Research on e-commerce user stratification and coupon intelligent placement optimisation based on multi-behavioural RFM and deep reinforcement learning

Abstract

With the ongoing accumulation of user behavioral data on e-commerce platforms, the precision of coupon allocation based on individual characteristics has emerged as a pivotal challenge for enhancing both resource allocation efficiency and marketing performance. Addressing the limitations of the traditional RFM model—particularly its reliance on single-dimensional user stratification and inability to capture complex interactive behaviors—this study introduces a novel customer segmentation approach that integrates a multi-behavioral RFM model with an enhanced self-organizing map (SOM) algorithm. This integrated method enables a more comprehensive identification of user value and facilitates fine-grained segmentation. Building on this stratification, an intelligent coupon delivery optimization framework powered by deep reinforcement learning (DRL) is further developed. In this framework, user stratification outcomes are embedded into the state space alongside real-time behavioral data, allowing dynamic learning of individual response patterns through a state-action-reward mechanism. Consequently, the system autonomously generates personalized and optimized coupon distribution strategies. Experimental results demonstrate that, under budgetary constraints, the proposed approach significantly increases the activity levels and repurchase frequencies of key users. Furthermore, it substantially outperforms static and random delivery strategies across critical metrics, including coupon conversion rate, delivery efficiency, and overall marketing effectiveness. These findings not only confirm the method’s flexibility and adaptability in practical e-commerce settings but also offer a viable pathway and robust theoretical support for advancing precision marketing and optimizing resource allocation.