Year: 2025
Type: conference-paper
Abstract: In a sponsored search system, high-quality ad creatives (such as texts and images) that can attract users play a crucial role in improving user engagement. This paper focuses on designing approaches to automatically generate ad texts, and the goals are to 1) improve the diversity of text creatives and 2) ensure the relevance between text creatives and search queries/keywords, which are the two key factors affecting user behaviors. We propose a framework named Ideation, which generates diverse ad texts given search queries/keywords and ad landing page information with a latent diffusion based sequence-to-sequence model. Experiments on a public industrial dataset for benchmarking ad text generation show that Ideation significantly outperforms previous approaches and possesses the ability of generating fluent and diverse text creatives.
Author: Penghui Wei
Institution: Baidu (China)
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4411549726