Communicative MARL-based Relevance Discerning Network for Repetition-Aware Recommendation

Abstract

The repeated user-item interaction now is becoming a common phenomenon in the e-commerce scenario. Due to its potential economic profit, various models are emerging to predict which item will be re-interacted based on the user-item interactions. In this specific scenario, item relevance is a critical factor that needs to be concerned, which tends to have different effects on the succeeding re-interacted one (i.e., stimulating or delaying its emergence). It is necessary to make a detailed discernment of item relevance for a better repetition-aware recommendation. Unfortunately, existing works usually mixed all these types, which may disturb the learning process and result in poor performance. In this paper, we introduce a novel Communicative MARL-based Relevance Discerning Network (CARDfor short) to automatically discern the item relevance for a better repetition-aware recommendation. Specifically, CARDformalizes the item relevance discerning problem into a communication selection process in MARL. CARDtreats each unique interacted item as an agent and defines three different communication types over agents, which are stimulative, inhibitive, and noisy respectively. After this, CARDutilizes a Gumbel-enhanced classifier to distinguish the communication types among agents, and an attention-based Reactive Point Process is further designed to transmit the well-discerned stimulative and inhibitive incentives separately among all agents to make an effective collaboration for repetition decisions. Experimental results on two real-world e-commerce datasets show that our proposed method outperforms the state-of-the-art recommendation methods in terms of both sequential and repetition-aware recommenders. Furthermore, CARDis also deployed in the online sponsored search advertising system in Meituan, obtaining a performance improvement of over 1.5% and 1.2% in CTR and effective Cost Per Mille (eCPM) respectively, which is significant to the business.