Bridging the Information Gap: Semantic Densification and Hindsight Distillation for Cold-Start Prediction
Artificial Intelligence
Summary
The authors address the challenge of predicting the value and likelihood of purchase from new users who have little interaction data on e-commerce platforms. They identify issues with existing methods: one that uses large language models (LLMs) produces messy explanations, and another that transfers knowledge from a detailed teacher model to a simpler student model struggles with inconsistent user data. Their solution, SemRaD, creates a cleaner, structured summary of user information and uses a special training method to transfer knowledge more effectively, adjusting for differences between users. Tests on real-world data show that SemRaD improves prediction accuracy and performs well even with less training data.
Authors
Hao Duong Le, Yifei Gao, Huan Li, Lun Jiang, Chen Bai, Ke Xing, Chen Zhang
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%.