Dynamics and Uncertainty in Networks and Web Data

Abstract

This thesis demonstrates how digital network analysis can illuminate complex regional economic transformations while advancing the methodological toolkit for large-scale, noisy web data and network data. Bridging these topics enables novel empirical economic research that is both data-rich and methodologically rigorous, with direct implications for understanding regional innovation and transition dynamics.
Firstly, this thesis uses Unfolded Adjacency Spectral Embedding, a novel method that enables comparisons of network embeddings across both space and time, to analyse directed, weighted yearly networks of commercial website hyperlink connections between geographic areas. A robust and scalable statistical testing procedure is developed for detecting changes between communities in dynamic networks where changes
are expressed in terms of known covariates.

Secondly, I address a fundamental challenge in network analysis: how to quantify uncertainty from a single observed network. I propose a distribution-free bootstrap method based on k-nearest neighbour smoothing, which produces bootstrapped embeddings that pass rigorous exchangeability tests. This methodological contribution provides tools for more reliable network analysis and visualisations.

Thirdly, the analysis of digital economic change is scaled beyond existing studies through the use of web archive data and AI-enabled analysis incorporating LLMs. A novel dataset to examine the twin transition (simultaneous digital and green economic transformation) spanning an average of 220,959 UK websites annually from 2014-2024 is introduced, alongside analysis that challenges assumptions about digital-led
transitions and highlights the fragmented nature of regional policy approaches.

Together, these contributions demonstrate how combining methodological innovation in network statistics with AI-driven web data analysis can yield new empirical insights into the evolving digital economy and the geography of technological change.

Date of Award	17 Mar 2026
Original language	English
Awarding Institution	University of Bristol
Supervisor	Oliver T Johnson (Supervisor), Emmanouil Tranos (Supervisor) & Daniel John Lawson (Supervisor)