OneTrans: Unified Feature Interaction and Sequence Modeling with One Transformer in Industrial Recommender

TL;DR

OneTrans is a unified Transformer model that integrates user-behavior sequence modeling and feature interaction, enabling efficient, scalable, and improved recommendation performance through shared parameters and caching strategies.

Contribution

It introduces a unified Transformer backbone with a shared tokenizer and parameter sharing for both sequence and feature interaction modeling, unifying and scaling recommendation tasks.

Findings

Outperforms strong baselines on industrial datasets

Achieves a 5.68% lift in per-user GMV in online tests

Scales efficiently with increasing model size

Abstract

In recommendation systems, scaling up feature-interaction modules (e.g., Wukong, RankMixer) or user-behavior sequence modules (e.g., LONGER) has achieved notable success. However, these efforts typically proceed on separate tracks, which not only hinders bidirectional information exchange but also prevents unified optimization and scaling. In this paper, we propose OneTrans, a unified Transformer backbone that simultaneously performs user-behavior sequence modeling and feature interaction. OneTrans employs a unified tokenizer to convert both sequential and non-sequential attributes into a single token sequence. The stacked OneTrans blocks share parameters across similar sequential tokens while assigning token-specific parameters to non-sequential tokens. Through causal attention and cross-request KV caching, OneTrans enables precomputation and caching of intermediate representations, significantly reducing computational costs during both training and inference. Experimental results on industrial-scale datasets demonstrate that OneTrans scales efficiently with increasing parameters, consistently outperforms strong baselines, and yields a 5.68% lift in per-user GMV in online A/B tests.