1. Introduction
With the strengthening trend of economic globalization and the advent of the era of knowledge economy, technological innovation has not only become a means for enterprises to make profits, but also the basis and driving force for their survival and development. High innovation efficiency motivates enterprises to continuously develop new technologies and products in line with market demand, which increases the added value of products and brings good economic benefits to enterprises [
1]. Meanwhile, with improved economic efficiency, enterprises will be able to increase their innovation investment, continue to improve their innovation efficiency, and gain a sustainable and competitive technological advantage [
2,
3]. As for technological innovation, it is closely related to human resources, capital investment, institutional protection and industrial structure. Therefore, technological innovation is a multi-input and multi-output process by using internal and external innovation resources such as human, financial, material, and institutional [
4,
5], and the process is dynamic and regionally heterogeneous. In the increasingly fierce business environment, how enterprises use advanced and intelligent methods, means that obtaining the optimal path for innovation and development is always a research topic in the field of national technology and economic strategy.
Current research on the efficiency of technological innovation focuses on two aspects: measurement methods and impact factors, as shown in
Table 1. Existing results on the efficiency measurement of technological innovation have been synthesized, and it was found that two basic measures are mostly used. One is mathematical programming-based data envelope analysis, which mainly applies a set of multiple input and output observations to estimate effective production fronts. The other is stochastic frontier analysis, which is developed based on knowledge production functions that require setting the production–function relationship between inputs and outputs in advance and using regression analysis and other methods to estimate the parameters of the production function. Moreover, in order to effectively promote the evolution of corporate innovation, scholars have explored the factors that influence the efficiency of corporate innovation at different levels and perspectives.
While existing studies have provided theoretical guidance and practical insights for technological innovation, further improvements are still needed. On the one hand, existing studies typically use traditional regressions to explore the net effect arising from a single variable, whereas changes in the efficiency of technological innovation are the result of multiple factors acting together. Therefore, when examining the path of efficiency improvements of technological innovation, not only should the degree of influence and direction of actionable individual factors be considered, but also focusing and mastering the joint effect of multiple influencing factors. On the other hand, enterprises are located in different innovation environments, and the direction and degree of influence exerted by factors may also produce differences. In the case of China’s eastern, central and western regions, the eastern region of the country, with its superior market environment and a high level of development capability factor, has moved away from its previous crude economic growth model, which was dominated by traditional factor inputs. Innovative activities are market-driven, and science and technology innovation has become a major engine of economic development. However, some provinces in Central and Western China are still falling behind in infrastructure construction, opening up to the outside world and having relatively low marketization. Although the innovation environment in the central and western regions has been improved by the policies of “Rise of Central China” and “Western Development”, it still has a large gap with the eastern regions. Therefore, regional heterogeneity should also be included in the study of technological innovation efficiency improvement, and regional innovation efficiency should also be improved according to local conditions.
To make up for the shortcomings of existing studies, we aim to dig deeper into the path of technological innovation efficiency enhancement in China’s high-tech industries. A qualitative approach is introduced to examine the joint effects of antecedent conditions, such as the level of economic development, human capital, government support, industrial structure, the level of openness to the outside world, the level of intellectual property protection, and MNC innovation on the efficiency of technological innovation. A neural network prediction model is used to explore the optimal path of innovation efficiency enhancement in China’s high-tech industries and to provide theoretical support for promoting China’s technological progress. The main contributions of this paper include: firstly, the fuzzy-set qualitative comparative analysis (fs/QCA) method, which is used to deeply explore the innovation efficiency enhancement paths of China’s high-tech industries and enrich the theoretical explanation of the antecedent conditions of technological innovation efficiency; and secondly, utilizing the Applying Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) to optimize the Back Propagation (BP) neural network to identify the optimal paths of innovation efficiency enhancement in China’s high-tech industries, which will open up new horizons in China’s technology innovation research, enrich and improve the methods and tools of China’s regional innovation development research, and provide support and suggestions for promoting China’s high-quality science and technology development.
