Abstract
Digital human recommendation system has been developed to help customers to find their favorite products and is playing an active role in various recommendation contexts. How to catch and learn the preferences of the customers at the right time and meet the exact requirements of the customer become crucial in the digital human recommendation. We design a novel practical digital human interactive recommendation agent framework based on reinforcement learning to improve the efficiency of interactive recommendation decision-making by leveraging both the digital human features and the superiority of reinforcement learning. The proposed framework learns through immediate interactions among the digital human and customers dynamically through stat-of-art reinforcement learning algorithms and embedding with multimodal and graph embedding to improve the accuracy of the personalization and thus enable the digital human agent to actively catch the attention of a customer timely. Experiments on real business data show that this framework can provide better-personalized customer engagement and better customer experiences etc.
Abstract (translated)
URL
https://arxiv.org/abs/2210.10638