DeepFFMS: A Parallel Model for Advertise Click-Through Rate Prediction

As a metric to measure the effect of advertising, the click-through rate has always been concerned by the company's operators, and the algorithm for estimating the click-through rate (CTR) of the advertisement can judge the expected benefits brought by the advertisement in advance. In this paper, we propose a parallel-structure click-through rate prediction model that enables selection of combined features. The proposed model, DeepFFMS, is mainly composed of the optimized field factorization machines and the popular neural network structure in recent years. Compared with other parallel-structure click-through rate prediction models, DeepFFMS uses parameters similar to attention mechanism training to eliminate useless feature interactions or feature interactions that have a negative impact on click-through rate prediction, thereby improving the accuracy of the model. After comprehensive experiments on open source advertising data sets, it is proved that DeepFFMS is superior to other similar models in the problem of advertising click-through rate estimation.