Analysis of deep learning-based technological innovation governance on the intelligent …

Research background

In the era of rapid digitalization and intelligentization, deep learning (DL) technology is reshaping the development landscape of various industries at an unprecedented pace. From healthcare to financial services, from transportation to cultural entertainment, the application scope of DL continues to expand, profoundly transforming production and lifestyles^1,2,3,4,5. In the high-technology (high-tech) industry, the effectiveness of innovation governance plays a decisive role in the long-term development of the industry and the enhancement of national competitiveness. Characterized by high knowledge density, rapid technological iteration, and intense market competition, this industry has become a core driver of national economic growth and a strategic high ground in international competition^6,7. Selecting high-tech industries as the research object is necessary, as they have three typical characteristics: high knowledge density, short innovation cycles, and intensive data assets. Relevant surveys indicate that high-tech enterprises have a high innovation failure rate, largely due to inherent limitations in conventional resource allocation frameworks. These traditional models fail to effectively reconcile the competing demands of accelerating technological iteration and market dynamics⁸. However, DL governance improves resource allocation foresight through real-time parsing of multi-source data streams. This capability addresses the key governance demand that differentiates high-tech industries from traditional sectors. However, the industry currently faces serious challenges in the innovation resource allocation. Traditional resource allocation models are constrained by insufficient depth in data mining, limited capability to handle intricate nonlinear relationships, and difficulties in flexibly responding to dynamic market environments. These limitations result in inefficient utilization of innovation resources, leading to resource idleness and misallocation, thus severely hindering the pace and quality of innovation development in the high-tech industry^9,10,11,12. The rise of DL technology has brought new hope for addressing these challenges. With its powerful data analysis abilities and model construction, DL can extract valuable insights from massive datasets, accurately capturing market dynamics and technological trends. Moreover, it can provide novel approaches and methods for intelligent innovation resource allocation^{13,14,15,16,17,18}.

Research question

Reviewing the progress in related research fields, early scholars primarily focused on applying traditional innovation resource allocation theories in the high-tech industry. However, with the swift advancement of technology and the increasing complexity of market environments, these theories have gradually revealed their limitations^19,20,21. In recent years, some studies have begun to explore the impact of emerging technologies on innovation resource allocation; DL technology receives growing attention due to its unique advantages. Existing research has preliminarily investigated the role of DL in optimizing innovation resource allocation in specific scenarios. Nevertheless, current research lacks a systematic examination of how DL fundamentally transforms intelligent innovation resource allocation. This gap is particularly evident within the complex governance structures of high-tech industry innovation ecosystems^22,23,24,25.

Research objectives

Based on this, this study delves into the impact mechanisms and practical effects of DL-based technological innovation governance on intelligent innovation resource allocation in the high-tech industry. In the research process, the study first thoroughly explores the key components and application models of DL in the context of high-tech industry innovation governance. The study also identifies the types of DL algorithms suitable for innovation resource allocation data analysis, the corresponding data collection, and processing workflows, laying a solid foundation for subsequent research. Next, the study systematically analyzes diverse factors influencing innovation resource allocation and their intrinsic relationships. It investigates how DL enhances the scientific rigor and rationality of innovation resource allocation decisions by precisely analyzing these factors. This includes multidimensional analyses of the synergistic relationships among different innovation resources and the dynamic impacts of external environmental factors. Simultaneously, the study comprehensively examines the effects of DL-based innovation resource allocation strategies on the quantity and quality of innovation outputs in high-tech enterprises; it also examines their role in enhancing enterprise market competitiveness. Comparative analyses with traditional allocation strategies are conducted to quantitatively assess the advantages and value of the new strategies. Finally, building on rigorous research findings, the study offers actionable recommendations and policy guidance for high-tech industry managers and policymakers. It facilitates the widespread application of DL technology in innovation governance, significantly advancing the intelligence level of innovation resource allocation. Consequently, it offers robust support for the sustainable innovation and development of the high-tech industry.

In the high-tech industry, innovation resource allocation has always been a central focus. Lo²⁶ underscored that the appropriate allocation of innovation resources was critical for high-tech enterprises to secure a competitive advantage. His analysis examined how different resource combinations influenced the innovation process. Expanding on this perspective, Ferrati and Muffatto²⁷ investigated the role of human capital in innovation resource allocation. They revealed that the quality and quantity of highly skilled professionals substantially impacted the resource utilization efficiency in high-tech innovation.

With the emergence of DL, its applications in high-tech innovation governance have garnered extensive attention. Rane et al.²⁸ proposed a novel DL algorithm specifically designed to analyze the intricate relationships among various innovation resources. Their findings demonstrated that this algorithm outperformed traditional methods in accurately identifying key factors influencing resource allocation. Similarly, Kinne and Lenz²⁹ focused on using DL to predict market demand for high-tech products. By analyzing market trends and consumer behavior data, their DL model provided valuable insights to guide resource allocation decisions. This ensured that innovation resources were strategically directed toward areas with high market potential.

Regarding the impact on intelligent allocation, Ma et al.³⁰ found that DL-based innovation governance enhanced resource allocation flexibility. They highlighted that such systems could rapidly adjust resource allocation in response to technological disruptions and market fluctuations, ensuring a more adaptive and dynamic approach. Hunt et al.³¹ examined the integration of DL with supply chain management in the high-tech industry. Their results indicated that this integration optimized the flow of innovation resources within the supply chain, ultimately enhancing overall innovation efficiency.

