Search papers, labs, and topics across Lattice.
This paper introduces the Discount-Aware Network (DANet), which addresses the challenges of data sparsity, sample selection bias, and delayed feedback in post-click conversion rate (CVR) prediction for e-commerce recommendation systems. By integrating a time-frequency transformation module, a distribution de-bias module, and a supervised regression auxiliary task, DANet effectively captures the influence of item discount rates on CVR. Experimental results show that DANet outperforms existing methods, achieving a 1.61% improvement in offline AUC and significant gains in online metrics, demonstrating its practical effectiveness in real-world applications on Alibaba Tmall APP.
A novel network that leverages discount awareness boosts conversion rate predictions by over 3% in live e-commerce environments.
Post-click conversion rate (CVR) is a crucial element in online recommendation systems, which addresses significant challenges such as data sparsity (DS), sample selection bias (SSB), and delayed feedback. However, the impact of item discount rate-a key factor influencing both pricing and user purchasing behavior, has received limited attention. In this paper, we introduce the Discount-Aware Network (DANet) to model the relationship between item discount rates and CVR. DANet comprises three main components: 1) a time-frequency transformation module that utilizes Fourier transform to derive the frequency spectrum and capture the long-term discount rate trends of items; 2) a distribution de-bias module designed to mitigate the biases in user-specific discount rates caused by various purchase combinations and promotional activities, as well as periodic deviations linked to different promotion periods on e-commerce platforms; and 3) a supervised regression auxiliary task that establishes the explicit item discount labels to enhance the model's performance in terms of value accuracy, facilitating an effective representation of item discount rates. Experimental results on real datasets demonstrate the superiority of DANet, with offline AUC improving by 1.61%, and online A/B test also shows that DANet achieves impressive gains of 3.63% on pCVR and 2.23% on GMV. DANet has been successfully deployed on Alibaba Tmall APP. The code is available at https://github.com/tangrc/DANet.
DAMO