Enhancing Generative Models with Hierarchical Architecture for Automated Product Advertising Content Generation

Khanh-Vinh Nguyen, Thanh-Tam Doan, Quang-Thuy Ha, Mai-Vu Tran

Abstract: Automated product advertising content generation aims to create compelling and persuasive ads to attract users. While various approaches exist to generate ad content, we adopt a summarization-based approach to address it. Despite the significant success of pre-trained models in the text summarization field, their input length limitations present significant challenges. The restriction potentially distorts advertisement outputs and omits crucial information. In this paper, we proposed a hierarchical architecture that harnesses generative models to produce concise advertising content that retains essential product details. Our design features a two-stage summarization process: the initial stage generates important information from the original inputs, and the second stage refines them to produce the final comprehensive summary. Both stages utilize the Vietnamese generative models BARTpho and ViT5. Additionally, we developed a dataset of Vietnamese smartphone product advertisements to assess the effectiveness of our proposed architecture. Experiments on real-world datasets demonstrate that our approach outperforms existing pre-trained models, underscoring its potential to enhance summary quality.

External IDs:dblp:conf/kse/NguyenDHT24