Analyzing E-Commerce Market Data Using Deep Learning Techniques to Predict Industry Trends

Introduction

With the rapid expansion of the e-commerce market, the rise of online retail has become an indispensable component of corporate sales, especially for some e-commerce retail enterprises. Online retail sales not only occupy a significant position but even constitute the entire sales revenue for these enterprises. As businesses increasingly rely on e-commerce platforms for sales, accurately predicting and assessing market demand has become a key factor in formulating appropriate production plans. In this context, the use of deep-learning technology to analyze e-commerce market data, particularly forecasting online sales trends, becomes crucial.

Historically, research has focused primarily on methods for forecasting product sales by analyzing historical sales data using time-series methods to identify changes in sales trends (Hong, 2021; Zhang et al., 2022). However, traditional time-series methods impose high requirements on the stationarity of data, and consumer behavior in e-commerce markets typically exhibits randomness and disorder. This makes time-series methods perform poorly when facing certain factors that influence user purchasing decisions, such as promotional strategies. In practical situations, the purchase-decision behaviors of individual users eventually converge into overall sales. However, due to the difficulty in acquiring factors related to individual purchase decisions in traditional sales-forecasting methods, these factors are rarely used for sales prediction. In the current internet environment, user browsing and purchasing behaviors are comprehensively recorded, providing businesses with the opportunity to utilize this data to identify factors influencing user purchasing decisions. Subsequently, deep-learning technology can be applied for product-sales forecasting.

In the e-commerce environment, the presentation of multimedia information is a crucial feature of transactions, especially in virtual environments where users find it challenging to directly observe and interact with actual products (Angeli et al., 2018; Gao et al., 2023b). User perception of products relies heavily on multimedia information presented on websites, including text, images, audio, and video. Among these, images, as the most intuitive and widely adopted form of information display, significantly influence user purchasing behavior through factors such as color and emotional appeal, making them a key determinant in user purchase decisions. Past studies have demonstrated that factors such as brightness, color, and emotions in product images can impact user purchase decisions. While research on utilizing image information for product-sales forecasting is currently limited, the rapid development of deep learning offers new methods for effectively handling image information. This development also provides technical support for integrating static and dynamic factors in product-sales forecasting.

Building upon the aforementioned analysis, this study focuses on leveraging deep-learning techniques to conduct in-depth analysis of e-commerce market data and predict industry development trends. The key contributions of this paper are outlined as follows:

• (1)

  Diverging from traditional sales-forecasting methods, this study employs three different types of neural networks—convolutional neural network (CNN), bidirectional long short-term memory (Bi-LSTM), and Fourier CNN (FCNN)—to abstract and represent market features at multiple levels. By combining dynamic and static features, this paper establishes an innovative sales-prediction model, offering a more accurate and comprehensive predictive tool for the industry.

• (2)

  This study maximizes the advantages of deep-learning methods in extracting image features and possessing strong learning capabilities. By applying these methods to sales forecasting, we achieve an end-to-end solution from raw data to final sales predictions. This approach facilitates the resolution of demand-forecasting issues in various industries, advancing research on industry development trends.

• (3)

  The proposed sales-forecasting method in this study relies entirely on learning from historical data, eliminating the impact of subjective judgments on predictions. Since the model design is based solely on product and sales data and excludes errors that may arise from managerial subjective factors, the predictions are more objective and trustworthy. This method allows for periodic sales forecasting, providing businesses with more reliable decision support.