Spatiotemporal Change-Points in Development Discourse: Insights from Social Media in Low-Resource Contexts

TL;DR

Analyzes the spatiotemporal evolution of development discourse in a low-resource setting using geotagged X data from Zambia and a mixed methods pipeline. Introduces the durable discourse concept and links online conversations to policy flashpoints such as COVID-19 and geothermal infrastructure to inform participatory development planning. The study identifies seven recurring themes, detects change-points linked to both acute crises and infrastructure interventions, and presents a framework for durable, geospatially-aware development sensing applicable to ICTD and HCI practice. Overall, it demonstrates that social media can serve as a real-time sensor of policy flashpoints, guiding more targeted, equitable development interventions in resource-constrained contexts.

Abstract

This study investigates the spatiotemporal evolution of development discourse in low-resource settings. Analyzing more than two years of geotagged X data from Zambia, we introduce a mixed-methods pipeline utilizing topic modeling, change-point detection, and qualitative coding to identify critical shifts in public debate. We identify seven recurring themes, including public health challenges and frustration with government policy, shaped by regional events and national interventions. Notably, we detect discourse changepoints linked to the COVID19 pandemic and a geothermal project, illustrating how online conversations mirror policy flashpoints. Our analysis distinguishes between the ephemeral nature of acute crises like COVID19 and the persistent, structural reorientations driven by long-term infrastructure projects. We conceptualize "durable discourse" as sustained narrative engagement with development issues. Contributing to HCI and ICTD, we examine technology's socioeconomic impact, providing practical implications and future work for direct local engagement.

Figures (5)

• Figure 1: PELT-based change-point detection of the public health topic in Lusaka, Zambia, highlighting a sharp shift from general health discussions (e.g., cholera, multistakeholder planning) to COVID-19–specific concerns (e.g., testing, masking) around mid-June 2020. The region between the two vertical dotted lines marks the peak discourse period, illustrating how an acute health emergency redirected attention from long-term priorities to immediate crisis management.
• Figure 2: PELT-based change-point detection of the government policy topic in Zambia’s Southern Province, showing a significant shift on March 29 2020. Earlier conversations centered on tourism and business protection, while later discourse emphasized resources, decision-making, and infrastructure in response to the Kalahari Energy–funded geothermal project. This illustrates how new development interventions can reorient local policy dialogue and public sentiment.
• Figure 3: The Policy Salience and Intervention Loop illustrates an iterative workflow connecting neural network-based topic modeling, interactive refinement, visual evaluation (including change-point detection), and policymaker action for continuous development monitoring.
• Figure 4: The Dual-Mode WhatsApp Sensing Framework. This hierarchical tree structure facilitates efficient information dissemination from policymakers to local communities (downward flow) while enabling bidirectional engagement through passive sensing (topic modeling of donated messages) and active sensing (deployment of polls and surveys)
• Figure A.1: Coherence scores (C_V) across 18 BERTopic model variants with varying HDBSCAN minimum cluster size parameters. Each point represents a unique model configuration, with the number of extracted topics on the x-axis and the corresponding coherence score on the y-axis. The top-performing model (highlighted in red) achieved a coherence score of 0.720 with a minimum cluster size of 115, yielding 103 topics.