Additionally, Verma and Singh³² explored the role of DL in facilitating interdisciplinary innovation resource allocation. Their research suggested that DL could break down traditional disciplinary barriers, enabling more seamless and efficient sharing of resources across different fields. Rahaman et al.³³ systematically explored the application of machine learning in business analysis, particularly emphasizing the promoting effect of statistical method innovation on data-driven innovation. Their study verified the ability of machine learning models to capture nonlinear relationships and proposed a collaborative framework of algorithm iteration and real-time data feedback. It provides a theoretical basis for this study to adopt the DL method to optimize the efficiency of innovation resource allocation.

Regarding DL-based dynamic innovation environment adaptability research and governance mechanisms, Rane et al.³⁴ used a dynamic governance framework to verify DL algorithms’ response capability to sudden market changes. Their study confirmed that a Long Short-Term Memory (LSTM)-based resource allocation model could enhance enterprises’ decision-making speed in response to technological disruptions while reducing resource idle rates. Sharifani and Amini³⁵ pointed out that DL-driven governance systems should be deeply coupled with organizational structures. Their case analysis of 128 tech enterprises showed that a sound governance system could improve the accuracy of DL models in resource allocation. This confirmed the moderating effect of “governance system completeness”.

In summary, existing research has made remarkable contributions to exploring the importance of innovation resource allocation and the potential of DL in this field. However, several research gaps remain that require further exploration. First, the literature on technological innovation governance predominantly focuses on traditional governance models; it lacks systematic research on DL-based technological innovation governance mechanisms and their impact on innovation resource allocation. Second, studies on intelligent innovation resource allocation are gradually increasing. However, they are mostly confined to static analyses of resource allocation efficiency, failing to fully reveal the adaptability and sustainability of intelligent allocation mechanisms in dynamic environments. Moreover, regarding the reverse impact of innovation resources on governance models, current literature predominantly examines the unidirectional influence of innovation resource endowments on governance structures. It overlooks the bidirectional interaction between resource allocation efficiency and governance system evolution. Based on the above limitations, this study innovatively proposes a DL-driven dynamic integration framework for technological innovation governance. This framework breaks through traditional research methods focusing on one-way influence; it systematically verifies the two-way interaction mechanism between the depth/breadth of DL application, the completeness of the governance system, and the efficiency/dynamic adaptability of innovation resource allocation. In addition, the study integrates dynamic capability theory and algorithm governance theory. This integration fills the gap in systematic research on DL governance mechanisms; it also provides a quantifiable theoretical model for technological innovation governance, realizing a theoretical shift from static description to dynamic coupling analysis. This theory offers new analytical dimensions and methodological tools for subsequent research.

Research methodology

Innovation governance framework

Before deeply exploring DL’s impact on intelligent innovation resource allocation in high-tech industries, it is necessary to define the “technological innovation governance framework”. This framework refers to a comprehensive management system aimed at optimizing the allocation and utilization of innovation resources through formulating strategies, processes, and organizational structures. Its main components include strategic planning, resource allocation, process monitoring, performance evaluation, and feedback adjustment. These components are interrelated, collectively promoting the effective conduct of innovation activities. Integrating DL technologies—particularly in predictive analysis, automated monitoring, and algorithmic decision support—can enhance the effectiveness of the technological innovation governance framework³⁶.

Research method design

This study employs a combination of empirical research and case study methods to thoroughly investigate the impact of DL-based technological innovation governance on intelligent innovation resource allocation in the high-tech industry. Empirical methods verify the universal laws between variables through large-sample statistical models to address the question of “whether there is a correlation”. Case studies involve in-depth interviews with enterprises to excavate practical details and dynamic processes, explaining the internal connections of “how the correlation works”. The selection of these methods is well-justified and appropriate for the research objectives.

The study leverages a stratified sampling approach to select 500 enterprises from five highly representative high-tech industries: information technology, biopharmaceuticals, new energy, advanced equipment manufacturing, and aerospace. This sampling method comprehensively considers enterprise size, age, technological innovation ability, and other factors. Enterprises of different sizes exhibit variations in resource reserves, market influence, and development strategies, leading to diverse patterns in innovation resource allocation. The age of an enterprise reflects its developmental stage, with startups, growth-stage enterprises, and mature enterprises each having distinct characteristics in innovation resource needs and utilization. Technological innovation ability directly influences the efficiency and direction of innovation resource allocation. Through this comprehensive and meticulous stratified sampling, the study ensures the diversity and representativeness of the sample, covering the actual conditions of various enterprises in the high-tech industry. This approach enhances the generalizability of the research findings and accurately reflects the patterns of innovation resource allocation across different contexts. Furthermore, this study’s main objective is to empirically test the effectiveness of enterprise-level practices. A multi-dimensional indicator system is constructed to accurately measure governance effects, as exhibited in Table 1.

The construction of a structural equation model (SEM) is a critical component of this study. The DL application depth/breadth and the completeness of the technological innovation governance system are treated as exogenous latent variables; The efficiency and the dynamic adaptability of intelligent innovation resource allocation are set as endogenous latent variables. Control variables such as the degree of technological innovation activities, enterprise size, age, and market competition intensity are also incorporated. The AMOS software is utilized for parameter estimation and goodness-of-fit tests to evaluate the model’s rationality and validity and to verify the related hypotheses. The innovation resource allocation in the high-tech industry is influenced by combining multiple factors, with intricate relationships and direct and indirect interactions. SEM is particularly effective in handling such intricate relationships among several variables. It accounts for measurement errors in variables and clearly illustrates the intrinsic connections between latent variables. This enables an in-depth analysis of the internal mechanisms and pathways through which DL-based technological innovation governance impacts intelligent innovation resource allocation. This approach helps researchers comprehensively understand the specific roles of various factors in the innovation resource allocation process, providing robust theoretical support for subsequent research and practical applications.