The remainder of the paper is organized as follows:
Section 2 is a literature review and theoretical framework, which mainly describes the mechanism of action between antecedent conditions and technological innovation efficiency;
Section 3 is the research design, which mainly uses the QCA method and GA-PSO-BP neural network to build the path optimization model;
Section 4 is empirical to identify the improvement path of the innovation efficiency of China’s high-tech industry and the optimal path of each region;
Section 5 provides further discussion; and finally, the conclusion and suggestion are provided in
Section 6.
3. Study Design
3.1. Research Methodology
3.1.1. QCA Method
The antecedents of technological innovation efficiency do not function in isolation, but will be correlated and affect each other; hence, the fs/QCA approach was chosen to construct the factor-linkage model of technological innovation efficiency in this study. QCA was explicitly proposed by sociologist Regin in 1987 as a case-oriented research method that integrates the advantages of qualitative and quantitative analysis. Fiss used it in 2007 for empirical analysis, and since then the QCA method has been widely used in solving management problems [
36]. QCA methods include clear-set qualitative comparative analysis (cs/QCA), multi-value set qualitative comparative analysis (mv/QCA), and fuzzy-set qualitative comparative analysis (fs/QCA). The cs/QCA and mv/QCA require dichotomous or multivariate segmentation of the variable data, which over-emphasizes the differences between groups and may indirectly lose a large amount of sample information, leading to serious outcome errors. Moreover, mv/QCA is difficult to handle continuous data, so these two methods are only suitable for specific studies. Compared with cs/QCA and mv/QCA, fs/QCA is good at dealing with analytical problems that are not dichotomous, defining the degree of affiliation of each variable in the range of [0,1]. It exhibits the characteristics of ambiguity and is suitable for continuous data studies, where the state of the variables can be more precisely described and the objectivity and confidence of the results can be improved. In this study, the fs/QCA method was used to select the innovation efficiency of China’s high-tech industries as the outcome variable, and the level of economic development, government support, human capital, industrial structure, innovation of multi-national corporations, openness and intellectual property protection as the conditional variables.
3.1.2. BP Neural Network
Artificial neural network models are mathematical representations formed by abstracting the information transfer of biological neurons using mathematical language. It has successfully solved many real-world problems in the fields of pattern recognition, intelligent robotics, automatic control, biology, and economics, and demonstrated powerful intelligent properties. BP neural network is a classical model of artificial neural network, which is a multilayer feed-forward neural network model. This neural network model uses the error backpropagation algorithm to solve the learning problem in the connection pairs of the hidden layers of a multilayer neural network [
37]. BP neural networks can be used to implement arbitrary nonlinear mapping from input to output by varying the network weights, layers, and nodes. The basic idea is the gradient descent method, which uses gradient search techniques to achieve the minimum mean-squared error between the actual output value and the desired output value of the network [
38]. The implementation process of BP neural network learning algorithm mainly includes the following three parts: firstly, the input sample forward propagation process. Here, the sample data is passed from the input layer through the implicit layer to the output layer, and the corresponding network output values are calculated based on the input samples. The sample data is passed from the input layer through the hidden layer to the output layer, and the corresponding network output value is computed based on the input samples. Secondly, output error back propagation process. Here, the sample data is forward propagated to obtain the output results, and the network weights and thresholds are corrected for learning by using the error between the output results and the expected ones. Thirdly, cyclic memory training process. Here, the input sample forward propagation procedure and the output error backward propagation procedure are repeated until the output error reaches the desired accuracy value of the ensemble. The specific procedure of the BP neural network is illustrated in
Figure 2.
