PCR-CA: Parallel Codebook Representations with Contrastive Alignment for Multiple-Category App Recommendation
Researchers have developed PCR-CA, a new framework designed to improve app recommendation systems by better handling apps with multiple categories. This approach uses a Parallel Codebook VQ-AE module to learn discrete semantic representations, allowing for the independent encoding of diverse app aspects. A contrastive alignment loss is applied to bridge semantic and collaborative signals, particularly benefiting long-tail apps. Experiments and online A/B testing on the Microsoft Store showed significant improvements in click-through and conversion rates. AI
IMPACT This new framework improves app recommendation accuracy and user conversion rates, particularly for multi-category and long-tail apps.