Beyond Pairwise Network Interactions: Implications for Information Centrality

• 成果类型: Article
• 署名作者: Lera, Sandro Claudio; Leng, Yan
• 署名单位: Southern University of Science & Technology; Southern University of Science & Technology; University of Texas System; University of Texas Austin
• 刊物名称: INFORMATION SYSTEMS RESEARCH
• ISSN/ISSBN: 1047-7047; 1526-5536
• DOI: 10.1287/isre.2024.1097
• 发表日期: 2026-06
• 关键词: hypergraph; centrality; Information diffusion; group interactions; network science; Investor sentiment; recommendation networks; empirical-analysis; design science; attention; TECHNOLOGY; diversity; KNOWLEDGE; models; IMPACT

摘要

Many sociotechnical systems feature group interactions that cannot be accurately represented by traditional network models because these reduce all interactions to pairwise links. As a result, classical network centrality measures often misidentify key actors in domains where diffusion occurs through higher-order group-based processes. To address this gap, we develop a hypergraph-based centrality framework that models diffusion using theory-informed random-walks. This approach incorporates domain-specific rules for the selection of nodes and edges, allowing the random-walk to align with the way group interactions unfold in real-world settings. We demonstrate the versatility of this hypergraph-based, theory-informed information centrality across three applications: opensource software development, where hyperedges link projects via shared contributors; high school social networks, where hyperedges represent observed group interactions among students; and financial news propagation, where articles form hyperedges connecting comentioned firms. In all three domains, our centrality measures outperform conventional graph-based metrics in predicting global outcomes of interest. Three insights emerge from our analysis. First, local diffusion mechanisms substantially influence global node rankings, underscoring the need for careful modeling of interaction rules. Second, centralities built on theory-informed hypergraph walks better explain collective outcomes than traditional graph-based benchmarks. Third, combining multiple centralities grounded in different local mechanisms improves predictive performance, highlighting the complementary contributions of distinct local interaction mechanisms. Together, these findings demonstrate the importance of directly modeling group interactions and provide a flexible theory-driven alternative to standard centrality measures.