Sensing Urban Dynamics Through Crowdsourced Data with the Support of Machine Learning Techniques

Crowdsourced data such as social media data, points of interest and geotagged images has attracted the attention of urban researchers, as it provides first-hand information regarding human activities, perception and the interaction with the built environment, helping researchers illuminate a richer sense of what cities are all about from data itself. Despite the appealing potential of crowdsourced data in answering urban-related questions, there are challenges inherent to the whole process of transforming immense data into accurate and actionable insights in the urban domain, which overshadow the benefits of this type of data in well-designed cases of use. This thesis focuses on leveraging crowdsourced data in sensing urban dynamics by implementing advanced machine learning techniques to deal with hitherto unsolved challenges. The main aim is to contribute to the current knowledge on crowdsourced data-driven studies for better management and planning of cities. The thesis starts with a systematic review of crowdsourced data-driven studies in the context of urban research in terms of concepts, data sources, applications and methods evidenced in the literature. The thesis further elaborates on the challenges of crowdsourced data mining and highlights the potential of state-of-the-art machine learning techniques in handling those challenges. An analytical framework of crowdsourced data-intensive studies for sensing urban dynamics is then formulated. Following the framework, this thesis then demonstrates its application by carrying out three empirical studies of the Greater London – sensing urban function from points of interest data, sensing activity pattern from location-based social media data, and sensing public opinion from social media data. These studies illustrate the capability of machine learning techniques in dealing with the challenges of crowdsourced data-driven studies in the context of urban research. Finally, this thesis discusses the implications of the three studies, summarised as evidence-based decision making for planning and near real-time urban management for governance.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2021

Author dc:creator

• Niu, Haifeng

Advisor dc:contributor.advisor

• Silva, Elisabete A

• Urban big data
• Machine learning
• Urban activity patterns
• Social media data
• Urban functional use
• Points of interest
• Public opinions

Licence

• https://www.rioxx.net/licenses/all-rights-reserved/

Language dc:language

eng

Author Identifier

0000-0002-0182-3573

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/334674