method. • Crime prediction can be formulated as a recommendation problem. • Recommendation methods mitigate the sparsity problem in crime prediction. • Biased matrix factorization is designed to incorporate social context. • CBMF outperforms state-of-the-art recommendation and non-recommendation methods. [ABSTRACT FROM AUTHOR]