Residual Connections Improve Prediction Performance
- Resource Type
- Conference
- Authors
- Bicici, Ergun; Kanburoglu, Ali Bugra; Turksoy, Ramazan Tarik
- Source
- 2023 4th International Informatics and Software Engineering Conference (IISEC) Informatics and Software Engineering Conference (IISEC), 2023 4th International. :1-5 Dec, 2023
- Subject
- Bioengineering
Communication, Networking and Broadcast Technologies
Computing and Processing
Robotics and Control Systems
Signal Processing and Analysis
Training
Visualization
Three-dimensional displays
Artificial neural networks
Computer architecture
Predictive models
Robustness
residual connections
neural networks
click-through rate
CTR
- Language
Click-through rate (CTR) prediction is a critical task in online advertising and recommendation systems, where the depth and complexity of learning models have become increasingly challenging. This paper addresses the challenges associated with CTR prediction by introducing residual connections to enhance CTR prediction models. In this paper, we investigate the integration of residual connections into CTR prediction models. Experiments involve the application of plain MaskNet and MaskNet enhanced with residual connections on benchmark datasets from both the company and Avazu. Our findings demonstrate that residual connections can be effectively integrated into CTR prediction models, increasing the AUC by 0.77% and decreasing the loss by 2.8%.