IBC2024 Tech Papers: The Power of Data-Driven Insights: Advancing Ad Targeting With Generative AI

In 2017, Netflix demonstrated the impact of personalized promotional messaging by adapting artwork based on user preferences and viewing history. The company’s content experts generated multiple images for every title and change them regularly to lure audiences based on their previous viewing history. The approach utilized a sophisticated “online reinforcement learning” strategy, optimizing the balance between exploiting known user preferences and exploring new data for improved recommendations. This dynamic methodology was essential in minimizing the cumulative “regret” (defined as the difference between the expected “payoff” (e.g. engagement) of the algorithm and the payoff of a single fixed strategy for selecting artworks) and enhancing viewer satisfaction over time.

Despite its efficiency, this approach epitomized a few key principles under the legacy Personalization 1.0 paradigm, namely the need to choose between a finite set of creatives (while being limited by multimodal content creation’s cost and complexity), the reliance on high-quality (and therefore expensive) human preference labels to optimize algorithmic tuning, and the use of shallow context information (e.g. transient signals like search history) as proxies for viewers’ interests. Today’s advancements in Generative AI necessitate a complete re evaluation of the algorithmic / architectural trade-off and its relevance.

This paper outlines...