3.1.3. GA-PSO Algorithm
GA is a computational model of a biological evolutionary process that mimics the mechanism of natural selection and genetics of biological evolution [
39]. It is an optimization algorithm that finds an optimal solution by simulating a natural evolutionary process, and is widely used in combinatorial optimization, machine learning, and signal processing. PSO is a global stochastic search algorithm based on simulating the intelligent behavior of a flock of birds in the process of predation [
40]. The GA-PSO algorithm is a hybrid algorithm formed by introducing the GA algorithm into the PSO algorithm. The GA algorithm has the advantage of having excellent global search ability, does not easily get stuck in local extrema, has good scalability, and can be mixed with other algorithms; however, it also has the drawback of having low local search ability and being prone to premature convergence. The PSO algorithm has the advantages of fast convergence, high generality, and small parameter settings, but it also has the drawbacks of easily conducting population convergence earlier than normal and having a slow convergence later than normal [
41]. Due to the scalability of the GA algorithm, the genetic operation operator in its algorithm is brought into the PSO algorithm to perform crossover operations and variation operations on the position and velocity vectors of the particles during the iterative process of the particle swarm, which helps maintain the diversity of the population and the population jumping out of the local extrema, and also greatly improves the global search performance of the population.
3.1.4. GA-PSO-BP Neural Network
The BP neural network has outstanding merits and can effectively solve the problem of adjusting the weights and thresholds of multilayer feed-forward neural networks. It features massively parallel processing, strong fault tolerance, and distributed storage [
42]. With a complete theoretical foundation and successful application cases, it implements a nonlinear mapping function from the input layer to the output layer and has become a hot topic for applications in several research areas. However, BP neural networks still have drawbacks, such as slow convergence, a tendency to get stuck in local extrema, and a tendency to overfit during training [
43]. As a result, a variety of improvements have emerged. For example, the integration of the additional momentum method and the LM (Levenberg–Marquardt) algorithm in the training of BP neural networks is very effective in improving the shortcomings of traditional neural BP. In addition, there are also optimization learning algorithms, such as the PSO algorithm (Partical Swarm Optimization) which is applied to the learning process of BP neural networks; however, the PSO algorithm still has shortcomings, such as low solution accuracy and easy scattering, so the performance of PSO-BP neural networks still needs to be improved. The GA and PSO algorithm were combined to speed up the convergence of the algorithm and solve the problem of early convergence [
44]. Therefore, the GA-PSO hybrid algorithm is used in this study to optimize the performance of BP neural networks. The GA-PSO-BP neural network algorithm is used to construct an improved path optimization model for technological innovation effectiveness. The flow chart of the GA-PSO-BP neural network algorithm is shown in
Figure 3.
3.2. Sample Selection and Data Sources
Based on the availability of data, 30 provinces in China were selected for this study (more data were missing in Tibet, so they were excluded). Regional heterogeneity is taken into account by dividing the 30 provinces in the sample into three regions: Eastern, Central and Western, according to the way the National Bureau of Statistics of China divides them. The eastern region specifically includes Beijing, Fujian, Guangdong, Hainan, Hebei, Jiangsu, Liaoning, Shandong, Shanghai, Tianjin and Zhejiang. The central region includes Anhui, Henan, Heilongjiang, Hubei, Hunan, Jilin, Jiangxi and Shanxi provinces. The western region includes Guangxi, Guizhou, Shaanxi, Sichuan, Yunnan, Chongqing, Inner Mongolia, Gansu, Qinghai, Ningxia and Xinjiang. The data on innovation efficiency, economic development level, human capital, government support, industrial structure, multinational corporation innovation, openness level and intellectual property protection level of China’s high-tech industries are mainly obtained from the “China High-tech industries Statistical Yearbook”, “China Industrial Statistical Yearbook”, “China Statistical Yearbook” and statistical yearbooks of each province. Since the statistical method of “China High-tech industries Statistical Yearbook” changed after 2008, this study selects 2009-2020 as the research period in consideration of the accuracy of the data.
