TransX: Scaling Transformer-based Recommendation via Behavioral and Serving Stream Crossings

2026-07-30

AI Summary

This work addresses the inefficiency and high computational cost in existing recommender systems that conflate heterogeneous long-term user behaviors with real-time service events into a single sequence, thereby obscuring their distinct causal roles and temporal characteristics. To resolve this, the authors propose TransX—a production-oriented encoder-decoder architecture that reframes recommendation as a cross-stream action transduction task between behavior and service streams. TransX explicitly decouples these streams and enables efficient interaction via cross-attention mechanisms. The approach integrates incremental behavior encoding, request-level key-value caching, and an amortized online serving strategy. Evaluated in large-scale A/B tests on LinkedIn, TransX achieves a 6.0% lift in click-through rate and a 4.4% improvement in conversion rate while reducing online computation by approximately 80%, maintaining service costs comparable to current production models.

Abstract

Modern industrial recommender systems (RecSys) increasingly adopt Transformer-based sequence models, with an emerging paradigm that frames recommendation as next-token prediction over a unified monolithic user sequence. However, collapsing heterogeneous data sources -- such as long-term user behaviors and real-time serving events -- into a single monolithic token stream that obscures their distinct causal roles and temporal characteristics, leading to inefficient modeling and elevated training and serving costs. We propose TransX, a production-oriented encoder-decoder architecture that reformulates recommendation as a sequence-to-sequence action transduction problem. TransX explicitly decouples behavior-stream modeling from serving-event modeling and conditions next-action decoding on scalable cross-attention between nearline behavior encodings and real-time serving representations. To enable low-latency, high-QPS deployment, TransX is co-designed with an amortized serving strategy that combines incremental behavior encoding with per-request key-value caching, rendering serving latency insensitive to behavior sequence length. Extensive offline experiments and large-scale online A/B tests on LinkedIn's recommender systems show that TransX consistently outperforms state-of-the-art DLRMs and sequential baselines, and delivers substantial CTR lift (+6.0%) and conversion gain (+4.4%) while maintaining serving costs comparable to existing production models where our co-designed serving strategy reduces online computation by approximately 80%.

Problem

• Transformer-based recommendation
• heterogeneous data sources
• monolithic token stream
• causal roles
• temporal characteristics

Innovation

• Transformer-based recommendation
• sequence-to-sequence transduction
• behavior-serving stream decoupling
• cross-attention mechanism
• amortized serving strategy