Mapping the character of places from georeferenced data

Hermond Tamba


Supervised by Chris B Jones; Moderated by Matthew J W Morgan

The characteristics of locations within cities can be viewed from various perspectives - for example with regard to activities and facilities (e.g. types of shopping. types of restaurants, parks, sports, cinema, cultural heritage, biodiversity etc). This project will attempt to evaluate the effectiveness of a topic modelling approach (using Latent Dirichlet Analysis - LDA), or similar, that uses geo-referenced text as input and that has been applied in recent research literature. Factors that could be considered include methods of selecting terms for input to the topic model, clustering of topics and the spatial resolution of the analysis. It is anticipated that geo-referenced Flickr tags could be used as input for which an existing dataset might be used or data might be mined from Flickr as part of the project. Alternatively other data sources could be considered. LDA could be done with a library such as scikit-learn in Python, but alternative methods of spatial analysis could also be considered.

Final Report (21/10/2022) [Zip Archive]

Publication Form