Entity Identification Based on Human Mobility Data

Ren, Huimin

Etd

Entity Identification Based on Human Mobility Data

Public Deposited

With the rapid development of the urbanization, the government has not only benefited from it, but also needs to deal with great challenges of urban governance, such as transportation issues, unhealthy economy development, public safety etc. To address the challenges, the government requires an accurate understanding of urban entities. For example, it is necessary to know whose work may involve criminal and where is the dangerous location. With the increasing adoption of smart devices and GPS modules, more and more human mobility data is generated in cities, such as trajectory data and check-in data, which makes it possible to infer the meaning of urban entities. In this dissertation, we propose and develop several novel machine learning techniques to identify entities in the urban environment based on human mobility data. In particular, we try to solve the entity identification problems in the following aspects. 1) Map-constrained Trajectory Recovery. First of all, most of the spatio-temporal data mining problems, including urban entity identification, rely on high-sampling-rate trajectories since fine-grained trajectories can provide more information. In this paper, we revisit the fundamental research topic, trajectory interpolation, to enhance the trajectory data and support the urban entity identification more effectively. We propose a Map-constrained Trajectory Recovery framework, MTrajRec, to recover the fine-grained points in trajectories and map match them on the road network in an end-to-end manner. Extensive experiments based on large-scale real-world trajectory data confirm the effectiveness and efficiency of our approach. 2) Human Mobility Signature Identification. Identifying human mobility signature is a significant task in the urban entity identification, which is helpful in many real-world applications, e.g., identifying drivers in ride-hailing service and criminal identification. In this topic, we propose the human mobility signature identification solution to identify human agents from their mobility data. We make the first attempt to match identities of human agents only from the observed location trajectory data by proposing a novel and efficient framework named Spatio-temporal Siamese Networks (ST-SiameseNet). Experimental results on a real-world taxi trajectory dataset show that our proposed ST-SiamesNet can achieve an F1 score of 0.8508, which significantly outperforms the state-of-the-art techniques. 3) Company Real Workplace Recognization. Local business development can help create jobs and increase tax revenue. Company real workplace identification is important for the government to manage companies. In this work, we recognize company's real workplace by implementing a novel two-step data mining method from employees' check-in data. Experimental results show that our proposed method significantly outperforms six baselines. And A real-world application system has been deployed in Nantong, China since September 2021, demonstrating the effectiveness of our solution. 4) Illegal Chemical Facility Detection. Detecting illegal chemical facilities is related to public safety. In this paper, we develop a system for illegal chemical facility detection based on hazardous chemical truck trajectories. We first generate candidate locations by clustering stay points extracted from trajectories and filtering out known locations. Then, we rank those locations in suspicion order by modeling whether it has loading/unloading events. ICFinder+ is evaluated over the real-world dataset from Nantong in China, and the deployed system identified 20 illegal chemical facilities in 3 months. 5) Learn to Stay. Stay points of trajectories are key to understanding location and human mobility. Most of the previous work focus on GPS trajectories which are limited since not all of the vehicles are equipped with GPS modules. In this work, we design a two-stage method named SAInf to detect stay points via surveillance camera records which can capture the moving patterns for all the vehicles. SAInf first detects which surveillance camera record pairs have stay events within them, and then uses a layer-by-layer stay area identification algorithm to infer the exact location of an object stay. Experimental results show that SAInf outperforms other baselines by 58%.

Creator