PSG: Pair-Space Generation for Efficient Generative Reranking

2026-07-28

AI Summary

This work addresses the high latency and train-test inconsistency inherent in autoregressive generative reranking for recommendation systems, which stem from long item sequences. To mitigate these issues, the authors propose a pairwise generative framework that elevates the generation unit from individual items to ordered item pairs, requiring only L/2 pair-tokens to be generated. This approach is augmented with a pretrained pairwise representation module. The method preserves expressive power over sequence distributions while theoretically achieving 2–4× speedup and reducing worst-case suboptimality by nearly fourfold, thereby alleviating train-test discrepancy. Deployed on Kuaishou’s platform serving over 400 million daily active users, the solution demonstrates a 1.83× empirical inference speedup and a statistically significant 0.178% increase in average user dwell time.

Abstract

Modern recommender systems adopt Generator-Evaluator (G-E) for list-wise reranking: a generator produces sequences from candidates and an evaluator scores them at sequence-level to filter out the optimal one for exposure. Auto-Regressive(AR), working as the backbone for generative recommendation, suffers two limitations. First, its complexity grows linearly with list length, forcing the system to generate fewer lists under rigorous latency constraints and thus limiting exploration. Second, teacher-forcing creates a train-test mismatch; cumulative errors worsen with length and degrade quality. To address these problems, we propose Pair-Space Generation (PSG), a reformulation that elevates the generation atom from individual items to ordered item pairs. Given $n$ candidate items, PSG operates over pair vocabulary of size $n(n-1)$ per request, generates only $L/2$ tokens. Pair token representations are produced on-the-fly by a pretrained pair-token representation module optimized over large scale exposure logs, eliminating the data sparsity that would otherwise plague a quadratic sized vocabulary. We establish three theoretical guarantees: (i) PSG is bijective with item-space generation and induces an equivalent family of sequence distributions, thus incurring no loss of expressiveness; (ii) generation in pair-token space achieves approximately a $2\times$ to $4\times$ speedup theoretically under moderate settings and $1.83\times$ in the real industrial environmental settings; and (iii) under outcome-only rewards, the worst-case suboptimality of PSG is bounded by $O((L/2)^2 \barε)$, representing a nearly $4\times$ improvement over item-space generation. Beyond benchmark-based validation, PSG has also been deployed on Kuaishou, delivering a 0.178\% lift in per-user stay time on the platform, which serves over 400 million daily active users.

Problem

• generative reranking
• auto-regressive generation
• latency constraints
• train-test mismatch
• sequence-level recommendation

Innovation

• Pair-Space Generation
• Generative Reranking
• Sequence-Level Recommendation
• Efficient Inference
• Pair-Token Representation