TEUM: Team Effect-aware Uplift Modeling for Online Games

Uplift modeling is an essential technique for providing accurate and unbiased estimates of a treatment's causal effect on individuals by modeling the Individual Treatment Effect (ITE). In the gaming industry, these techniques are frequently used to evaluate the effectiveness of various engagement incentives, such as equipment resources or daily rewards, aimed at enhancing player retention. However, existing uplift modeling methods have largely overlooked the team effects arising from teammates' treatment propensity (i.e., the likelihood of teammates receiving the treatment) and their influence on individual players' responses to the treatment. Such team effects can heavily influence players' responses to external treatments in team-based multiplayer games, e.g., in-game resources attained by a player can be shared with the teammates, thus affecting the teammates' response and vice versa. To tackle this issue, we propose a Team Effect-aware Uplift Modeling framework (TEUM) that incorporates the awareness of causal team effects into the ITE modeling process. TEUM consists of three key components: 1) a team effect awareness module that integrates the features and the estimated treatment propensity of the teammates into an individual player's embedding, thereby embedding team effect awareness into the uplift estimation process; 2) a specialized uplift forecasting network guided by a decoupled team effect objective function, which explicitly models and disentangles team causal effects from the overall causal effect space; 3) a GAN-based knowledge distillation module that extracts nuanced team effect patterns from our trained model, removing the dependency on instance-specific team information during inference and ensuring a more flexible and efficient deployment process. We conduct extensive offline experiments on two real-world online multiplayer gaming datasets, demonstrating the superior performance of TEUM compared to the state-of-the-art methods. In addition, TEUM has been deployed in the automatic incentive distribution system of a large online FPS game, resulting in a notable improvement.