Deep Field Relation Neural Network for click-through rate prediction

Click-Through Rate (CTR) prediction is crucial in calculating advertisements and recommendation systems. To effectively predict CTR, it is important to properly model the interaction among features of data. This work tends to fully utilise the interaction information among features while employing deep neural networks for CTR prediction. To this end, we propose a Deep Field Relation Neural Network (DFRNN), which models feature interaction via a 3-dimensional relation tensor. The proposed method is evaluated on real data sets and compared with related methods. The results demonstrate that our method could be used to derive significant information contained in feature interaction and achieve an accurate CTR prediction.