Exploration of the Impact of Different Deep Learning Network Architectures on Click-Through Rate Prediction Models Based on Gated Deep Cross Networks

Abstract

This paper investigates advertisement click-through rate (CTR) prediction models based on Gated Deep Cross Networks and analyzes the impact of different deep learning network architectures on prediction performance. The paper provides an introduction to the background and related research of CTR prediction tasks, as well as a detailed discussion of the model structure and design. In the experimental section, experiments were conducted using publicly available datasets. Various network structure configurations were tested, including changes in activation functions and dropout rates. The results indicate that certain configurations in the experimental group exhibited superior performance compared to the control group, showing significant advantages. Overall, the paper demonstrates the strong performance of Gated Deep Cross Networks in CTR prediction and explores how different network structures affect model performance. Finally, the research work is summarized, and future research directions are proposed, providing valuable insights and references for CTR prediction tasks.