Optimizing the Operation Mechanism of Public Data Assets and Data-Driven Decision Models in the Digital Economy With Deep Neural Networks

Introduction

In this new era of the digital economy, public data assets play a pivotal role in driving social and economic innovation and development. As various technologies, such as big data, the Internet of Things, and cloud computing, advance rapidly, some profound changes have occurred with respect to the acquisition (Thara et al., 2022; Umer & Sharif, 2022), storage, and application of public data. These data, which have extremely high social value and massive economic potential, cover important information in various fields related to government, enterprises, and society, including transportation (Saouabe et al., 2024), medical care, environment, and education. However, despite the wide application scenarios of public data, efficiently integrating, analyzing, and using these data remains an outstanding issue. Challenges remain in extracting valuable information from data and making scientific decisions (Ooi et al., 2025), especially in the context of huge data volume, diverse types of data, and dynamic changes.

With the continuous development of data analysis methods, data-driven decision models have gradually become recognized as an important tool to solve complex problems in all walks of life. It is imperative to support decision-making through accurate data analysis for governments and enterprises engaged in resource management (Liu & Vakharia, 2024; Wang & Srivastava, 2020), urban planning, public health, and so on, especially in the public sector. However, a majority of the existing decision-making models rely on the traditional methods of data analysis, which perform poorly in processing large-scale time series data and multimodal data (Huang & Vakharia, 2024). As a result, it is often difficult to fully tap the potential value in the data, and there are limitations in dealing with dynamically changing environments (Sakib et al., 2025; Soltis et al., 2018). Current research has noted that it is challenging to address these shortcomings and improve the accuracy and stability performance of data-driven decision-making models.

In recent years, there has been a significant progress made by deep learning technology in many fields, especially the processing of time series data and image data (He et al., 2025; Xing et al., 2025). Convolutional neural networks (e.g., a residual network [ResNet]) and long short-term memory (LSTM) networks represent two key technologies. ResNet is applicable to processing image data and extracting high-level features, especially in image recognition tasks, whereas LSTM excels in modeling time series data (Li et al., 2025) and can be used to capture long-term dependencies, which facilitates time series prediction. These two technologies can be combined to better address the complex scenario in which image data and time series data coexist in public data (Shiau et al., 2024; Xia et al., 2024). In addition, various hyperparameter-optimization methods have been proposed in recent years, such as Bayesian optimization and grid search, which makes deep learning models suitable to adjusting model hyperparameters more efficiently, thus further improving the model’s predictive ability and generalization performance.

Despite the important progress that has been made by deep learning technology in public data analysis, the effective combination of these technologies to overcome the limitations of existing models in the actual application of the digital economy and public decision-making remains an outstanding issue. To achieve this aim, in this article we propose an innovative hybrid deep learning model, namely xxxx (ReLSTM-Opt). ReLSTM-Opt combines ResNet, LSTM, and hyperparameter optimization technology to enhance the efficiency and accuracy of public data asset analysis and data-driven decision-making. To be specific, the ReLSTM-Opt model starts by extracting image features from public data through ResNet. Then, the dynamic changes in time series data are captured through LSTM, and model parameters are adjusted through hyperparameter optimization to ensure the model’s optimal performance when large-scale complex data are processed.

The innovative contributions of this article can be summarized as follows: