MOEA for Viral Marketing: Balancing Revenue and Costs via Agent-Based Influence Maximization

TL;DR

This research shifts the "Influence Maximization" (IM) paradigm from a pure graph-theory problem to a realistic business optimization task. By combining Evolutionary Multiobjective Optimization (EMO) with Agent-Based Modeling (ABM), the authors demonstrate how to target social media influentials to maximize sales revenue while strictly controlling the costs of free product samples.

Problem & Motivation: Beyond "Maximum Spread"

In the classic IM problem, the goal is simple: find $k$ nodes to maximize the spread of information. However, real-world marketing faces two harsh realities:

1. Budget Constraints: Targeting "hubs" (highly connected users) isn't free; it often involves discounts or free samples.
2. Human Complexity: Real consumers aren't just "active/inactive" nodes. They have preferences, they get bored (awareness deactivation), and they experience uncertainty.

Existing SOTA methods often treat the number of seeds as a fixed input and ignore the cost-revenue trade-off. This paper argues that the "best" marketing strategy is actually a Pareto front of solutions, allowing a brand to choose between a "high-risk, high-reward" aggressive campaign or a "low-cost, conservative" one.

Methodology: The Core Engine

The framework stands on two legs: Evolutionary Search and Simulated Market Dynamics.

1. The Decision Space

Instead of selecting specific node IDs, the algorithm optimizes the criteria for what makes an influential. It evolves weights ($\omega$) for local network metrics:

• Degree: Direct reach.
• 2-steps: Proximity to the "friend-of-a-friend" network.
• Clustering Coefficient: Local density, which affects the speed of "word-of-mouth."
• Seed Count ($s$): Dynamically determining how many people to target.

2. The Simulation (ABM)

The fitness of a strategy is evaluated through a 365-day simulation. Agents (consumers) use four heuristics to decide whether to buy: Repetition, Deliberation, Imitation, and Social Comparison.

Figure 1: Conceptual overview of the viral marketing optimization process.

Experiments & SOTA Comparison

The authors tested their approach on various topologies, from Scale-Free (SF) networks to a real-world APP network with 20,000 users.

• MOEA/D vs. NSGA-II: Both outperformed greedy baselines. MOEA/D showed a slight edge in networks with high diffusion potential (like Scale-Free networks), while NSGA-II was more robust in regular lattices where information travels slowly.
• The "Greedy" Failure: Traditional greedy algorithms, which pick nodes based on degree alone, were consistently dominated. They are too "eager" and fail to find the nuanced trade-off between targeting one global hub vs. several local influencers.

Figure 2: The Pareto front for a Scale-Free network, illustrating the trade-off between Revenue (NPV) and Costs.

Critical Insight: Who are the Influentials?

The study reveals a fascinating "topological dependency." In "Email-style" networks, the optimal strategy heavily weights Degree (finding hubs). In "Regular Lattice" networks (where everyone has similar local connections), the strategy shifts toward increasing the number of seeds ($s$) because a single node's reach is physically limited by the network's diameter.

Conclusion & Future Outlook

This work provides a professional blueprint for Decision Support Systems in marketing. While highly effective, the study highlights its own limitation: it treats seeding as a "one-shot" event at $t=0$. The next frontier is Sequential Seeding, where marketers adaptively release samples over time as they observe the cascade's progress.

For practitioners, the message is clear: Stop looking for the "single best" influencer. Start evolving a portfolio of targeting strategies that respect your budget.