The design of consumer behavior prediction and optimization model by integrating DQN and LSTM.

Abstract

Amidst the rapid evolution of e-commerce and the growing abundance of consumer shopping data, accurately identifying consumer interests and enabling targeted outreach has become a critical focus for merchants and researchers. This study introduces the RL-Trans framework, an innovative approach integrating Deep Reinforcement Learning (DQN) with Transformer to capture and analyze consumer interests intelligently. By leveraging consumer profiles and transactional records, the RL-Trans framework dynamically adapts to evolving consumer behavior, enabling personalized interest propagation. The framework initially employs a Transformer network to process consumer behavioral data using a multi-headed attention mechanism, It then integrates DQN to optimize the model culminating in an enhanced prediction layer that refines consumer interest analysis. Experimental results demonstrate the framework's superior performance to conventional LSTM-based approaches, while achieving competitive efficacy relative to state-of-the-art methods. This study advances academic discourse by introducing a novel perspective and methodology for consumer behavior analysis. It provides theoretical foundations and practical insights for enhancing personalized services and marketing strategies in e-commerce.