<![CDATA[This study compares three similarity-based edge construction methods for village welfare networks from household welfare attributes in Data Desa Presisi. Households are represented as nodes, while edges denote computational similarity between household welfare profiles rather than social relationships. Five Village Welfare Index dimensions were discretized and transformed into binary one-hot representations. Pairwise similarities were calculated using Cosine Similarity, Jaccard Index, and Pearson Correlation, followed by automatic thresholding to retain strong relationships. Network evaluation focused on the Largest Connected Component, while community structure was assessed using Louvain modularity across 14 villages. All methods produced analyzable networks, achieving a 100% success rate and 97.17% average node coverage. Jaccard achieved the highest mean modularity (0.5074) and win rate (71.43%; 10 of 14 villages). A Friedman test confirmed differences among methods (χ² = 8.71, p = 0.013). Holm-corrected Wilcoxon tests showed that Jaccard significantly outperformed Cosine and Pearson, whereas Cosine and Pearson did not differ significantly. These findings indicate that attribute-overlap-based edge construction provides most consistent representation of household welfare-profile proximity under the tested binary representation and automatic-thresholding scheme, while emphasizing that this advantage is context-dependent rather than universal.]]>
Copyrights © 2026
