FARM: Frequency-Aware Model for Cross-Domain Live-Streaming Recommendation

TL;DR

FARM addresses data sparsity in cross-domain live-streaming recommendation by leveraging a frequency-aware intra-domain module that separates high-frequency (sparse but valuable) behaviors from low-frequency signals, and a preference align-before-fuse strategy to transfer user preferences from short-video to live-streaming domains. The framework combines an embedding layer, a DFT-based decomposition with trainable frequency weights, a contrastive cross-domain alignment module, and a cross-domain fusion module that uses tailored attention to produce a cross-domain representation fed into a multi-task predictor. Empirical results on Kuaishou show offline gains in AUC, UAUC, and GAUC and online improvements in gifts, likes, and follows, with notable benefits for low-active users, validating practical impact. FARM has been deployed to serve hundreds of millions of users, demonstrating real-world scalability and revenue implications for live-streaming platforms.

Abstract

Live-streaming services have attracted widespread popularity due to their real-time interactivity and entertainment value. Users can engage with live-streaming authors by participating in live chats, posting likes, or sending virtual gifts to convey their preferences and support. However, the live-streaming services faces serious data-sparsity problem, which can be attributed to the following two points: (1) User's valuable behaviors are usually sparse, e.g., like, comment and gift, which are easily overlooked by the model, making it difficult to describe user's personalized preference. (2) The main exposure content on our platform is short-video, which is 9 times higher than the exposed live-streaming, leading to the inability of live-streaming content to fully model user preference. To this end, we propose a Frequency-Aware Model for Cross-Domain Live-Streaming Recommendation, termed as FARM. Specifically, we first present the intra-domain frequency aware module to enable our model to perceive user's sparse yet valuable behaviors, i.e., high-frequency information, supported by the Discrete Fourier Transform (DFT). To transfer user preference across the short-video and live-streaming domains, we propose a novel preference align before fuse strategy, which consists of two parts: the cross-domain preference align module to align user preference in both domains with contrastive learning, and the cross-domain preference fuse module to further fuse user preference in both domains using a serious of tailor-designed attention mechanisms. Extensive offline experiments and online A/B testing on Kuaishou live-streaming services demonstrate the effectiveness and superiority of FARM. Our FARM has been deployed in online live-streaming services and currently serves hundreds of millions of users on Kuaishou.

Figures (4)

• Figure 1: An illustration of (a) relative percentages of different user behaviors; (b) relative percentages of exposure content in our live-streaming services on Kuaishou.
• Figure 2: The overall framework of FARM. The intra-domain frequency-aware module enable the model to perceive user's sparse yet valuable behaviors in both domains. The preference align before fuse strategy first apply the contrastive learning to align user preference in both domains, and further adopt a serious of tailor-designed attention mechanisms to fuse user preference in both domains, thus achieving the preference transfer across the short-video and live-streaming domains.
• Figure 3: A statistical result of the cosine similarity with respect to (a) w/o C-PA variant, and (b) FARM.
• Figure 4: Visualization of the fused user preference.