Source-linked AI summary
Multirelational Organization of Large-scale Social Networks in an Online World
Michael Szell, Renaud Lambiotte, Stefan Thurner
TL;DR
Large-scale social-network studies often overlook the multiple relations connecting individuals, despite their importance for describing social organization. This paper analyzes six interaction networks among roughly 300,000 online-game players and finds that aggressive relations differ from non-aggressive ones, while multiplex patterns support structural balance theory.
Problem
Large-scale social-network studies have largely overlooked the multiple relations connecting individuals because of limited empirical data.
Method
The study constructs six directed interaction networks from longitudinal data covering nearly the entire population of an online game and analyzes their topology, overlap, correlations, and triads.
Results
Negative interactions show lower reciprocity, weaker clustering, and fatter-tailed degree distributions, while friendship–enmity triads empirically support the weak formulation of structural balance.
Takeaways & Limitations
Multiplex social networks exhibit distinct interaction-dependent organization, with individuals often occupying different roles across relational networks.
Takeaways & Limitations
Correlation measures require caution because they may be biased by time spent in the game and by ignoring link weights for messages or traded money.
Abstract
from arXiv · showhide
The capacity to collect fingerprints of individuals in online media has revolutionized the way researchers explore human society. Social systems can be seen as a non-linear superposition of a multitude of complex social networks, where nodes represent individuals and links capture a variety of different social relations. Much emphasis has been put on the network topology of social interactions, however, the multi-dimensional nature of these interactions has largely been ignored in empirical studies, mostly because of lack of data. Here, for the first time, we analyze a complete, multi-relational, large social network of a society consisting of the 300,000 odd players of a massive multiplayer online game. We extract networks of six different types of one-to-one interactions between the players. Three of them carry a positive connotation (friendship, communication, trade), three a negative (enmity, armed aggression, punishment). We first analyze these types of networks as separate entities and find that negative interactions differ from positive interactions by their lower reciprocity, weaker clustering and fatter-tail degree distribution. We then proceed to explore how the inter-dependence of different network types determines the organization of the social system. In particular we study correlations and overlap between different types of links and demonstrate the tendency of individuals to play different roles in different networks. As a demonstration of the power of the approach we present the first empirical large-scale verification of the long-standing structural balance theory, by focusing on the specific multiplex network of friendship and enmity relations.
Results
Across six interaction networks, positive ties are more reciprocal and clustered, whereas negative ties show lower reciprocity, weaker clustering, and power-law degree distributions. Cross-network overlap and correlations reveal distinct relational roles, while friendship–enmity triads support the weak form of structural balance.
- Single-network properties: Positive networks are strongly reciprocal, while enmity, attack, and bounty networks show significantly smaller reciprocity.The positive networks are friendship, private messages, and trade; lower reciprocity in enemy networks may partly reflect deliberate refusal of reciprocation.
- Single-network properties: Negative interactions have heavier-tailed degree patterns: power laws appear for aggression, enmity, and bounty, but not for positive or passive links.Power laws occur for attack out-degree, enmity in-degree, and both bounty in- and out-degree; they are absent for friendship, communication, trade, and being attacked.
- Single-network properties: Positive ties cluster more strongly than negative ties, suggesting that triadic closure is not dominant for negative interactions.The lower clustering of negative links is attributed to the balance of signed motifs.
- Inter-network correlations and overlap: Communication overlaps strongly with friendship and attack, while trade–enmity overlap vanishes and friends and enemies are never the same relation.Communication accompanies aggression; players who trade almost never become enemies, and friendship–enmity shows substantial degree correlation despite vanishing overlap.
- Inter-network correlations and overlap: Low cross-network degree correlations indicate that hubs in one network are not necessarily hubs in another, so agents play different relational roles.Correlation estimates require caution because player time and ignored link weights may bias them.
- Structural balance: Signed-triad frequencies support the weak formulation of structural balance: +++ and +−− triads are over-represented, ++− is under-represented, and −−− is less under-represented.The z-score comparisons are relative to pure chance, and the pattern favors the weak formulation over Heider’s original formulation.
Discussion
The work addresses a gap in large-scale social-network research by quantitatively measuring the multidimensionality of human relationships. It shows that aggression-driven relations differ systemically from non-aggressive relations, while different networks reveal distinct human roles.
- Motivation: The study argues that analyzing multiplexity is essential because multiple forces acting on social agents shape large-scale social systems.Many empirical studies focus on node properties while overlooking the multiple types of links connecting agents.
- Findings: Aggression-driven relations produce markedly different systemic characteristics from non-aggressive relations.This finding demonstrates macroscopic consequences of interaction type.
- Findings: Interactions between network types reveal non-trivial multidimensional structure, with humans playing very different roles across relational networks.The dataset enables quantitative exploration of this multi-dimensionality.
Materials and Methods
The study uses nearly complete, longitudinal interaction data from over 300,000 Pardus players to construct six directed social networks. It analyzes each network with standard structural measures and compares network overlap using Jaccard similarity.
- Social Network Data: The dataset records practically all actions of Pardus players since 2004, enabling an almost complete, longitudinal mapping of multiplex relations in an entire society.Because players were unaware their actions were logged, interviewer effects were absent and measurement errors were practically negligible.
- Network Construction: Friendship and enmity networks are snapshots from day 445, whereas the other networks aggregate any link occurring between days 1 and 445.Networks are represented as unweighted and directed; undirected links exist when at least one directional link is present.
- Network Measures: The six networks are analyzed separately using node and link counts, reciprocity, in–out-degree correlation, average degree, and clustering coefficients.Clustering is also normalized against the corresponding random graph.
- Network Interactions: Interactions between network types are quantified with the Jaccard coefficient, defined as the intersection of two link sets divided by their union.This measures the tendency for links to occur simultaneously in both networks.