Following the completion of quantitative empirical research, case studies are conducted on 15 representative high-tech enterprises. These enterprises differ in industry, size, DL application level, and technological innovation governance models. The study conducts on-site investigations and in-depth interviews with personnel from various departments within the enterprises. It comprehensively examines the background and motivations for applying DL technology in technological innovation governance. Additionally, it analyzes the implementation processes and the challenges faced. The study also explores the strategies adopted and assesses the resulting innovation outcomes and economic benefits. Case studies offer a unique perspective for understanding the actual conditions of individual enterprises, capturing details and special circumstances that may be overlooked in empirical research. Given the unique operational characteristics of each enterprise, case studies can reveal personalized experiences and issues encountered by different enterprises in optimizing innovation resource allocation using DL. These case studies complement and validate the findings of empirical research. By analyzing and summarizing multiple cases, the study identifies common patterns and lessons learned from successful enterprises, offering more practical and actionable guidance for other high-tech enterprises. This promotes the widespread application and effective implementation of DL technology in intelligent innovation resource allocation within the high-tech industry.

Research model

This study employs the SEM as the core analytical framework to test the macro-path relationship of “DL governance and resource allocation”. Meanwhile, it uses Analysis of Moment Structure (AMOS) software to estimate the standardized coefficients between latent variables, answering the questions of “whether the influence is significant” and “the intensity of influence”. As a practical tool for enterprises, the application effect of DL technology is presented through the actual deployment scenarios of case enterprises. The variables of “DL application depth/breadth” in the SEM model are derived from the integrated measurement of these practical data^37,38,39,40. Among them, DL application depth refers to algorithm complexity and the intensity of computing resource investment; it is measured by the proportion of advanced models like Transformer and the proportion of computing power of GPU clusters. DL application breadth reflects the scope of technical coverage, measured by the application coverage rate of DL technology in innovation links such as R&D project approval, process monitoring, and result evaluation. Depth mainly focuses on technological advancement, while breadth emphasizes scenario penetration. Its measurement indicators have passed a reliability test with Cronbach’s α > 0.85 and have been cross-validated through management log audits and executive interviews. In addition, when constructing the research model, the study focuses on how DL enhances specific elements in the technological innovation governance framework. It is hypothesized that the application of DL technology can improve the effectiveness of innovation governance in the following ways:

1. (1)

   DL models are utilized to deeply mine massive data, predict the success probability and market potential of R&D projects. Thus, this optimizes the allocation of R&D resources and ensures priority support for high-value projects.

2. (2)

   Real-time monitoring of the progress and effectiveness of innovation activities is conducted through DL algorithms. Potential risks and problems are automatically identified, providing timely decision support for managers and enhancing the dynamic adaptability of resource allocation.

3. (3)

   Combining DL with optimization algorithms, the study provides intelligent decision suggestions for innovation resource allocation. This improves the scientificity and accuracy of decision-making while accelerating the decision-making process and enhancing overall innovation efficiency. To verify these hypotheses, this study incorporates the DL application depth/breadth as exogenous latent variables in the model. At the same time, it explores their impacts on the efficiency and dynamic adaptability of innovation resource allocation. Moreover, this study focuses on the completeness of the technological innovation governance system and how it acts as a moderator to enhance the effect of DL application⁴¹.

Empirical research design

1. (1)

   Sample selection

This study selects enterprises from five representative high-tech industries—information technology, biopharmaceuticals, new energy, advanced equipment manufacturing, and aerospace—as the research sample. A stratified sampling method ensures comprehensiveness and diversity, categorizing enterprises based on size, age, and technological innovation ability. Enterprise size is classified into three categories. Large, medium, and small enterprises have more than 5000, 1000–5000, and fewer than 1000 employees, respectively, with annual revenue exceeding 10 billion RMB, 1–10 billion RMB, and below 1 billion RMB. Enterprise age is divided into startups (established within the past three years), growth-stage enterprises (established for 3–10 years), and mature enterprises (established for over 10 years). Additionally, technological innovation ability is stratified into three levels based on the average annual patent application growth rate over the past five years: high, medium, and low innovation abilities. The average annual patent application growth rate for these levels over the past five years exceeds 20%, ranges from 10 to 20%, and is less than 10%, respectively. Ultimately, 500 enterprises are selected, with 100 from each industry. This ensures that the sample adequately reflects the innovation resource allocation under varying industry characteristics, enterprise sizes, development stages, and levels of innovation ability.

1. (2)

   Model construction

A SEM is developed based on the defined variables and measurements. In the SEM, the DL application depth/breadth and the comprehensiveness of the technological innovation governance system are treated as exogenous latent variables. The efficiency of intelligent innovation resource allocation and the dynamic adaptability of resource allocation are considered endogenous latent variables. Control variables encompass enterprise size, industry-level technological innovation activity, enterprise age, and market competition intensity. An SEM path diagram is drawn to clarify the hypothesized relationships between variables. For example, it is hypothesized that the DL application depth/breadth directly and positively influences the efficiency of intelligent innovation resource allocation and its dynamic adaptability. To support this hypothesis, a review of related studies reveals that the expansion of DL in different dimensions remarkably impacts innovation governance⁴². Therefore, depth and breadth may exhibit certain homogeneity in some contexts. However, their unique roles in promoting technological innovation governance cannot be ignored, and the adoption of two separate indicators is fully logically reasonable. Moreover, the comprehensiveness of the technological innovation governance system is posited to play a moderating role, strengthening the positive impact of DL applications on resource allocation efficiency. The control variables are also expected to exert direct or indirect influences on the dependent variables. AMOS software is used for parameter estimation and goodness-of-fit testing of the model. The model’s rationality and validity are assessed based on fit indicators, and the proposed hypotheses are verified through path coefficients and significance tests. This analysis reveals the intrinsic mechanisms and pathways through which DL-based technological innovation governance influences intelligent resource allocation in high-tech industries. The final constructed model is shown in Fig. 1. Among them, H1 represents the relationship between DL application and allocation efficiency; H2 denotes the relationship between DL application and dynamic adaptability; H3 refers to the relationship between the governance system and allocation efficiency; H4 indicates the relationship between the governance system and dynamic adaptability.

