CADCN: A click-through rate prediction model based on feature importance

Abstract

Recommendation systems are widely used in real-world advertising recommendations. In traditional recommendation system prediction models, click-through rate plays a crucial role. However, traditional recommendation systems cross-combine original features to make the linear model memorable and generalizable while taking into account the importance of features requires a lot of computational cost, and it uses manual cross-combination of features on data, which requires a lot of time and effort. Traditional recommendation systems simply learn the relationship between features without considering the importance of features. We combine deep crossover network and AFM network, dynamically assign weights to different features prior to feature crossover using a synthetic incentive network, and introduce attention mechanism based on feature crossover explicitly. We then propose an advertisement click-through rate prediction based on feature importance model, and the experimental results demonstrate that the algorithm is superior to the deep crossover network in predicting the click-through rate of advertisements.