2024阿里妈妈技术年刊

Investment Rating - The report does not explicitly provide an investment rating for the industry or company Core Insights - The report highlights the introduction of AIGB (AI-Generated Bidding) as a new paradigm for automated bidding in online advertising, which has been successfully implemented on the Alibaba advertising platform, achieving significant results [5][49] - The online advertising market reached a size of $626.8 billion in 2023, with automated bidding being a key driver for its growth [7][49] - AIGB utilizes generative models to optimize bidding strategies, avoiding the errors associated with traditional reinforcement learning methods, particularly in long-sequence decision-making scenarios [9][48] - The report discusses the successful deployment of the DiffBid model, which has shown improvements in GMV (Gross Merchandise Value) by 3.6% and ROI by 5.0% in various scenarios [13][49] - The AuctionNet benchmark has been developed to facilitate research in decision-making within large-scale auction environments, providing a comprehensive dataset and evaluation framework [21][46] Summary by Sections Decision Intelligence - AIGB was introduced as a new iterative paradigm for automated bidding, marking the first application of generative AI in this field [5] - The report outlines the successful deployment of AIGB on Alibaba's advertising platform, leading to significant performance improvements [49] - The DiffBid model, part of the AIGB framework, has been validated through multiple online experiments, demonstrating its effectiveness in optimizing bidding strategies [13][17] Benchmark Development - AuctionNet serves as a benchmark for decision-making in large-scale auctions, providing a realistic environment and a dataset of over 500 million records [21][41] - The benchmark has been utilized in the NeurIPS 2024 competition, allowing participants to evaluate their algorithms in a controlled setting [43][46] Generative Model Application - The report emphasizes the advantages of using generative models for automated bidding, particularly in capturing the relationships between historical bidding data and optimization objectives [9][48] - AIGB's framework allows for the generation of bidding trajectories that maximize specified goals while adhering to constraints [49][50]