The model structure.

Full size image

Case study

Upon completing the quantitative empirical analysis, 15 representative high-tech enterprises are selected for case studies. These enterprises span multiple industries and vary in size, levels of DL application, and technological innovation governance models. Field investigations are conducted at these enterprises, involving in-depth interviews with senior management, R&D directors, technical experts, and key personnel from relevant functional departments. The objective is to gain granular insights into the background, motivations, and implementation processes associated with the adoption of DL technologies in their technological innovation governance processes. Additionally, the study explores critical technological and managerial challenges, strategic responses, and the resulting innovation outcomes and economic benefits.

Data collection and parameter setting

This study constructs a comprehensive dataset to support experimental analyses. The dataset is primarily sourced from multiple public databases and internal enterprise data to ensure diversity and representativeness.

Standard & Poor’s (S&P) Global Market Intelligence Database: This database provides detailed financial information on publicly listed companies. Key financial indicators of high-tech enterprises, such as R&D expenditures, revenue, and asset data, are extracted. These indicators are crucial for understanding the enterprises’ financial inputs and scale in innovation activities, reflecting their investment in innovation resources. For example, R&D expenditure data can directly measure the financial resources allocated by enterprises to innovation projects, reflecting an enterprise’s commitment to technological advancement.

United States Patent and Trademark Office (USPTO) Database: This database offers extensive patent information. Patent application and authorization data from high-tech enterprises in relevant fields are collected, encompassing patent numbers, application dates, technology classification codes, and citation relationships. Patent data effectively reflects an enterprise’s innovation output and technological innovation capabilities. For instance, the number of patents in specific technological fields indicates an enterprise’s R&D achievements and competitive positioning within those areas.

Crunchbase Dataset: This dataset focuses on startup and emerging high-tech enterprises, offering information such as founding dates, industry classifications, funding rounds, and investor details. These data facilitate the analysis of innovation environments and growth trajectories of newly established high-tech enterprises. In particular, funding round data indirectly reflects the external resource support and market expectations for their innovation activities.

Additionally, this study establishes a data cooperation mechanism with seven sample enterprises to make the research results more accurate. It aims to obtain the full-volume operation logs of their innovation resource management systems from 2019 to 2023. These logs include non-public fields such as resource allocation decision flows, project progress tracking, and abnormal intervention records. After desensitization, this data is integrated with public data for modeling, which can enhance the granularity and credibility of enterprise-level practice analysis.

Data cleaning and preprocessing procedures are implemented to ensure data quality and completeness. Duplicate data is eliminated, while missing values are filled or corrected using appropriate interpolation methods. Furthermore, data standardization and normalization are applied to enhance consistency and compatibility for subsequent experimental analyses and model training. This multi-source dataset is systematically collected and processed. A robust analytical foundation is thereby established for accurately assessing the impact of DL-based technological innovation governance on intelligent innovation resource allocation in high-tech industries.

The experiments are conducted on a high-performance computing cluster comprising multiple compute nodes, each equipped with powerful Central Processing Units (CPUs) and GPUs. Specifically, the CPUs used are Intel Xeon Platinum processors, characterized by high clock speeds and multi-core capabilities; this enables them to handle complex data preprocessing and general computational tasks. The GPUs, such as NVIDIA Tesla V100 or higher-end models, are essential for accelerating the training and inference processes of DL models. Each GPU has a large memory capacity to store model parameters and intermediate results during training. The compute nodes are interconnected via high-speed InfiniBand networking technology, ensuring fast and seamless data transmission and communication between nodes. This hardware configuration efficiently handles large-scale datasets and supports the training of intricate DL architectures.

Model validation and result analysis

Cross-validation

To test the robustness of the SEM, this study adopts the K-fold cross-validation method (k = 5) to evaluate the goodness of fit. Cross-validation is employed to ensure the assessment results’ reliability and generalizability. The dataset is divided into k folds (k = 5). The model is trained k times, with each fold serving as the validation set in turn, while the remaining k-1 folds are used as the training set. Performance indicators are calculated for each fold, and the model’s final performance is the average of the k-fold performances. This method helps to reduce the impact of overfitting and provides a more accurate estimate of the model’s performance on unknown data. For the hyperparameter tuning process of the model, the study uses a Bayesian optimization framework instead of the traditional grid search. The learning rate ranges from [1e−5, 1e−2], the number of hidden layers ranges from [2, 6], and the Dropout rate ranges from [0.1, 0.5]. The convergence condition is met when the validation set loss decreases by less than 0.1% in consecutive iterations. Meanwhile, 120 groups of parameter combinations are evaluated in parallel on the NVIDIA A100 cluster.

The results of the cross-validation are presented in Fig. 2.

The results of the cross-validation.

Full size image

In Fig. 2, the F1 score, precision, recall, and accuracy remain consistently stable across all folds. The average accuracy reaches 0.83, indicating that the model has a high prediction accuracy level. Additionally, the precision, F1 score, and recall fall within an acceptable range, suggesting that the model effectively identifies positive instances and maintains balanced predictive performance. These results confirm the model’s robustness, reliability, and strong generalization ability.

