Group Modeling and Recommendation Based on Multi-Behavior Interactions in Live Streaming E-Commerce

The live streaming e-commerce paradigm is experiencing a meteoric surge, distinguished by hosts dynamically showcasing products via live videos and fostering interactive engagement with their viewers. Although a streamer may not cater to every individual’s tastes, it is imperative that it resonates with the majority of viewers’ preferences. Nevertheless, a common challenge arises when the products featured in the live streams often fail to align with the collective interests of the viewers in the live rooms, leading to a significant portion of them disengaging from the broadcast due to a lack of interest in the streamed content. To address this issue, we introduce an Attention-based Group Recommendation for E-commerce Live Stream viewers (AGES). Specifically, AGES employs a dual-attention network that meticulously captures the nuanced preferences of viewer groups in a live streaming room. Furthermore, the product representation is intricately crafted from viewer-product interaction data, coupled with insights into the similarities among products broadcasted by the streamers. To validate the efficacy of AGES, we conduct a comprehensive comparison against existing live recommendation models on two real datasets. The results show its superior performance and demonstrate its ability to better align with the diverse preferences of e-commerce live stream viewers.