One Size, Many Fits: Aligning Diverse Group-Wise Click Preferences in Large-Scale Advertising Image Generation

TL;DR

The paper tackles the limitation of one-size-fits-all CTR optimization in advertising image generation by introducing One Size, Many Fits (OSMF). It proposes Product-Aware Adaptive Grouping (PAAG) to form dynamic, product-conditioned user groups and Preference-Conditioned Image Generation (PCIG) that generates group-tailored images via a Group-aware Multimodal LLM (G-MLLM) and Group Direct Preference Optimization (Group-DPO). A new GAIP dataset with about 610k groups across 40M users enables large-scale evaluation of group-wise preferences. Empirical results show state-of-the-art performance in both offline and online settings, with PCIG achieving meaningful CTR gains and GRM improving reward modeling. The work includes releasing GAIP to foster research on group-aware advertising and demonstrates the practical impact of aligning generation with diverse user group preferences at scale.

Abstract

Advertising image generation has increasingly focused on online metrics like Click-Through Rate (CTR), yet existing approaches adopt a ``one-size-fits-all" strategy that optimizes for overall CTR while neglecting preference diversity among user groups. This leads to suboptimal performance for specific groups, limiting targeted marketing effectiveness. To bridge this gap, we present \textit{One Size, Many Fits} (OSMF), a unified framework that aligns diverse group-wise click preferences in large-scale advertising image generation. OSMF begins with product-aware adaptive grouping, which dynamically organizes users based on their attributes and product characteristics, representing each group with rich collective preference features. Building on these groups, preference-conditioned image generation employs a Group-aware Multimodal Large Language Model (G-MLLM) to generate tailored images for each group. The G-MLLM is pre-trained to simultaneously comprehend group features and generate advertising images. Subsequently, we fine-tune the G-MLLM using our proposed Group-DPO for group-wise preference alignment, which effectively enhances each group's CTR on the generated images. To further advance this field, we introduce the Grouped Advertising Image Preference Dataset (GAIP), the first large-scale public dataset of group-wise image preferences, including around 600K groups built from 40M users. Extensive experiments demonstrate that our framework achieves the state-of-the-art performance in both offline and online settings. Our code and datasets will be released at https://github.com/JD-GenX/OSMF.

Figures (7)

• Figure 1: Current methods use overall CTR as unified preference targets, failing to capture group-specific preferences. (a) In smartphone advertising, male and female users show distinct preferences; (b) In shoe advertising, young and old users exhibit distinct preferences. These image preferences are often conflicting in both cases.
• Figure 2: Overview of the OSMF framework: (a) Product-Aware Adaptive Grouping (PAAG) organizes users into preference-coherent groups; (b) Preference-Conditioned Image Generation (PCIG) tailors advertising images for each group using Group-DPO.
• Figure 3: Comparison with SOTA reward models.
• Figure 4: Qualitative results of OSMF. $\mathbf{G}_{*,k}$ denotes the $k$-th group feature for product "$*$". Group assignments are product-dependent, so rows correspond to different groups.
• Figure 5: More visualization examples generated by our method across diverse product categories.