Comparison with baseline models

It is compared with several baseline models to evaluate the DL model’s superiority. These baseline models include traditional statistical models (e.g., linear regression, decision trees) and classical machine learning models (e.g., support vector machines). The performance of the DL and these baseline models is assessed using the same evaluation indicators and dataset. The adopted DL model architecture includes 3 hidden layers (with 256, 128, and 64 neurons in each layer, respectively). The study uses the ReLU activation function and a Dropout rate of 0.1 to prevent overfitting. The training process employs the Adam optimizer (with a learning rate of 0.001), conducting iterative training for 200 epochs on the NVIDIA A100 cluster with a batch size of 64. The early stopping mechanism is set to terminate training when the validation set loss does not decrease for 5 consecutive times. All hyperparameters are determined through a Bayesian optimization framework.

Figure 3 displays the results of the comparison with the baseline models.

The results of the comparison with the baseline models.

Full size image

Figure 3 reveals that the DL model has a remarkable advantage in all evaluation indicators. Its accuracy, precision, recall, and F1 scores are all higher than those of the other baseline models, and it has a lower mean squared error (MSE). These results indicate that the DL model can make more precise predictions when handling innovation resource allocation data, demonstrating a stronger ability to handle complexity and provide accurate insights.

Sensitive analysis

A sensitivity analysis is conducted to investigate the model’s robustness and sensitivity. This analysis examines how variations in input variables or model parameters influence performance. For example, features used in the model, such as different types of innovation resources or external factors, can be altered. At the same time, changes in performance indicators can be observed. The results of the sensitivity analysis are exhibited in Table 2.

In Table 2, sensitivity analysis data demonstrate that incorporating external market demand factors significantly enhances various indicators related to input variables. The results highlight this variable’s positive effect on model prediction accuracy. In contrast, relying solely on human innovation resources or excluding financial innovation resources leads to a decline in performance. When adjusting hyperparameters, changes in the learning rate and the number of hidden layers affect model performance; Excessively high or too low learning rates and inappropriate numbers of hidden layers can cause fluctuations in key indicators. Carefully fine-tuning these parameters optimizes model performance, underscoring the critical role of precise parameter tuning in enhancing predictive accuracy.

Case study

To gain deeper insights into the practical application of DL-driven technological innovation governance in intelligent innovation resource allocation within high-tech industries, 15 representative high-tech enterprises are selected for case studies. These case studies further validate the reliability and generalizability of the previous quantitative research results while also uncovering unique mechanisms and practical experiences. The results of the case analysis are detailed in Table 3.

Table 3 suggests that enterprises with higher composite scores in the DL application depth/breadth and the completeness of technological innovation governance systems tend to perform better. For example, Enterprises 3 (new energy) and 11 (information technology) perform prominently in the DL application depth/breadth (scoring: 7) and the completeness of governance systems (scoring: 9.0/9.2). Their innovation resource allocation efficiency (0.78/0.80) and dynamic adaptability (0.85/0.88) are significantly higher than the sample averages (efficiency: 0.69, adaptability: 0.77). In-depth interviews with these two enterprises reveal that Enterprise 3 has shortened the new product R&D cycle by 30% through DL models. In contrast, Enterprise 11 has further reduced the market response time to less than 1/3 of the industry average with real-time resource scheduling algorithms. This result intuitively reflects the key role of DL methods in improving the effectiveness of resource allocation. In contrast, enterprises with relatively shallow DL applications or less well-developed technological innovation governance systems show slightly inferior performance in innovation resource allocation. This partially confirms the hypotheses regarding the relationships between variables in the previously constructed SEM. In other words, the DL application depth/breadth positively impacts the efficiency and dynamic adaptability of intelligent innovation resource allocation. Moreover, the completeness of the technological innovation governance system plays a moderating role, enhancing these positive effects. Additionally, the level of technological innovation activity and market competition intensity in different industries exert varying degrees of influence on enterprise innovation resource allocation. Thus, these can offer a reference for enterprises to formulate innovation resource allocation strategies based on their respective diverse industry environments. Through in-depth analysis of these case enterprises, common characteristics and best practices can be distilled from successful cases. For instance, it emphasizes the in-depth application of DL technology in multiple stages and establishes a comprehensive technological innovation governance system to cope with internal and external environmental changes. These insights offer practical, actionable guidance for high-tech enterprises, fostering the broader adoption and deeper integration of DL-driven technological innovation governance within the high-tech industry.

Robustness test

To enhance the rigor of causal inference, this study adopts the Difference-in-Difference (DID) method to design a quasi-natural experiment. The study utilizes the “AI Demonstration Enterprise” policy (implemented by a provincial science department in 2024) as an exogenous shock for causal identification. The treatment group comprises 10 enterprises mandated to adopt DL technology under this policy. The control group includes 10 matched enterprises (same industry/scale) not subject to the DL mandate. The final experimental results are detailed in Table 4.

Table 4 indicates that the coefficients of the core variable Treat × post are significantly positive in all three models (0.191–0.203, p < 0.01). This suggests that the innovation efficiency of enterprises in the treatment group increases by an average of 19.7–20.3% after the policy implementation. The results of the full and matched samples are highly consistent (R² are all greater than 0.68), and the dynamic effect test shows that the treatment effect is persistent. Among the control variables, the impact of enterprise size is not significant, indicating that the efficiency improvement mainly stems from the application of DL technology rather than scale factors. The control of industry and quarterly fixed effects effectively eliminates the impact of potential confounding factors, enhancing the reliability of causal inference.

This study examines the impact of DL-driven technological innovation governance on intelligent innovation resource allocation in the high-tech industry. By integrating multiple methods and selecting enterprise samples from various industries, the study constructs models and conducts detailed case studies to comprehensively explore these impacts.

