Researchers have developed TransX, a novel encoder-decoder architecture for recommender systems that addresses the challenges of integrating long-term user behaviors with real-time serving events. This approach decouples the modeling of these distinct data streams and uses cross-attention for more efficient decoding. TransX is designed for low-latency, high-throughput deployment with an amortized serving strategy that makes serving latency independent of behavior sequence length. Online A/B tests on LinkedIn's recommender systems demonstrated that TransX significantly improved click-through rates by 6.0% and conversion rates by 4.4%, while maintaining comparable serving costs and reducing online computation by approximately 80%. AI
IMPACT Introduces a more efficient architecture for Transformer-based recommenders, potentially improving user experience and reducing operational costs in large-scale systems.
RANK_REASON The item is a research paper published on arXiv detailing a new architecture for recommender systems. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
