Report on Recent Developments in Online Advertising and Marketing Research

General Trends and Innovations

The field of online advertising and marketing is witnessing a significant shift towards more sophisticated and integrated models that cater to diverse stakeholders within the digital ecosystem. Recent advancements are characterized by the development of novel frameworks that leverage deep learning, graph neural networks, and advanced uplift modeling techniques to optimize advertising strategies, enhance user engagement, and maximize platform revenue.

One of the primary trends is the move towards joint advertising models that allow for collaborative bidding between different entities such as online store owners and brand suppliers. This approach not only addresses the dual needs of these stakeholders but also introduces new neural network architectures designed to optimize auction mechanisms, ensuring near dominant strategy incentive compatibility and individual rationality.

Another notable development is the application of end-to-end cost-effective incentive recommendation models under budget constraints. These models utilize uplift modeling to strategically assign incentives to individual customers, addressing the challenges of monotonic and smooth response curves and the optimality gap inherent in two-stage approaches. The integration of differentiable allocation modules and integer linear programming within these models demonstrates a significant advancement in optimizing budget allocation.

The modeling of reference-dependent choices using graph neural networks represents a breakthrough in understanding and quantifying consumer preferences. This approach integrates theoretical insights from Prospect Theory into data-driven frameworks, capturing complex interactions between interest-inspired and price-inspired preferences. The introduction of attribute-level willingness-to-pay measures further enhances the granularity and accuracy of preference quantification.

Multi-treatment multi-task uplift modeling is emerging as a powerful tool for enhancing user growth in online platforms. By estimating treatment effects in multi-task scenarios, this approach addresses the limitations of single-treatment settings and provides a more nuanced understanding of user responses to various treatments. The use of multi-gate mixture of experts networks and treatment-user feature interaction modules enhances the model's ability to capture inter-task relations and correlations.

Noteworthy Papers

• Joint Auction in the Online Advertising Market: Introduces a novel joint advertising model and a neural network architecture for optimal auction design, significantly improving platform revenue.
• End-to-End Cost-Effective Incentive Recommendation under Budget Constraint with Uplift Modeling: Proposes a novel end-to-end model that integrates differentiable allocation and integer linear programming, outperforming existing two-stage approaches.
• Modeling Reference-dependent Choices with Graph Neural Networks: Develops a deep learning framework for quantifying reference-dependent preferences, introducing innovative measures like attribute-level willingness-to-pay.
• Multi-Treatment Multi-Task Uplift Modeling for Enhancing User Growth: Proposes a multi-task uplift network that captures complex treatment effects, demonstrating effectiveness in both offline and online settings.
• Enhancing Uplift Modeling in Multi-Treatment Marketing Campaigns: Leverages score ranking and calibration techniques to refine uplift predictions, providing actionable insights for campaign optimization.

These papers represent significant contributions to the field, advancing our understanding and capabilities in online advertising and marketing.