Based on the experimental results, cross-validation shows that the model fit indicators are stable. This indicates that the DL-driven technological innovation governance framework can effectively explain the internal relationships of innovation resource allocation. This finding echoes the “dynamic governance response theory” proposed by Rane et al. When the depth of DL application increases by 1 unit, the dynamic adaptability of resource allocation also increases, confirming the key role of algorithmic complexity in coping with market fluctuations. Compared with traditional methods, the DL-SEM model has a lower innovation resource mismatch rate, an advantage derived from its non-linear feature capture capability. As Sharifani and Amini stated, DL could identify the implicit collaborative relationships between human capital resources by integrating and analyzing patent knowledge graphs and financial time-series data. However, traditional models often ignore such complex interaction effects.

Sensitivity analysis illustrates that external market demand factors substantially enhance model performance. Market demand serves as a crucial guide for the allocation of innovation resources. When market demand exhibits a clear definition and stability, enterprises can allocate innovation resources more effectively. This enables the model to learn and predict the relationship between resource allocation and output. Financial investment in innovation resources is indispensable; hyperparameter tuning also exerts a significant impact, underscoring the importance of precise adjustments. Sufficient funding ensures the smooth implementation of innovation activities, providing the model with richer data support. In addition, appropriate hyperparameter tuning optimizes the model’s structure and performance, allowing it to better adapt to data and task requirements. Case studies demonstrate that enterprises with a high level of DL application and a well-established technological innovation governance system exhibit superior resource allocation efficiency and dynamic adaptability. These case studies validate the hypothesized relationships between variables. These enterprises effectively leverage DL technology to extract data value, while a robust governance system ensures the scientific rigor and effectiveness of resource allocation decisions. This enables enterprises to respond rapidly to market changes and optimize resource allocation. Additionally, industry environmental factors exert multifaceted influences, and the common characteristics of successful enterprises provide valuable insights for others. Industry-specific factors such as market competition intensity and the pace of technological advancement vary across sectors, impacting enterprises’ innovation resource allocation strategies and outcomes. The effective measures implemented by successful enterprises to address these environmental factors offer useful references for others.

In the practical allocation of innovation resources within the high-tech industry, the powerful data-processing abilities of DL can uncover latent connections within innovation-related data. For example, DL can provide precise strategies for resource allocation by analyzing vast datasets on market trends, technological advancements, and enterprise capabilities. However, variations in data quality and distribution can undermine model stability, necessitating rigorous data management by enterprises. Low-quality data may contain errors or missing values, while imbalanced data distribution can lead to learning biases, ultimately affecting model stability and prediction accuracy. A well-developed technological innovation governance system can complement DL applications by ensuring effective resource allocation through strategic planning, organizational optimization, resource integration, and risk management. Governance frameworks provide directional guidance and structural support for DL applications, enhancing resource allocation efficiency through synergy. However, the inherent characteristics of DL-based decision-making limit its applicability in certain contexts, highlighting the need for methods to improve interpretability. In scenarios that demand high decision-making transparency, such as major investment decisions or policy formulation, the DL model’s black-box nature poses a significant barrier to adoption, necessitating urgent solutions. Moreover, the level of technological innovation activities and the intensity of market competition vary across industries. Enterprises adjust their DL applications and governance frameworks based on specific conditions and environments. Only through such adaptations can enterprises fully leverage DL’s advantages, optimize governance systems, and achieve intelligent and efficient allocation of innovation resources. This, in turn, fosters the sustainable development and competitiveness of the high-tech industry.

Compared with similar studies in the literature, the uniqueness of this study lies in its comprehensive analysis, incorporating multi-industry samples and multiple methodological approaches. It not only validates the advantages of DL in innovation resource allocation but also delves into the challenges encountered during its application and explores potential solutions. Previous studies may have focused on a single industry or a single methodology, lacking sufficient discussion on the intricate issues arising in real-world applications. This study enriches the existing literature by offering more practice-oriented insights. It highlights critical considerations such as data management and governance system synergy when applying DL for innovation resource allocation. Future research in this field could concentrate on improving the interpretability of DL models, thus exploring ways to enhance decision-making transparency without compromising model performance. Moreover, further investigations into how industry-specific characteristics influence DL applications and governance frameworks could provide enterprises with more tailored solutions.

This study has achieved a series of valuable contributions in DL-driven technological innovation governance and intelligent innovation resource allocation in the high-tech industry. From a theoretical perspective, it establishes a systematic research framework that clearly defines core concepts and their interrelationships, laying a solid theoretical foundation for subsequent studies. Utilizing SEM, this study provides an in-depth analysis of the intrinsic mechanisms through which DL influences innovation resource allocation, significantly enriching the theoretical landscape in this domain. By offering a clearer understanding of this intricate process for both academia and industry, the study provides a clear direction for future research. This illustrates that scholars can build upon these findings to explore the relationship between DL and innovation resource allocation from additional dimensions and at greater depth, refining the theoretical framework.

From a methodological perspective, the study employs a research design that combines stratified sampling with empirical and case studies, ensuring the reliability and generalizability of the findings. The carefully designed multidimensional variable measurement system serves as a valuable reference for future research in this field. The establishment of this methodological framework provides a more scientific and standardized foundation for data collection, analysis, and variable measurement in subsequent studies. It contributes to the standardization and scientific advancement of research methodology in this domain.

At the practical application level, this study offers actionable recommendations for high-tech enterprises. The analysis identifies key success patterns among enterprises excelling in DL applications and innovation governance, including multi-stage deep integration of technology, comprehensive strategic planning, and optimized organizational structures. These empirically observed best practices enable enterprises to optimize innovation resource allocation and bolster market competitiveness through structured adoption. For policymakers, these findings offer critical insights to inform policy formulation, supporting intelligent innovation resource allocation and the sustainable development of the high-tech industry. The research outcomes demonstrate measurable operational impacts for enterprise management and actionable insights for regulatory policy formulation. These contributions provide renewed developmental impetus to the high-tech industry while strengthening its competitiveness in the global market.

