Search papers, labs, and topics across Lattice.
This paper introduces STAR, a novel framework for post-click conversion rate (PCVR) prediction that effectively integrates structured tokenization and target-aware interest representation within a HyFormer-style multi-sequence architecture. The approach addresses challenges posed by high-cardinality sparse features and inconsistencies between training and inference, leading to significant improvements in ranking AUC through enhanced temporal context and user-item interaction tokens. Key findings reveal that while temporal context provides the most substantial gains, additional benefits arise from contrastive alignment and target-aware encoding strategies.
Temporal context boosts PCVR prediction accuracy significantly, revealing the critical importance of user behavior dynamics in recommender systems.
Post-click conversion rate (PCVR) prediction is a core ranking task in industrial recommender systems. Modern ranking models must jointly capture heterogeneous non-sequential features, multi-behavior user sequences, and target-item-aware user interests, while remaining robust to high-cardinality sparse features, missing values, and train-inference inconsistencies. In this paper, we present STAR (Structured Tokenization and Target-Aware Interest Representation), a practical framework for the KDD Cup 2026 Tencent UniRec Challenge. STAR combines structured feature tokenization with target-aware interest representation on top of a HyFormer-style multi-sequence backbone. It introduces high-cardinality signal recovery, explicit user-item interaction tokens, target-aware sequence decoding, and a weighted user-item contrastive auxiliary objective inspired by InfoNCE. We further align the training and inference pipelines by reconstructing feature remapping tables and structural hyperparameters from the saved training configuration. Experiments on the challenge dataset identify the components that most reliably improve ranking AUC, while LogLoss is reported as a calibration diagnostic. The main ablation study shows a large gain from temporal context, with smaller but useful contributions from contrastive alignment, target-aware interest encoding, and high-cardinality sequence feature recovery.