3.3. Variable Definition and Measurement
The specific measures of antecedent conditions and outcome variables for this study are listed below. Innovation efficiency of China’s high-tech industries (IE). Data related to domestic enterprises and state-owned enterprises in China’s high-tech industries were selected to measure the innovation efficiency of China’s high-tech industries. The innovation efficiency of China’s high-tech industries is measured using the ultra-efficient SBM model with full-time equivalent research and development personnel, internal R&D funding and new product development funding as input indicators, and the number of patent applications and valid inventions and new product sales revenue as output indicators. Economic development (ED): Gross domestic product (GDP) per capita is used as a proxy variable to measure the level of economic development, which can reflect the state of economic development and potential of a region. Human capital (HC): Human capacity directly affects innovation efficiency, as well as the ability of innovation agents to externally learn, assimilate and digest new technologies. Thereby, this can indirectly affect innovation efficiency. This study expresses human capital in terms of the ratio of the number of researchers to the number of employed people. Government support (GS): The higher the indicator, the more favorable it is for improving the motivation of the subject to carry out innovative activities. In this study, the percentage of government funding in R&D grants was used to indicate the strength of government support. Industrial structure (IS): Industrial structure is an important influence on intra-regional efficiency, as measured by the ratio of tertiary industry output to regional GDP. MNC innovation (MNCI): MNC innovation has helped activate China’s regional innovation system and has had a major impact on corporate technology innovation. The innovation efficiency of MNCs in China was chosen to represent MNC innovation. Openness (OP): The high level of foreign development is conducive to improving the efficiency of foreign exchange and resource factor accretion, which has a significant impact on scientific and technological innovation. This study uses the ratio of total imports and exports to regional GDP to express the level of openness. Intellectual property protection (IPP): The intellectual property protection system is an effective guarantee against malicious infringement of knowledge and technology and helps to increase the innovation-driven effect. In this study, the share of technology market transactions to regional GDP is used to express the level of intellectual property protection.
6. Conclusions and Recommendations
6.1. Conclusions
This study uses the fs/QCA method to excavate the multi-factor innovation efficiency enhancement path of China’s high-tech industry, which enriches the theoretical results of the antecedent conditions of technological innovation efficiency, deepens the systematic study of technological innovation efficiency, and extends the dimension and depth of technological innovation efficiency research. The optimization model of technological innovation efficiency enhancement in China’s high-tech industries is constructed based on GA-PSO-BP neural network. According to the group-path warp scenario setting and simulation, the optimal path for innovation efficiency improvement of high-tech industries in the eastern, central and western regions of China is obtained. To develop innovation management strategies for China’s high-tech industries based on the level of regional economic development and regional endowment characteristics, it is of great practical significance to efficiently utilize innovation resources and enhance the innovation capacity of localities and enterprises, and help to promote sustainable progress in the innovation efficiency of China’s high-tech industries, which in turn promotes high-quality development of regional innovation in China. The findings of the study include the following main conclusions:
- (1)
The improvement of technological innovation efficiency is revealed as a con-current mechanism of multi-factor combination through a configuration perspective. There is no single condition among economic development, government support, human capital, industrial structure, MNC innovation, openness or intellectual property protection that is necessary to promote the efficiency of innovation in China’s high-tech industries. Only an effective combination of conditions can achieve the goal of increasing the level of innovation efficiency in China’s high-tech industries.
- (2)
Four configuration paths of innovation efficiency improvement in China’s high-tech industries were identified by applying the fs/QCA method. They are: the “MNC innovation—economic development—government support” linkage type, the “MNC innovation—government support” linkage type, the “Economic development—government support” linkage type, and the “Economic development” driven type.
- (3)
Regional heterogeneity makes differences in the optimal paths of innovation efficiency improvement in high-tech industries in Eastern, Central and Western China. The “MNC innovation—economic development—government support” linkage type has the best effect on improving the efficiency of innovation in high-tech industries in Eastern China. The “MNC innovation—government support” linkage type has the best effect on improving the efficiency of innovation in high-tech industries in Central China. The “Economic development” driven type has the best effect on improving the efficiency of innovation in high-tech industries in Western China.
6.2. Recommendations
The innovation management strategies of China’s eastern, central and western regions cannot be generalized. Combined with the region’s own resource conditions and development stage, a corresponding management policy for the improvement of enterprises’ innovation and self-sufficiency should be formulated. This will lead to a coordinated, balanced and stable development of Chinese innovation in the region. Therefore, the following recommendations for regional differentiation management are presented in this study.
