Parallel Ranking of Ads and Creatives in Real-Time Advertising Systems

Abstract: "Creativity is the heart and soul of advertising services". Effective creatives can create a win-win scenario: advertisers can reach target users and achieve marketing objectives more effectively, users can more quickly find products of interest, and platforms can generate more advertising revenue. With the advent of AI-Generated Content, advertisers now can produce vast amounts of creative content at a minimal cost. The current challenge lies in how advertising systems can select the most pertinent creative in real-time for each user personally. Existing methods typically perform serial ranking of ads or creatives, limiting the creative module in terms of both effectiveness and efficiency. In this paper, we propose for the first time a novel architecture for online parallel estimation of ads and creatives ranking, as well as the corresponding offline joint optimization model. The online architecture enables sophisticated personalized creative modeling while reducing overall latency. The offline joint model for CTR estimation allows mutual awareness and collaborative optimization between ads and creatives. Additionally, we optimize the offline evaluation metrics for the implicit feedback sorting task involved in ad creative ranking. We conduct extensive experiments to compare ours with two state-of-the-art approaches. The results demonstrate the effectiveness of our approach in both offline evaluations and real-world advertising platforms online in terms of response time, CTR, and CPM.

• The paper proposes the Peri-CR architecture that decouples ad and creative ranking, reducing latency while enhancing system efficiency.
• The joint optimization model JAC and the novel NSCTR metric enable more accurate CTR estimation and reliable offline evaluation.
• Experimental results demonstrate a 1.58% CTR improvement over traditional methods, leading to higher revenue and performance.

Parallel Ranking of Ads and Creatives in Real-Time Advertising Systems

Introduction

Online advertising has become a crucial revenue stream for e-commerce platforms by leveraging personalization based on user preferences. Creatives, vital elements in advertisement strategies, facilitate the connection between potential buyers and products through diverse formats like images and videos. With the progress in AI-Generated Content (AIGC), advertisers can now produce numerous creative variations at minimal costs. This paper introduces an innovative architecture for parallel ranking of ads and creatives within real-time advertising systems, addressing the inefficiencies of traditional serial ranking methods.

Methodology

Parallel Architecture for Ranking

The proposed architecture, termed "Peri-CR," decouples the creative ranking module from the conventional ad retrieval and ranking process. Unlike traditional "Post-CR" and "Pre-CR" methods, where ad and creative rankings are performed sequentially, Peri-CR conducts these processes in parallel. This separation reduces the system's end-to-end latency by allowing more sophisticated models for each component without the time constraints imposed by previous architectures.

Joint Model for Offline Optimization

An offline joint optimization model called JAC (Joint Ad-Creative) facilitates mutual awareness between ads and creatives during CTR estimation. The offline model integrates both ad and creative features, enabling collaborative optimization. The architecture is designed for scalability, dividing into two parallel models for online application. This two-part model supports efficient and accurate creative bias estimation by leveraging historical statistical data and creative attributes.

Novel Evaluation Metrics

The paper addresses challenges in evaluating creative ranking offline. Traditional metrics like AUC are less effective for creative performance analysis, leading to the introduction of Normalized Simulated CTR (NSCTR). NSCTR measures creative ranking quality while maintaining sample distribution fidelity, offering a more reliable offline metric compared to sCTR, particularly in correlating with online performance outcomes.

Experimental Results

The proposed Peri-CR architecture demonstrates superiority in both offline and online settings. Extensive experiments reveal the following:

• Response Time: Peri-CR exhibits negligible computational overhead compared to no-creative ranking baselines, maintaining efficient response times.
• CTR Improvement: The architecture achieves a higher CTR and Revenue Per Mille (RPM) than both Pre-CR and Post-CR methods. For example, Peri-CR enhances CTR by 1.58% over Pre-CR.
• System Efficiency: The Peri-CR approach successfully reduces latency, an outcome not easily achieved with sequential creative and ad ranking methods.

Implications and Future Directions

The implications of this research extend to improving the architectural and operational efficiency of online advertising systems. By decoupling creative and ad ranking, platforms can employ complex models without compromising system speed. Future research could explore real-time creative generation integrated with ranking systems, potentially enhancing the personalization and responsiveness of advertising platforms.

Conclusion

This paper presents a novel parallel ranking architecture for ads and creatives, showing clear advantages over traditional sequential methods in terms of efficiency and efficacy. The Peri-CR approach, supported by a joint optimization model, reduces latency while enhancing CTR prediction accuracy. Through comprehensive evaluations, the proposed methodology proves superior in real-world applications, offering significant contributions to the field of online advertising systems.