However, this study has certain limitations. Regarding data sources, public databases may not accurately reflect internal enterprise details, and access to proprietary corporate data is subject to various restrictions. This may introduce biases in the accuracy and completeness of the findings. The limited interpretability of the model makes intricate decision-making processes difficult to understand, reducing trust among enterprise managers and constraining broader adoption. Additionally, the scope of industries covered in this study is relatively limited. The high-tech industries cover most core fields of the national strategic emerging industries directory. However, emerging fields such as quantum computing and brain-computer interfaces have not yet been included. Future research can be conducted in three main directions to address these limitations. First, optimizing data collection and processing by strengthening collaborations with a broader range of enterprises enables access to more accurate data. Integrating and cleaning data using big data technologies can further enhance data quality. Second, improving model interpretability by developing algorithms or tools, such as feature visualization techniques, increases transparency in the decision-making process. Third, the high-tech industry has expanded into other fields. The resource allocation path from technology incubation to industrialization has been tracked vertically to verify the DL governance model’s migration ability in innovation scenarios. Thus, it provides more comprehensive and in-depth support for innovation and development in the high-tech industry. Through these follow-up studies, the proposed framework in this field can be further improved, offering stronger support for the continued advancement of the high-tech industry.

DL:

Deep learning

SEM:

Structural equation model

S&P:

Standard & Poor’s

USPTO:

United States Patent and Trademark Office

CPU:

Central processing unit

GPU:

Graphics processing unit

1. Moss, E. & Metcalf, J. High tech, high risk: Tech ethics lessons for the COVID-19 pandemic response. Patterns 1(7), 1 (2020).

   Google Scholar 

2. Ding, S., Tao, Z. & Hu, J. Forecasting the economic indices of the high-tech industries in China using the grey multivariable convolution model. Appl. Soft Comput. 126, 109301 (2022).

   Google Scholar 

3. Sahar, T. et al. Anomaly detection in laser powder bed fusion using machine learning: A review. Results Eng. 17, 100803 (2023).

   CAS  Google Scholar 

4. Tang, C. et al. Deep learning in nuclear industry: A survey. Big Data Min. Anal. 5(2), 140–160 (2022).

   Google Scholar 

5. Zhou, H. et al. An improved grey multivariable time-delay prediction model with application to the value of high-tech industry. Expert Syst. Appl. 213, 119061 (2023).

   Google Scholar 

6. Luo, B. A method for enterprise network innovation performance management based on deep learning and internet of things. Math. Probl. Eng. 2022(1), 8277426 (2022).

   Google Scholar 

7. Jha, S. B. & Babiceanu, R. F. Deep CNN-based visual defect detection: Survey of current literature. Comput. Ind. 148, 103911 (2023).

   Google Scholar 

8. Guerzoni, M., Nava, C. R. & Nuccio, M. Start-ups survival through a crisis. Combining machine learning with econometrics to measure innovation. Econ. Innov. New Technol. 30(5), 468–493 (2021).

   Google Scholar 

9. Mikhaylov, A. & Grilli, M. L. Machine learning methods and sustainable development: Metal oxides and multilayer metal oxides. Metals 12(5), 836 (2022).

   Google Scholar 

10. Mukhamediev, R. I. et al. Review of artificial intelligence and machine learning technologies: Classification, restrictions, opportunities and challenges. Mathematics 10(15), 2552 (2022).

    Google Scholar 

11. Kiangala, S. K. & Wang, Z. An effective adaptive customization framework for small manufacturing plants using extreme gradient boosting-XGBoost and random forest ensemble learning algorithms in an industry 4.0 environment. Mach. Learn. Appl. 4, 100024 (2021).

    Google Scholar 

12. Guliyev, H. Artificial intelligence and unemployment in high-tech developed countries: New insights from dynamic panel data model. Res. Glob. 7, 100140 (2023).

    Google Scholar 

13. Tikhonov, A. I., Sazonov, A. A. & Kuzmina-Merlino, I. Digital production and artificial intelligence in the aircraft industry. Russ. Eng. Res. 42(4), 412–415 (2022).

    Google Scholar 

14. Rane, N. L., Kaya, O. & Rane, J. Artificial intelligence, machine learning, and deep learning technologies as catalysts for industry 4.0, 5.0, and society 5.0. Artif. Intell. Mach. Learn. Deep Learn. Sustain. Ind. 5, 2–2 (2024).

    Google Scholar 

15. Capatina, A., Bleoju, G. & Kalisz, D. Falling in love with strategic foresight, not only with technology: European deep-tech startups’ roadmap to success. J. Innov. Knowl. 9(3), 100515 (2024).

    Google Scholar 

16. Chien, C. F., Lin, Y. S. & Lin, S. K. Deep reinforcement learning for selecting demand forecast models to empower industry 3.5 and an empirical study for a semiconductor component distributor. Int. J. Prod. Res. 58(9), 2784–2804 (2020).

    Google Scholar 

17. Luo, S. & Choi, T. M. Great partners: How deep learning and blockchain help improve business operations together. Ann. Oper. Res. 339(1), 53–78 (2024).

    MathSciNet  Google Scholar 

18. Zeng, C., Zhang, F. & Luo, M. A deep neural network-based decision support system for intelligent geospatial data analysis in intelligent agriculture system. Soft. Comput. 26(20), 10813–10826 (2022).

    Google Scholar 

19. Pajpach, M. et al. Low-cost education kit for teaching basic skills for industry 4.0 using deep-learning in quality control tasks. Electronics 11(2), 230 (2022).

