Big Data Analytics in E-Commerce using Distributed Deep Neural Networks

Currently, there is a pressing need to emphasize big data analytics (BDA) within the e-commerce sector. Despite its potential, BDA remains not sufficiently explored, which obstructs both its theoretical and practical advancement. This paper systematically reviews the literature on BDA in e-commerce, offering a comprehensive framework that addresses its definitional aspects, characteristics, types, business values and challenges within the e-commerce context. Furthermore, the paper encourages broader discussions regarding future research challenges and opportunities in both theoretical and practical domains. Overall, the findings provide a thorough understanding of various BDA concepts, offering deeper insights into its diverse applications in e-commerce. In recent years, distributed deep neural networks (DDNNs) and neural networks (NNs) have demonstrated exceptional performance across a wide range of applications. For instance, deep convolutional neural networks (DCNNs) have continuously acquired new and additional features for various tasks in computer vision. Concurrently, a large number of end devices including Internet of Things (IoT) devices, has expanded significantly. These devices present attractive targets for machine learning applications due to their frequent connection to sensors such as cameras, microphones, and gyroscopes, which record substantial amounts of input data in a streaming mode. This study delineates the architecture of a Distributed Deep Neural Network (DDNN) that encompasses various computational hierarchies, including multiple end devices, edge nodes, and cloud platforms. The proposed concept is recognized for being new due to its belief in two fundamental layers: the convolutional layer and the pooling layer. The primary aim of integrating these two layers is to achieve optimal performance outcomes. Our findings demonstrate that the proposed methodology yields superior results in terms of accuracy and costeffectiveness, achieving a precision rate of 99% and an economical cost of 25. Consequently, we conclude that these results surpass those reported in recent literature.