- (1)
Eastern China has a higher level of economic development, human capital, and intellectual property protection. The government’s financial support for the eastern region is stronger, and the attraction for R&D investment by MNCs is stronger. Therefore, more attention should be paid to absorbing and digesting MNC innovations and enhancing independent innovation capabilities to promote the high-quality development of regional innovation. As economic development and government policies change, the advantages of high openness and robust infrastructure in the eastern region will gradually diminish, and the role of human capital will gradually come to the forefront. Therefore, the eastern region should focus on introducing MNCs with a large proportion of R&D-oriented, high-tech and highly educated workers. More attention should be paid to the cultivation of high-quality talent, the development of corresponding talent introduction strategies, and the reduction of talent consumption costs. In this way, we can attract and retain talent and enhance our human capital edge. Eastern regions should break regional monopolies, cross-regional information exchange on innovation should be strengthened, local attributes of innovation spillover should be reduced, geographical radius of innovation spillover should be increased, and coordinated and balanced development of innovation in China’s regions should be promoted.
- (2)
Although Central China is weaker than the eastern region in all aspects, the economy, capital and technology are in the stage of rapid development, which belongs to the state that should focus on improving learning, absorbing and imitating advanced foreign technology. The government should give appropriate policy support and implement favorable policies to attract multinational companies to the central region, so that the central region can get more financial support for research and development, and cultivate and import higher-quality technical talent. The central region should continue to expand the size of the market, tap the market potential and make the trading market system robust. Market information exchange network platforms, innovation achievement transformation platforms and intermediary service agencies should be constructed. High-tech industries in the central region should accelerate the conversion of the innovation model based on “imitation innovation” to the innovation model based on “independent innovation”. Self-competitive advantages should be enhanced and industries with distinctive advantages should be formed. Meanwhile, we should strengthen the docking cooperation with the innovation main body in the eastern and undertake the quality industries transferred from the eastern. The eastern region plays its radiating and leading role and can realize the rapid improvement of innovation efficiency of the high-tech industries in the central region.
- (3)
Compared with the eastern and central regions, the western region has a lower level of economic development, an incomplete talent pool, and an imperfect construction of project support facilities and the innovation policy system. Meanwhile, the western region is opening up to the outside world later and in a slower process. Therefore, the western region belongs to the phase of improving the innovation environment and enhancing the technological learning capability of high-tech industries. The western region should strengthen hard environmental conditions and improve the environment for investment in research and development as soon as possible. The construction of transport and its communication infrastructure should be accelerated. Regional aviation networks should be constructed, and telecommunications and information networks should be improved between the western and central regions and abroad. This way, we can build convenient platforms for research, production and operations needed by enterprises. The learning and absorption capacity of the western region should be enhanced. Governments at all levels should continue to invest more in education and the introduction of talent, and advocate for senior talent and university students to go to the western, the grassroots and places that are struggling in terms of development commitment and progress. Local enterprises should actively organize training and conduct regular exchange seminars with MNCs and high-tech industrial enterprises in the central and eastern. This allows local enterprises to learn and summarize new technologies, knowledge and ideas that advanced enterprises have. Resource advantages were translated into new economic development advantages, and efforts were made to enhance their own level of innovation and development capacity.
6.3. Limitations and Prospects
There are still some limitations in this paper. (1) Based on the availability of data, this paper examines the innovation efficiency enhancement path of China’s high-tech industries using provincial data. The research object can be expanded in the future, for example, the research can be carried out from Chinese cities and manufacturing industries, and the research can also be deepened from the perspective of high-tech industry subdivision, which will help to explore the path of technological innovation efficiency improvement more deeply and make the research conclusion more scientific and comprehensive. (2) In this paper, the research sample is divided into three regions in the eastern, central and western parts of China according to the way the China Bureau of Statistics is divided. According to the physical geography and economic development level, China’s regions can also be divided into eight economic geographical regions, including northeast, north, central, southwest, central and south, east, south and northwest. Meanwhile, China has three major economic circles, including the Yangtze River Delta Economic Circle, the Pearl River Delta Economic Circle and the Bohai Sea Economic Circle. Future research can be conducted from the above eight economic geographic regions or three major economic spheres to expand the breadth and depth of research.