Taxi demand can be divided into pick-up demand and drop-off demand, which are firmly related to human’s travel habits. Accurately predicting taxi demand is of great significance to passengers, drivers, ride-hailing platforms and urban managers. Most of the existing studies only forecast the taxi demand for pick-up and separate the interaction between spatial correlation and temporal correlation. In this paper, we first analyze the historical data and select three highly relevant parts for each time interval, namely closeness, period and trend. We then construct a multi-task learning component and extract the common spatiotemporal feature by treating the taxi pick-up prediction task and drop-off prediction task as two related tasks. With the aim of fusing spatiotemporal features of historical data, we conduct feature embedding by attention-based long short-term memory (LSTM) and capture the correlation between taxi pick-up and drop-off with 3D ResNet. Finally, we combine external factors to simultaneously predict the taxi demand for pick-up and drop-off in the next time interval. Experiments conducted on real datasets in Chengdu present the effectiveness of the proposed method and show better performance in comparison with state-of-the-art models.

With the rapid development of sensor acquisition technology, more and more data are collected, analyzed, and encapsulated into application services. However, most of applications are developed by untrusted third parties. Therefore, it has become an urgent problem to protect users’ privacy in data publication. Since the attacker may identify the user based on the combination of user’s quasi-identifiers and the fewer quasi-identifier fields result in a lower probability of privacy leaks, therefore, in this paper, we aim to investigate an optimal number of quasi-identifier fields under the constraint of trade-offs between service quality and privacy protection. We first propose modelling the service development process as a cooperative game between the data owner and consumers and employing the Stackelberg game model to determine the number of quasi-identifiers that are published to the data development organization. We then propose a way to identify when the new data should be learned, as well, a way to update the parameters involved in the model, so that the new strategy on quasi-identifier fields can be delivered. The experiment first analyses the validity of our proposed model and then compares it with the traditional privacy protection approach, and the experiment shows that the data loss of our model is less than that of the traditional k-anonymity especially when strong privacy protection is applied.