Click-Through-Rate Prediction Using Deep Neural Networks and Efficient Channel Attention Mechanisms

The progress of technology and the Internet has brought the development of online advertising into a developed period. To optimize the accuracy of advertisement, click-through-rate rate, the research proposed an advertisement click-through-rate prediction model based on deep neural network combined with efficient channel attention network. This model consists of four parts: embedding layer, interaction layer, efficient channel attention layer, and prediction layer. The embedding layer is responsible for passing the feature vectors to the interaction layer. higher-order feature interactions are learned through deep neural networks and efficient channel attention networks are introduced for lower-order feature interactions. higher-order feature interactions can capture the nonlinear and complex relationships between original features, while low-level feature interactions mainly focus on the relationships between a few features. The channel attention layer integrates the original features with the features that have already been interacted with by the interaction layer. The prediction layer uses perceptrons to predict click-through-rates. The proposed model is compared with logistic regression, deep feature crossover network, and deep factorization machine on Criteo, Avazu, KDD12, and MovieLens-1M datasets. The results showed that when the network depth was 1, the area under the curve of the proposed model was 0.8377, which was 10.4% higher than that of the logistic regression model. The average logarithmic loss was 0.1985, which was lower than that of the comparison model. The UC value of the model in the KDD12 dataset was 0.7879 and the logarithmic loss value was 0.4478. Taken together, the proposed model of the study is able to predict click-through-rates more accurately and has better model performance.