Robust Spatio-Temporal Purchase Prediction via Deep Meta Learning

Purchase prediction is an essential task in both online and offline retail industry, especially during major shopping festivals, when strong promotion boosts consumption dramatically. It important for merchants to forecast such surge of sales have better preparation. This a challenging problem, as the purchase patterns festivals are significantly different from usual cases also rare historical data. Most existing methods fail at this problem due extremely scarce data samples well inability capture complex macroscopic spatio-temporal dependencies city. To address we propose Spatio-Temporal Meta-learning Prediction (STMP) model festivals. STMP meta-learning based multi-task deep generative model. adopts framework with few-shot learning capability spatial temporal representations. A component then uses extracted representation input infer results. Extensive experiments demonstrate generalization ability STMP. outperforms baselines all cases, which shows effectiveness our