Beyond the Expected: Mapping and Explaining Outlier Communities in Highly Homophilous Social Networks

Most studies, like Srivastav et al. (2024), confirm that homophily is a dominant organizing principle in social networks, leading to communities of similar individuals. But what about those unusual communities that don’t fit the mold—where members are notably dissimilar by interests, demographics, or beliefs? This project would build on existing community detection (Louvain, LPA) and develop algorithms to systematically “flag” outlier communities that violate expected homophily patterns (e.g., mixing across highly disparate MBTI types as seen in Ayyoubzadeh & Shahnazari, 2025). Qualitative follow-up (interviews or content analysis) would investigate why these outliers form—are they driven by shared external goals, bridging personalities, or algorithmic nudges? This flips the usual homophily lens, using outliers as windows into hidden network dynamics and sources of innovation or resilience.

References:

1. Analyzing Interest-Based Homophily in Online Social Networks Using Community Detection Methods. Manoj Kumar Srivastav, Somsubhra Gupta, Subhranil Som (2024). Educational Administration: Theory and Practice.
2. Can Invisible Psychological Traits Organize Visible Network Structure? A Complex Network Analysis of Myers-Briggs Type Indicator-Based Interaction Patterns in Anonymous Social Networks. Seyed Moein Ayyoubzadeh, Kourosh Shahnazari (2025). arXiv.org.