Abstract
Social network services facilitate relationship building, knowledge sharing, and social management, offering significant business and social value. Effective research in this domain utilizes modeling and optimization strategies. This paper surveys the application of game theory to social networks, categorizing research into four key areas: information diffusion, behavior analysis, community detection, and information security. We provide a structured overview of these applications and discuss current limitations and future research directions.
Introduction
Traditional social network analysis (SNA) often relies on static graph theory, which excels at mapping topological connections but struggles to capture the underlying motivations of human behavior. Social networks are inherently dynamic; users are not merely passive nodes but strategic agents who make choices about sharing information, joining groups, and protecting their privacy.
Game theory—the mathematical study of strategic interaction—fills this gap by providing a framework to model these interactions. By treating users as "selfish" agents who maximize their own utility, researchers can predict network-wide phenomena. The challenge lies in selecting the appropriate game model (e.g., cooperative vs. non-cooperative, static vs. dynamic) to fit specific social behaviors like viral marketing or cyber-violence.
Taxonomy of Research Fields
The application of game theory in social networks is structured around four functional pillars, as illustrated in the taxonomy below.
This classification allows researchers to apply specific mathematical tools—such as Nash Equilibrium for stability or Shapley values for centrality—to distinct social network problems.
Key Application Scenarios
Information Diffusion
Information diffusion models how messages or innovations spread based on "payoffs" for sharing. These payoffs can include social capital or "limbic rise cost." In these models, users "adopt" a message only if it increases the message's valuation for themselves or their neighbors.
As shown in Figure 2, the adoption process is modeled as a strategic choice where the value of a message is modified by individual player contributions, influencing the decision of connected neighbors to adopt or ignore the content.
Community Detection
Community formation is viewed as a dynamic game where nodes choose to join or leave groups to maximize individual utility. This approach handles overlapping communities more naturally than traditional clustering by focusing on the incentives of individual nodes to remain within a group.
Behavior Analysis
This area identifies influential or "covert" nodes and models social issues like cyber-violence. Evolutionary game theory is particularly effective here, explaining how behaviors like "cooperate" or "defect" (from the classic Prisoner's Dilemma) scale within a network. For instance, research using a cyber-violence dataset of over 60,000 posts has modeled how abusive behavior propagates through strategic interactions.
Information Security
Security research analyzes the strategic tug-of-war between users, attackers, and data controllers. Game theory is used to develop access control and privacy strategies, particularly in multi-controller scenarios where different users (e.g., friends tagged in a photo) have conflicting privacy preferences.
Key Insights and Results
The "selfish node" assumption is the most critical insight, allowing researchers to predict macro-level shifts by calculating individual incentives. Key quantitative successes include:
- Centrality Accuracy: Shapley values in cooperative games provide a more robust method for identifying crucial nodes compared to traditional degree centrality.
- Covert Node Identification: Payoff matrices combined with Bayes' theorem have successfully identified "infected" or "covert" nodes in complex networks.
- Conflict Resolution: Game-theoretic adjustment algorithms can resolve privacy conflicts among multiple data owners by finding a point that maximizes collective profit while minimizing untrusted access.
Limitations and Future Work
Current models face several limitations:
- The "Human Factor": Most models assume perfect rationality, neglecting "irrational" factors like emotion and psychological vulnerability.
- Static Constraints: Many current models are restricted to "pure strategies." Real-world networks require more complex, multi-strategy dynamic games.
- Cross-Platform Analysis: Most research is siloed within single platforms. Future work should explore how games play out across multiple, interconnected social software systems simultaneously.
Summary updated