RCLRec: Reverse Curriculum Learning for Modeling Sparse Conversions in Generative Recommendation

2026-03-30

AI Summary

This work addresses the challenge of sparse conversion signals in large-scale recommender systems, which hinders effective modeling of user decision-making. To tackle this issue, the authors propose RCLRec, a novel framework that introduces reverse curriculum learning into generative recommendation for the first time. For each conversion target, RCLRec extracts a relevant subsequence from the user’s historical interactions in reverse chronological order to serve as a semantic prefix, which is then jointly generated with the target conversion token to better capture critical decision processes. The approach integrates behavior-aware attention mechanisms and a curriculum-quality-aware loss function to mitigate data sparsity. Extensive offline experiments and online A/B tests demonstrate significant improvements, yielding a 2.09% increase in ad revenue and a 1.86% uplift in order volume.

Abstract

Conversion objectives in large-scale recommender systems are sparse, making them difficult to optimize. Generative recommendation (GR) partially alleviates data sparsity by organizing multi-type behaviors into a unified token sequence with shared representations, but conversion signals remain insufficiently modeled. While recent behavior-aware GR models encode behavior types and employ behavior-aware attention to highlight decision-related intermediate behaviors, they still rely on standard attention over the full history and provide no additional supervision for conversions, leaving conversion sparsity largely unresolved. To address these challenges, we propose RCLRec, a reverse curriculum learning-based GR framework for sparse conversion supervision. For each conversion target, RCLRec constructs a short curriculum by selecting a subsequence of conversion-related items from the history in reverse. Their semantic tokens are fed to the decoder as a prefix, together with the target conversion tokens, under a joint generation objective. This design provides additional instance-specific intermediate supervision, alleviating conversion sparsity and focusing the model on the user's critical decision process. We further introduce a curriculum quality-aware loss to ensure that the selected curricula are informative for conversion prediction. Experiments on offline datasets and an online A/B test show that RCLRec achieves superior performance, with +2.09% advertising revenue and +1.86% orders in online deployment.

Problem

• sparse conversions
• generative recommendation
• conversion sparsity
• recommendation systems
• behavior modeling

Innovation

• Reverse Curriculum Learning
• Generative Recommendation
• Sparse Conversions
• Intermediate Supervision
• Behavior-aware Modeling