Enhancing Click-Through Rate Prediction: A Composite Approach Integrating DNN with DCN and FM-NN

Click-through rate (CTR) is a crucial measure used in system of recommendations and online advertising. It assesses how effective advertisements or content are by gauging the percentage of users who engage with them. It is a key performance indicator, influencing advertising revenue, content engagement, and user interaction. Furthermore, it empowers businesses to maximize return on investment and boost user engagement by targeting the right audience with the right content. The accurate CTR prediction can unlock a multitude of advantages, including the optimization of ad placements, content personalization, and improved resource allocation. The use of deep learning models will estimate and enhance the likelihood that a user will check on the specific advertisement or link. In this paper, an innovative approach to improve CTR prediction through the use of composite models was introduced. This paper delves into the core rationale behind CTR and its prediction. The first model combines a Deep Neural Network (DNN) with a Deep Cross Network (DCN), effectively merging the DNN’s ability to learn features with the DCN’s knack for capturing complex feature interactions. Through extensive experimentation and evaluation, the proposed work aims to demonstrate the superior predictive performance of these composite models for enhancing CTR prediction. Additionally, this PAPER introduces a second composite model that brings together a DNN with a sophisticated composite model incorporating Factorization Machines (FMs) and a Neural Network (NN). This unique combination takes advantage of FM’s capability to capture intricate feature interactions, complemented by the DNN’s proficiency in learning intricate data patterns The achieved results were exceptionally promising, showcasing an impressive AUC Score of 0.97 for the DNN-(FM-NN) model and AUC Score of 0.96 for the DNN-DCN model.