Published May 29, 2026 | Version 1

Measuring spatial injustices - Supplementary Material

Authors/Creators

Description

This repository provides the supplementary material for the article "Measuring spatial injustices: a systematic review of topics, metrics, normative foundations and planning implications in applied geography", to be submitted for publication. 

File name

Content/Info

Scopus search string Copyable to Scopus document search
WoS search string

Copyable to Web of Science Core Collection document search

Additional filters used:

+ Publication Date: years 2015 - 2025 

+ doc type: article

+ language: English

Literature search results (Excel)

 All references to included papers in table format for custom ordering/search/extraction

Literature search results (PDF)

All references to included papers in bibliography format
Code_BERTopic_Analysis_Measuring_Spatial_Injustices
BERTopic pipeline to extract topics and medoids. The model can be run as is on Google Colab (using CPU).

Notes

Note on model testing

The most common and advised clustering method in BERTopic is Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN). HDBSCAN was tested but rendered unsuitable for the problematic number of outliers it produced (papers not assigned to any topic). As all input documents have been manually screened before to assess their fit, a high outlier rate indicated that HDBSCAN does not fit this use case. The centroid-based clustering algorithm k-Means clusters all documents without generating outliers, such that this method was deemed appropriate.

HDBSCAN determines the number of topics independently, whereas k-Means allows selecting the number of topics. The tests with HDBSCAN resulted in between 20 and 25 topics, which was used as a baseline when selecting the final number in k-Means. As an additional test, the silhouette score was calculated. The silhouette score measures how similar a data point is to its cluster, compared to other clusters, and shows if data points are in overlapping clusters, such that the assignment to one cluster instead of the other might be arbitrary. According to the silhouette score, ten topics would result in the best-defined clusters for this text sample. After comparison of the different assignments, the number of topics was set to 20 to capture diverse but well-defined, interpretable topics. As all papers exhibit thematic and methodological overlaps, mild overlaps between clusters are credible.      

Files

Scopus search string.txt

Files (1.8 MB)

Name Size
md5:008de174dd2c89007c41af7de7d1eeb5
380.1 kB Preview Download
md5:1b54cde95388560dcbbfa36cb90f4e48
328.9 kB Download
md5:96352895c45e408c335fb354053f9b37
1.1 MB Preview Download
md5:59ec18e3d9e2217dfe91efe3cecaf1b8
1.5 kB Preview Download
md5:4cfc21f19907327fbd06517884e14ac1
1.4 kB Preview Download

Additional details

Dates

Created
2026-05-29