Using Deep Machine Learning to Understand Green Space Popularity, Activities, Functionality, and Design Implications: a comparison of Chicago and Stockholm

Abstract

<p>Green spaces have a significant effect on urban living environments, providing shared natural areas and entertainment spaces among other benefits. Investigating their popularity and functionality is valuable for supporting green space design and guiding land use development around public green spaces. Conventional methods used to extract use and functionality information on public green spaces have typically relied on questionnaires and in-site observations, both resource and time consuming processes. These approaches can also be context dependent and produce less transferable data across regions. This study utilizes deep machine learning techniques to mine social media text and image data to produce useful information on public green space popularity, activities, functionality and design.  Convolutional neural networks (CNN), an advanced machine learning technique, is used to analyze large scale data sets (Yelp and publicly accessible images) of public green spaces in Chicago, Illinois, US. The coupling of the two types of data enables the extraction of a rich and comprehensive analytical frame for understanding how green spaces are used and how they might be improved. The technique also utilizes a complex transfer learning process to pretrain the model and allow its quick adaptation to other regions around the world. The process will be replicated in Stockholm, Sweden. We use the analysis generated to compare open spaces in the 2 cities. The process has the potential to substantially improve nature based strategies on green space development and design.</p>