    Google Scholar 

20. Guan, X. et al. Construction of science and technology achievement transfer and transformation platform based on deep learning and data mining technology. J. Intell. Fuzzy Syst. 39(2), 1843–1854 (2020).

    Google Scholar 

21. Polini, A. & Moroni, L. The convergence of high-tech emerging technologies into the next stage of organ-on-a-chips. Biomater. Biosyst. 1, 100012 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

22. Du, L. et al. Asymmetric effects of high-tech industry and renewable energy on consumption-based carbon emissions in MINT countries. Renew. Energy 196, 1269–1280 (2022).

    CAS  Google Scholar 

23. Peng, J. et al. Machine learning and deep learning modeling and simulation for predicting PM2.5 concentrations. Chemosphere 308, 136353 (2022).

    CAS  PubMed  Google Scholar 

24. Azeem, M., Haleem, A. & Javaid, M. Symbiotic relationship between machine learning and industry 4.0: A review. J. Ind. Integr. Manag. 7(03), 401–433 (2022).

    Google Scholar 

25. Zhu, D. J. et al. Knowledge graph and deep learning based pest detection and identification system for fruit quality. Internet Things 21, 100649 (2023).

    Google Scholar 

26. Lo, F. T. H. The paradoxical transparency of opaque machine learning. AI Soc. 39(3), 1397–1409 (2024).

    Google Scholar 

27. Ferrati, F. & Muffatto, M. Entrepreneurial finance: Emerging approaches using machine learning and big data. Found. Trends® Entrep. 17(3), 232–329 (2021).

    Google Scholar 

28. Rane, N. L. et al. Emerging trends and future research opportunities in artificial intelligence, machine learning, and deep learning. Artif. Intell. Ind. Soc. 5, 2–96 (2024).

    Google Scholar 

29. Kinne, J. & Lenz, D. Predicting innovative firms using web mining and deep learning. PLoS ONE 16(4), e0249071 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

30. Ma, Q., Li, H. & Thorstenson, A. A big data-driven root cause analysis system: Application of machine learning in quality problem solving. Comput. Ind. Eng. 160, 107580 (2021).

    Google Scholar 

31. Hunt, J. O. S., Myers, J. N. & Myers, L. A. Improving earnings predictions and abnormal returns with machine learning. Account. Horiz. 36(1), 131–149 (2022).

    Google Scholar 

32. Verma, S. & Singh, V. Impact of artificial intelligence-enabled job characteristics and perceived substitution crisis on innovative work behavior of employees from high-tech firms. Comput. Hum. Behav. 131, 107215 (2022).

    Google Scholar 

33. Rahaman, M. M. et al. Machine learning in business analytics: Advancing statistical methods for data-driven innovation. J. Comput. Sci. Technol. Stud. 5(3), 104–111 (2023).

    Google Scholar 

34. Rane, N. L. et al. Artificial intelligence, machine learning, and deep learning for advanced business strategies: A review. Partn. Univers. Int. Innov. J. 2(3), 147–171 (2024).

    Google Scholar 

35. Sharifani, K. & Amini, M. Machine learning and deep learning: A review of methods and applications. World Inf. Technol. Eng. J. 10(7), 3897–3904 (2023).

    Google Scholar 

36. Mishra, N. K., Mishra, N. & Sharma, P. P. Unraveling the relationship between corporate governance and green innovation: A systematic literature review. Manag. Res. Rev. 48(5), 825–845 (2025).

    Google Scholar 

37. Eom, T., Woo, C. & Chun, D. Predicting an ICT business process innovation as a digital transformation with machine learning techniques. Technol. Anal. Strateg. Manag. 36(9), 2271–2283 (2024).

    Google Scholar 

38. Sujitha, B. et al. Optimal deep learning based image compression technique for data transmission on industrial internet of things applications. Trans. Emerg. Telecommun. Technol. 32(7), e3976 (2021).

    Google Scholar 

39. Rajesh, S. et al. Detection of features from the internet of things customer attitudes in the hotel industry using a deep neural network model. Meas. Sens. 22, 100384 (2022).

    Google Scholar 

40. Parvez, M. O. Use of machine learning technology for tourist and organizational services: High-tech innovation in the hospitality industry. J. Tour. Futur. 7(2), 240–244 (2021).

    MathSciNet  Google Scholar 

41. Shamim, M. M. I. Artificial intelligence in project management: Enhancing efficiency and decision-making. Int. J. Manag. Inf. Syst. Data Sci. 1(1), 1–6 (2024).

    MathSciNet  Google Scholar 

42. Buhmann, A. & Fieseler, C. Deep learning meets deep democracy: Deliberative governance and responsible innovation in artificial intelligence. Bus. Ethics Q. 33(1), 146–179 (2023).

    Google Scholar 

Download references

Authors and Affiliations

1. School of Humanities and Law, Northeastern University, Shenyang, 110169, China

   Xiaoxuan Yu

2. School of Marxism, Northeastern University, Shenyang, 110169, China

   Zhen Chen

3. CFHI Innovation Research Institute Beijing, Beijing, 100071, China

   Xinyu Zhang

Contributions

Xiaoxuan Yu: Conceptualization, methodology, software, validation, formal analysis, investigation, resources, data curation, writing—original draft preparation Zhen Chen: writing—review and editing, visualization, supervision, project administration, funding acquisition Xinyu Zhang:methodology, software, validation, formal analysis.

Corresponding author

Correspondence to Zhen Chen.

Competing interests

The authors declare no competing interests.

Ethics approval

This article does not contain any studies with human participants or animals performed by any of the authors. All methods were performed in accordance with relevant guidelines and regulations.

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions