Deep Filter Context Network for Click-Through Rate Prediction

The growth of e-commerce has led to the widespread use DeepCTR technology. Among various types, deep interest network (DIN), evolution (DIEN), and session (DSIN) developed by Alibaba have achieved good results in practice. However, above model’s filtering for user’s own historical behavior sequences insufficient context features lead reduced recommendation effectiveness. To address these issues, this paper proposes a novel article model: filter (DFCN). This improves efficiency attention mechanism adding out data sequence that differs greatly from target advertisement. DFCN pays through two local activation units. model expressiveness model, offering strong environment-related attributes adaptive capability with significant improvement up 0.0652 AUC metric when compared our previously proposed DICN under different datasets.