In major urban cities, government authorities have developed various service-requesting systems to report non-emergency public issues related to urban rare events such as noise, blocked driveways, illegal parking, etc. For certain events, request volumes can surge significantly, and timely response depends on accurate prediction. In this paper, we investigate how long it takes to resolve service requests by the agencies. This paper introduces NERPS, a non-emergency response system designed to forecast service request response time. Leveraging urban data, the model establishes connections between historical and future response times. In time series data, applying boot-strapping on the reminder component for generating synthetic data with original time series before fitting the model has been viewed to be effective. The NERPS integrates Holt-Winters with the Moving Block Bootstrap (MBB+HW) model for forecasting the service requests in the NYC dataset. Proposed model forecasts to generate 100-time series values and final prediction obtained by averaging the forecast set. The optimal block size is estimated via the flat-top lag windows technique. This research extends beyond prior studies by comparing the forecasting performance of proposed statistical methods with MI/DL approaches on complex and nonlinear time series data. We consider SARIMA, ARIMA, FB-Prophet, linear regression and basic LSTM as baseline models for response time forecasting and compare the proposed model with multistep ahead point forecasts. The results show that in most cases, the NERPS achieves low RMSE, MAE and Relative Errors among top complaint types and agencies.

We propose a semi-supervised learning method called Cformer for automatic clustering of text documents in cases where clusters are described by a small number of labeled examples, while the majority of training examples are unlabeled. We motivate this setting with an application in contextual programmatic advertising, a type of content placement on news pages that does not exploit personal information about visitors but relies on the availability of a high-quality clustering computed on the basis of a small number of labeled samples.