Adversarial Graph Embeddings for Fair Influence Maximization over Social Networks (2005.04074v2)

Abstract: Influence maximization is a widely studied topic in network science, where the aim is to reach the maximum possible number of nodes, while only targeting a small initial set of individuals. It has critical applications in many fields, including viral marketing, information propagation, news dissemination, and vaccinations. However, the objective does not usually take into account whether the final set of influenced nodes is fair with respect to sensitive attributes, such as race or gender. Here we address fair influence maximization, aiming to reach minorities more equitably. We introduce Adversarial Graph Embeddings: we co-train an auto-encoder for graph embedding and a discriminator to discern sensitive attributes. This leads to embeddings which are similarly distributed across sensitive attributes. We then find a good initial set by clustering the embeddings. We believe we are the first to use embeddings for the task of fair influence maximization. While there are typically trade-offs between fairness and influence maximization objectives, our experiments on synthetic and real-world datasets show that our approach dramatically reduces disparity while remaining competitive with state-of-the-art influence maximization methods.

• The paper introduces a novel Adversarial Graph Embedding methodology for fair influence maximization over social networks.
• The approach uses adversarial training to learn node embeddings that minimize influence disparities across groups based on sensitive attributes.
• Experiments demonstrate the method maintains competitive influence while significantly reducing fairness disparities on real-world datasets.

Adversarial Graph Embeddings for Fair Influence Maximization

This paper tackles the challenge of influence maximization within social networks, aiming at equitable distribution of influential effects across diverse groups defined by sensitive attributes. The authors introduce a novel methodology employing Adversarial Graph Embeddings to achieve this goal, a first in leveraging embeddings for fair influence maximization.

Overview of Methodology

The paper's central contribution is an innovative embedding framework designed to minimize disparities in influence spread based on sensitive attributes such as race or gender. Traditional influence maximization techniques, often deterministic and reliant on network topology features, do not account for fairness, thereby potentially exacerbating existing societal biases.

By combining graph embeddings with adversarial training—a concept inspired by Generative Adversarial Networks (GANs)—the method generates node representations that are similarly distributed across different subgroups. An auto-encoder captures the network structure, while a discriminator seeks to identify sensitive attribute discrepancies in embeddings, pushing the auto-encoder to produce fair representations. The final selection of initial influential seeds is determined using cluster centroids derived from these adversarially trained embeddings, emphasizing demographic diversity.

Experimental Validation and Performance

The paper substantiates the proposed methodology through experiments on both synthetic and real-world datasets (e.g., Rice-Facebook dataset), comparing results against established benchmarks like the Greedy algorithm and Tsang et al.'s work on group fairness in influence maximization.

• Results: The method consistently demonstrates competitive influence maximization performance while substantially reducing disparity between demographic groups in terms of influenced node fractions. The synthetic dataset showed the majority of nodes, irrespective of their community size, influenced equally, highlighting the fairness of the method without sacrificing influence effectiveness.

Implications and Future Directions

The integration of fairness into influence maximization through adversarial embeddings not only advances network diffusion strategies but also enriches the methodological arsenal for addressing equity in algorithmic processes. The approach provides foundational concepts that can be extended to various social network applications, including fair clustering or node classification.

Looking forward, this paper suggests potential pathways for further refinement and application of adversarial embedding techniques, potentially exploring multi-attribute setups and dynamic network environments. As fairness in AI continues to gain prominence, such methodologies are pertinent across domains where equitable information distribution is crucial.

Overall, this work offers a significant step towards harmonizing influence maximization and fairness, paving the way for more socially aware algorithmic designs in network science.