MOEF: Modeling Occasion Evolution in Frequency Domain for Promotion-Aware Click-Through Rate Prediction

Promotions are becoming more important and prevalent in e-commerce to attract customers boost sales, leading frequent changes of occasions, which drives users behave differently. In such situations, most existing Click-Through Rate (CTR) models can’t generalize well online serving due distribution uncertainty the upcoming occasion. this paper, we propose a novel CTR model named MOEF for recommendations under occasions. Firstly, design time series that consists occasion signals generated from business scenario. Since discriminative frequency domain, apply Fourier Transformation sliding windows upon series, obtaining sequence spectrums is then processed by Occasion Evolution Layer (OEL). way, high-order representation can be learned handle uncertainty. Moreover, adopt multiple experts learn feature representations aspects, guided via an attention mechanism. Accordingly, mixture obtained adaptively different occasions predict final CTR. Experimental results on real-world datasets validate superiority A/B tests also show outperforms representative significantly.