A mixed-methods matrix approach to explainable GNNs in migrant network capital

Abstract Social networks play a central role in facilitating the circulation of information, resources, trust, and opportunities during cross-border migration processes. Although explainable artificial intelligence (XAI) approaches address concerns regarding privacy, surveillance, and algorithmic fairness in graph neural networks (GNNs), they have rarely been applied in migration research, and their relevance to migration studies remains largely underexplored. The purpose of this study was to examine how an XAI–GNN model reinforced network capital (NC) among Laotian migrant workers in Bangkok, Thailand. A mixed-methods matrix design was employed, consisting of a qualitative phase followed by a quantitative phase. Purposive sampling was employed for 15 participants, whereas stratified sampling was used to select 280 Laotian migrant workers from 50 districts in Bangkok. A qualitative matrix was developed through qualitative network analysis, whereas a quantitative matrix was constructed using GNNs and a local linear regression model. Findings from the QNA of 15 participants highlighted the roles of network diversity, tie strength, reciprocity, and bridging capital among Laotian migrants in Bangkok, Thailand . The quantitative matrix strand, conducted on 280 samples, indicated an XAI intercept of 0.112 (∅₀ = 0.098), while NNExplainer, PGExplainer, GraphLIME, SHAP, and Grad-CAM/Saliency produced scores of 0.792, 0.767, 0.780, 0.791, and 0.821, respectively. The findings provide evidence supporting the effectiveness of XAI-driven network analysis using GNNs for identifying latent dimensions of migrant NC in host countries.