LLM-HYPER: Generative CTR Modeling for Cold-Start Ad Personalization
Summary: arXiv:2604.12096v1 Announce Type: new
Abstract: On online advertising platforms, newly introduced promotional ads face the cold-start problem, as they lack sufficient user feedback for model training. In this work, we propose LLM-HYPER, a novel framework that treats large language models (LLMs) as hypernetworks to directly generate the parameters of the click-through rate (CTR) estimator in a training-free manner.
Introduction
The cold-start problem in online advertising has been a significant hurdle for marketers and advertisers, particularly when launching new promotional campaigns. These campaigns often lack sufficient historical data to predict user engagement effectively. LLM-HYPER aims to address this challenge by leveraging advancements in natural language processing and machine learning.
Key Features of LLM-HYPER
- Hypernetwork Framework: LLM-HYPER utilizes large language models to serve as hypernetworks that generate parameters for the CTR estimator without the need for extensive training.
- Few-shot Chain-of-Thought Prompting: This innovative approach enables the model to infer feature-wise weights for a linear CTR predictor using minimal examples of multimodal ad content, including both text and images.
- Semantic Similarity Retrieval: By using CLIP embeddings, LLM-HYPER can retrieve semantically similar past campaigns, providing context that aids in understanding customer intent and content relevance.
- Normalization and Calibration: To maintain numerical stability, LLM-HYPER incorporates techniques that align generated weights with production-ready CTR distributions, ensuring that the output is both stable and applicable in real-world scenarios.
Performance and Results
Extensive offline experiments demonstrate that LLM-HYPER significantly outperforms traditional cold-start baselines, achieving an impressive increase of 55.9% in NDCG@10. Additionally, a real-world online A/B test conducted on one of the top e-commerce platforms in the U.S. highlighted the robustness of LLM-HYPER, showcasing its ability to drastically reduce the cold-start period while delivering competitive performance.
Conclusion
LLM-HYPER represents a substantial advancement in the realm of ad personalization, particularly for cold-start situations. Its successful deployment in production environments marks a critical step forward for advertisers seeking to optimize their campaigns in real-time. As online advertising continues to evolve, frameworks like LLM-HYPER will play a pivotal role in shaping future strategies.
For further details, refer to the original paper on arXiv.