PhD opportunity: Advancing emotional geographies using artificial intelligence

Supervisory team

• Dr Stef De Sabbata
• Dr Katy Bennett
• Prof Ivan Tyukin

Project Highlights

1. To develop a ground-breaking, interdisciplinary approach to exploring the emotional relationship between people and place
2. To harness the capabilities of recent advances in deep learning to conduct multimodal textual and visual analysis including street-view imagery
3. To develop a predictive model capable of estimating how places are experienced.

Project Overview

The project aims to better understand people’s emotional connections with places by implementing and testing geospatial, multimodal text and image processing models, expanding previous work developed for social media analysis to analyse virtual walking interview data we generated through the Mapping Multiculture project.

In a previous work (Liu and De Sabbata, 2021), we demonstrated how multimodal autoncoders and graph-convolutional neural networks can effectively spatially analyse text and images from social media. We illustrated how to generate models that combine qualitative content analysis with large-scale quantitative analysis to predict specialised and project-specific topics in social media content. In the Mapping Multiculture project (Bennett et al., forthcoming), we have illustrated the effectiveness of street-view imagery in an immersive virtual reality environment as a stimulus in the qualitative interview process and generated a rich dataset of personal experiences of Leicester. That opened up opportunities to explore the emotional connections between people and place, identifying distinct emotional ‘atmospheres’ associated with different parts of the city. The key idea behind the project here proposed is to apply approaches developed in the first project to the data and context of the second project.

Successful completion of this project will lead to significant advances in geospatial artificial intelligence, thus generating a better understanding of where tensions (and easiness) are experienced, especially in places with particularly dynamic, diverse populations such as Leicester. Exploring people’s emotional reactions to places is crucial to further our understanding of their experiences of heritage and culture in a digital age (WP1). However, we need geospatial, multimodal text and image processing tools to effectively link interview transcripts and street-view photography from interviews, along with relevant big data, to identify people’s emotional reactions to places (WP2). Finally, we need to explore the generalisability and predictive capabilities of the developed models to critically assess the practical and ethical implications of their use (WP3).

The results from these studies will help us further our understanding of the relationship between place and emotion and the role that geospatial artificial intelligence can take in conducting digital geographical research on these topics at scale.

Methodology

We will conduct three Work Packages (WP) in parallel, which will feedback one another.

WP1: Content Analysis. Qualitatively analyse data generated during the Mapping Multiculture project through virtual walking interviews using Google Earth VR to assess emotional connection with places. Collect street-view imagery. Identify gaps, data imbalances and opportunities for further interviews to support WP2 and WP3. WP2: Modelling. Explore supervised and semi-supervised machine learning approaches such as multimodal autoncoders and graph-convolutional neural networks to develop geospatial, multimodal text and image processing tools that effectively link interview transcripts and street-view photography from interviews (WP1) to identify people’s emotional reaction to places. WP3: Evaluation and critical assessment. Apply the developed models (WP2) in new scenarios to test their generalisability and predictive capabilities. Explore the use of big data from user-generated content platforms (e.g., social media, Wikipedia) as supplementary data for generalisation. Critically assess the practical and ethical implications of their use.

References

• Ballatore, A., & De Sabbata, S. (2020). Los Angeles as a digital place: The geographies of user‐generated content. Transactions in GIS, 24(4), 880-902.
• Bennett K, Gardner Z and De Sabbata (forthcoming). Digital geographies of everyday multiculturalism: ‘Let’s go Nando’s!’ Social and Cultural Geography
• De Sabbata, S., & Liu, P. (2019). Deep learning geodemographics with autoencoders and geographic convolution. In Proceedings of the 22nd AGILE conference on Geographic Information Science, Limassol, Greece.
• Liu, P., & De Sabbata, S. (2021). A graph-based semi-supervised approach to classification learning in digital geographies. Computers, Environment and Urban Systems, 86, 101583.
• Neal S, Bennett K, Cochrane A and Mohan G (2018) The Lived Experiences of Multiculture: the new spatial and social relations of diversity, Routledge.