AmpSum: Adaptive Multiple-Product Summarization towards Improving Recommendation Captions

Tuan Quoc Truong, Tong Zhao, Changhe Yuan

Abstract:

In e-commerce websites, multiple related product recommenda- tions are usually organized into “widgets”, each given a name, as a recommendation caption, to describe the products within. These recommendation captions are usually manually crafted and generic in nature, making it difficult to attach meaningful and informative names at scale. As a result, the captions are inadequate in help- ing customers to better understand the connection between the multiple recommendations and make faster product discovery.
We propose an Adaptive Multiple-Product Summarization frame- work (AmpSum) that automatically and adaptively generates widget captions based on different recommended products. The multiplic- ity of products to be summarized in a widget caption is particularly novel. The lack of well-developed labels motivates us to design a weakly supervised learning approach with distant supervision to bootstrap the model learning from pseudo labels, and then fine-tune the model with a small amount of manual labels. To validate the ef- ficacy of this method, we conduct extensive experiments on several product categories of Amazon data. The results demonstrate that our proposed framework consistently outperforms state-of-the-art baselines over 9.47-29.14% on ROUGE and 27.31% on METEOR. With case studies, we illustrate how AmpSum could adaptively gen- erate summarization based on different product recommendations.