XCrossNet: Feature Structure-Oriented Learning for Click-Through Rate Prediction

Click-Through Rate (CTR) prediction is a core task in nowadays commercial recommender systems. Feature crossing, as the mainline of research on CTR prediction, has shown promising way to enhance predictive performance. Even though various models are able learn feature interactions without manual engineering, they rarely attempt individually representations for different structures. In particular, mainly focus modeling cross sparse features but neglect specifically represent dense features. Motivated by this, we propose novel Extreme Cross Network, abbreviated XCrossNet, which aims at learning and an explicit manner. XCrossNet structure-oriented model leads more expressive representation precise not only interpretable, also time-efficient easy implement. Experimental studies Criteo Kaggle dataset show significant improvement over state-of-the-art both effectiveness efficiency.