基于群智感知的犯罪风险预测及应用

对犯罪风险进行预测能够帮助预防和减少犯罪。然而，基于数据的犯罪风险预测由于犯罪数据稀疏性而存在严峻挑战。随着群智感知技术快速发展，人们能够采集到多种城市数据，帮助缓解数据稀疏并提升预测准确性。本项目基于群智感知理论和方法，探索犯罪风险预测方法及应用场景。利用多源群智感知异构数据提取与融合重要特征，多视角全面表征犯罪风险；针对目前研究粗粒度问题，提出细粒度犯罪风险预测方法，构建犯罪风险网络并动态识别犯罪风险区域，利用时空图神经网络进行犯罪风险预测，有效缓解数据稀疏性；针对城市犯罪数据缺失问题，利用其他城市数据学习犯罪风险预测模型，并利用无监督领域自适应模型迁移到该城市进行跨城市犯罪风险预测；最后，依托杭州城市大脑平台开展基于犯罪风险的应用研究来验证犯罪风险预测方法的实用性，包括个性化路线推荐和巡逻路线优化。研究成果将提升社会公共安全，为推进智慧城市建设提供依据，具有重要的学术价值和现实意义。

Public safety has been an increasingly critical and severe problem worldwide nowadays, bringing tremendous damage to people’s properties, psychological conditions, and lives. Crime risk prediction can be an effective strategy for crime prevention and reduction. However, the data sparsity issue of crime data poses great challenges to crime risk prediction. With the massive advance of crowdsensing techniques, large amounts of urban open data can be collected to help alleviate the data sparsity and improve crime prediction accuracy. In this study, we plan to explore crime risk prediction methods and applications leveraging crowdsensing data. Using multi-source crowdsensing heterogeneous data, we extract useful features and fuse these multi-scale features to represent crime risk comprehensively. Previous crime prediction studies mainly focus on coarse-grained prediction, which fail in fine-grained road-level prediction due to data sparsity. We propose a fine-grained crime risk prediction method. We construct crime risk networks, identify dynamic crime risk areas and then propose a spatio-temporal graph neural network for crime risk prediction. Furthermore, for crime risk prediction problems in cities without labeled crime data, we learn crime risk prediction model using data from other cities, and then propose an unsupervised domain adaptation model to conduct a cross-city crime risk prediction. Finally, we verify the practicality of proposed crime risk prediction methods through the Hangzhou City Brain platform with two crime risk-based applications, i.e., risk-aware personalized route recommendation and police patrol route optimization. This research will help us promote public safety and benefit smart city construction.