The FFCP Model for Predicting Click-Through Rate

In advertisement recommendations, click-through rate (CTR) prediction is a hugely sparse problem where embedding techniques and interaction models are needed to extract useful information from the data. Several CTR prediction and interaction models have been developed over the years that consider different parts of low-order and higher-order interactions. We develop the FFCP model, which combines multiple beneficial layers into a single model for better performance. An ablation study is performed to show the contribution of each model component. Our experiments demonstrate that FFCP achieves better AUC and loss than the 6 other models we compared on both of the two commonly used CTR prediction datasets and improves F1 results. The improvements reach 0.29% in AUC and 0.31% in loss.