A Hybrid Deep Recommendation System for Issuing Digital Coupons in Real-Time

Abstract

In the field of marketing, offering discount coupons to encourage purchases is one of the most frequently used strategies. In particular, with the rise of online shopping malls surpassing offline ones, digital coupons have become increasingly prevalent. The key challenge in issuing coupons is identifying customers with a high risk of churn and providing them with coupons of the right category at an optimal discount rate. Manually addressing this task is impractical because online shopping platforms serve millions of users and offer a wide variety of coupon types. Accordingly, this study proposes a time series deep learning framework that identifies customers at high risk of churn in real time and provides them with personalized coupons for relevant product categories at appropriate discount rates. To address the data class imbalance problem, additional dependent variables were introduced, and static customer meta-information was combined with real-time page click features to enhance accuracy. In addition, we tested the improvement in financial profits of an actual shopping mall. After various ablation model experiments, the results show the superiority of our proposed framework, which has brought great profits to the platform. The proposed framework achieved an AUC of 0.747 for purchase intention prediction, improved coupon conversion rates by up to 32.65%, and increased the relative profit index from 100.0 to 103.15 in real-world deployment experiments.