Contextual Bandits Optimize Word-of-Mouth Marketing in Social Networks

Stimulated word-of-mouth (WOM) campaigns aim to encourage information sharing, often through prompts or incentives. A key challenge in optimizing these campaigns within social networks is accurately identifying and targeting users who are not only receptive themselves but also likely to generate "spillover" – where their recommendations influence their connected users. The effectiveness of this spillover varies significantly among individuals and their connections, creating a complex heterogeneity. To address this, researchers have developed a novel contextual multi-armed bandit framework. This framework is designed to learn the unique spillover probabilities for individual users. By understanding these probabilities, the system can then rank connected users to precisely target those who will maximize the overall rewards from stimulated WOM. Experiments conducted on real-world network datasets demonstrate the framework's superior performance. It significantly enhances the precision of targeting the most influential connected users, leading to increased rewards compared to traditional methods that do not account for individual spillover effects.