Know in AdVance: Linear-Complexity Forecasting of Ad Campaign Performance with Evolving User Interest

TL;DR

AdVance addresses Campaign Performance Forecasting in Real-time Bidding by integrating a time-aware, three-layer approach: a Transformer-based local auction encoder to produce per-auction embeddings, a fatigue-aware user-interest model to capture evolving preferences, and a conditional State Space Model that summarizes long sequences with linear complexity. By coupling auction-level representations with a global campaign summary, AdVance delivers accurate, horizon-robust cost and yield predictions while maintaining scalability for industrial deployment. Empirical results on Tencent data show superior forecasting accuracy versus state-of-the-art baselines, significant ablations confirming the importance of modeling both clicked and displayed items, and an online A/B uplift of 4.5% in ARPU. The framework's real-world impact is underscored by its deployment on the Tencent Advertising platform and its successful handling of irregular time intervals and long contexts, enabling more informed budgeting and bidding strategies.

Abstract

Real-time Bidding (RTB) advertisers wish to \textit{know in advance} the expected cost and yield of ad campaigns to avoid trial-and-error expenses. However, Campaign Performance Forecasting (CPF), a sequence modeling task involving tens of thousands of ad auctions, poses challenges of evolving user interest, auction representation, and long context, making coarse-grained and static-modeling methods sub-optimal. We propose \textit{AdVance}, a time-aware framework that integrates local auction-level and global campaign-level modeling. User preference and fatigue are disentangled using a time-positioned sequence of clicked items and a concise vector of all displayed items. Cross-attention, conditioned on the fatigue vector, captures the dynamics of user interest toward each candidate ad. Bidders compete with each other, presenting a complete graph similar to the self-attention mechanism. Hence, we employ a Transformer Encoder to compress each auction into embedding by solving auxiliary tasks. These sequential embeddings are then summarized by a conditional state space model (SSM) to comprehend long-range dependencies while maintaining global linear complexity. Considering the irregular time intervals between auctions, we make SSM's parameters dependent on the current auction embedding and the time interval. We further condition SSM's global predictions on the accumulation of local results. Extensive evaluations and ablation studies demonstrate its superiority over state-of-the-art methods. AdVance has been deployed on the Tencent Advertising platform, and A/B tests show a remarkable 4.5\% uplift in Average Revenue per User (ARPU).

Figures (4)

• Figure 1: AdVance disentangles user interests as time-stamped click sequences representing user preference and fatigue vectors compressing all displayed items (Sec. \ref{['user_interest']}). The attention mechanism compresses auctions into dense embeddings, and AdVance accumulates auction-level performance (Sec. \ref{['local_auction']}). A global SSM recurrently summarizes all embeddings, and AdVance returns final results based on the summary and accumulated performance (Sec. \ref{['global_SSM']}). During training, a causal mask blocks out "future" records after the current time-stamp (Sec. \ref{['offline_training']}).
• Figure 2: The AUC of three baselines and AdVance on five campaigns from various industries.
• Figure A-1: Real-time bidding workflow and the funnel-shaped structure.
• Figure A-2: The CTR trends vary with the number of exposures to ads of the same category. We present three categories, all showing a decline when over-exposed. We normalize the data for business privacy.