Bridging the Information Gap: Semantic Densification and Hindsight Distillation for Cold-Start Prediction

2026-07-19

AI Summary

This work addresses the challenge of predicting customer lifetime value (LTV) and conversion rate (CVR) for new users in e-commerce cold-start scenarios by proposing the SemRaD framework. The approach innovatively transforms free-form reasoning from large language models into structured semantic user profiles and introduces a semantic-gated encoder alongside a hindsight-aware distillation network. By incorporating a Distillation Experts module that leverages privileged information during training to guide the student model, SemRaD effectively bridges the information gap between teacher and student models while accounting for user heterogeneity. Experiments on large-scale industrial data demonstrate a 1.9% improvement in LTV (measured by Gini) and a 1.0% gain in CVR (measured by AUROC). Online A/B tests confirm consistent gains of 1.0% in LTV and 0.43% in CVR, and the model achieves baseline-level LTV performance using only 9% of the training data while improving CVR by 0.8%.

Abstract

New-user cold-start is a critical bottleneck for e-commerce platforms: predicting user lifetime value (LTV) and conversion rate (CVR) for users with sparse interaction history. Two prior directions -- LLM-based semantic augmentation and learning using privileged information (LUPI) -- each face a key limitation. First, LLM augmentation produces unstructured rationales that are noisy and hard to operationalize in production. Second, naive student-teacher distillation can be brittle due to an information gap between the privileged teacher and the sparse student; moreover, this gap is heterogeneous across users. We propose SemRaD, a Semantic Reasoning-aware Distillation framework addressing both limitations. First, a Structured Semantic Reasoning Pipeline replaces free-form rationales with a structured schema built via a discover-curate-audit workflow, producing per user a Densified Semantic Profile (consumed by the deployed student via a Semantic-Gated Encoder that focuses on the most informative dimensions) and a Hindsight Distillation Target reconciled from pre- and post-conversion reasoning (used only at training). Second, to bridge this gap and handle its heterogeneity, a Hindsight-Aware Distillation Network transfers privileged knowledge via the hindsight target, with Distillation Experts improving transfer under per-user variability. On a large-scale industrial dataset, SemRaD lifts +1.9% LTV (Gini) and +1.0% CVR (AUROC) over a production-grade base; a four-week online A/B at Keeta confirms +1.0% LTV / +0.43% CVR. SemRaD also matches the production system's LTV using only 9% of the training data while improving CVR by 0.8%.

Problem

• cold-start
• user lifetime value
• conversion rate
• semantic augmentation
• privileged information

Innovation

• Semantic Densification
• Hindsight Distillation
• Cold-Start Prediction
• Structured Semantic Reasoning
• Knowledge Distillation