AdnFM: an Attentive DenseNet Based Factorization Machine for Click-Through-Rate Prediction

In this paper, we introduce a novel deep learning-based model named AdnFM, to attack the Click-Through-Rate (CTR) prediction problem. The AdnFM includes two important components to extract both low-order and high-order features of users and items, to jointly learn a comprehensive representation for the prediction. It further combines residual learning and an attention mechanism, to enable high-order features interactions and weight their importance dynamically. We conduct extensive experiments to evaluate the performance of the proposed architecture. Results show that the AdnFM outperforms popular baselines on two offline dataset. We deploy our model on an online CTR prediction application. Online A/B test demonstrates that the proposed AdnFM achieves remarkable performance and significantly outperform other benchmarks.