Abstract
With the advent of the Internet, individuals have access to an ever-increasing array of external information sources, and as such, advertising information has undergone a significant transformation. In this study, we present a highly accurate, high-throughput, and high-yield ad recommendation system that is capable of precisely targeting users for personalized ad recommendations. Our approach involves enhancing the user interest preference model, introducing ad keywords as labels into the similarity calculation of Query pages, and employing a weighted integrated similarity measure of Query pages to mitigate sparsity within the similarity matrix. Furthermore, we optimize the system process of ad promotion and devise an ad recommendation strategy based on multiple parallel recall and uniform sorting distribution. Our model training results indicate that the proposed algorithm elevates Precision, Recall, and F1 by 27%, 25%, and 28%, respectively, compared to the traditional model. Additionally, our system test results demonstrate enhanced scalability, with approximately four times higher concurrent performance, as well as improvements in high expansion, low latency, and strong stability. Such results hold particular significance in guiding the design of actual ad recommendation system projects.
Publisher
Agora University of Oradea
Subject
Computational Theory and Mathematics,Computer Networks and Communications,Computer Science Applications