The digital economy is a concept closely related to digitalization, which is the development and extension of the information economy. Machlup (1962) realized the particularity of information products and services earlier and pointed out that enterprises providing information products or services to the market were an important economic sector. Although his perception of the information economy has exceeded the technical level, the essence of the information economy has not yet been explained. Don Tapscott (1996) studied the effect of information technology on the economy and formally proposed the concept of a digital economy for the first time. The Group of Twenty Summit (2016) pointed out that the essence of the digital economy is a series of economic activities that take digital knowledge and information as key production factors, modern information networks as the important carriers, and information and communication technology as a main driving force for efficiency improvement and economic structure optimization. Xu, Ge, Wang & Skare, 2021 used bibliometric analysis methods and visualization software to explore technology adoption, and found that technology adoption has a variety of applications in digital fields such as online shopping, mobile learning, mobile banking, and e-learning, which shows that the integration of digital technology and traditional field has become an important development trend. The overall digitalization of China is in the middle of the range globally, but China has emerged as a global leader in some key digital industries such as e-commerce, fintech, cloud computing and ICT exports (Alikberova & Alikberov, 2020). In the industrial sector, the advanced manufacturing sector is more digitalized (L. Zhang & Chen, 2019).

Scholars have used different methods to measure the degree of digitalization. Based on the maturity model, Schumacher, Erol and Sihn (2016) used the Likert Scale method to evaluate the importance of digitalization degree from 62 indicators in 9 dimensions. They divided the enterprise digital transformation into five levels, from never realizing digitalization to fully realizing digitalization. Schuh, Anderl, Gausemeier, Hompel and Wahlster (2017) analyzed four modules of resources, information systems, organizational structure, and culture by radar chart. They divided the enterprise digital transformation into six levels: computerization, connectivity, visibility, transparency, prediction ability, and adaption. Additionally, Marinko Škare and Małgorzata Porada-Rochon’(2021) constructed a summary index using the standardized inverse-covariance weighted average of growth of capital services provided by computers and information technology (ICT) assets, the contribution of capital services provided by (ICT) assets to GDP, and share of (ICT) capital compensation in the gross domestic product (GDP) to test the digital convergence of 129 countries. They identified digital leaders and followers by log t convergence test. Skare and Riberio Soriano (2021) also proposed the digital adoption index (DAI) from three dimensions, DAI Business, DAI People, and DAI Governments, to measure the digital adoption of 183 countries.

Digitalization has realized the decentralization of industrial structure, the digitalization of economic activities, and the connection of social life with the Internet of Things (Besson & Rowel, 2012; Tan, Pan, Lu & Huang, 2015). It is a resource available to enterprises and the derivation of enterprise capabilities (McAfee & Brynjolfsson, 2012). Some scholars think that the significance of digitalization to the real economy lies in the renewal of the organizational mind, which emphasizes the reconstruction of traditional industries by digitalized thinking (H. Li, Tian & Li, 2014). Digitalization endows enterprises with four new capabilities: information acquisition and integration capabilities based on the Internet of Things, cloud computing, data mining, artificial intelligence, and other new-generation information technologies (Basole & Park, 2019; Wang & Wang, 2020), market perception ability generated by the direct interaction between enterprises and consumers owing to the “disintermediation” function of the Internet (Dey, Yen & Samuel, 2020), relationship integration ability resulting from enterprise information transparency and the digital connection between enterprises (Zhao, Wu & Liu, 2019), and advanced prediction ability generated by real-time processing and application of super large data in mobile Internet (McAfee & Brynjolfsson, 2012). Digitalization forms more and faster “new competitive means” by giving enterprises new capabilities. Subsequently, it subverts the original value-creation process of enterprises.

Digitalization can promote the innovation ability of enterprises. Digital information and knowledge can be generated, shared, and exchanged among enterprises at low cost, rapid, and real-time via various digital platforms (Menon, Kärkkäinen & Wuest, 2019); it enables enterprises to obtain increased information at a reduced cost, increases knowledge stock, and accumulates additional knowledge for innovation (C. Zhang, Xue & Dhaliwal, 2016). Additionally, digital technology allows the accelerated diffusion and dissemination of knowledge within enterprises, making it easier for information and knowledge to be transformed into high-quality innovation results, thus promoting enterprise innovation (Paunov & Rollo, 2016). The inherent dynamism and plasticity of digital technologies may drive knowledge combinations, which leads to new digital knowledge generation and digital innovation (Hanelt, Firk, Hilebrandt & Kolbe, 2020). Kaufmann, Lehner and Tödtling (2003) empirical analysis of Australian enterprises showed that as a typical representative of digital technology, Internet technology has a significant positive effect on enterprise innovation cyberspace expansion. Purdy, Qiu and Chen (2017) have shown that digital technology could improve the innovation ability of enterprises.

Another significant impact of digitalization on enterprises is that it can effectively reduce the cost of enterprises. With the help of digital platforms, enterprises can use free knowledge, information, and low-paying or even zero-paying public power to reduce the cost of enterprise information acquisition (García, Guillot, Zimányi & Langman, 2012). Digital technology has transformed “information scarcity” into “information democracy” (Dey et al., 2020). The connection function of digital technology can reduce the information asymmetry between consumers and producers (Cappellini & Yen, 2013), remove unnecessary intermediate channels, and accurately connect the consumers and producers, which can reduce the information search cost (García-Dastugue & Lambert, 2003) and transaction cost of both consumers and producers (Howe, 2008). Additionally, in the digital economy era, everyone is self-Media; the media-based marketing behavior of consumers can effectively reduce the marketing cost of enterprises (Dan, Qu, Liu, Zhang & Zhang, 2014).

Digitalization can also improve productivity. Brynjolfsson, Hitt & Kim, 2011 used the survey data of 179 large listed enterprises to analyze the effect of data and business analysis on enterprise productivity. They found that the average productivity of enterprises making decisions using big data and business analysis was high. The empirical analysis of Tunisian enterprises by Mouelhi (2009) showed that using information and communication technology could promote the production efficiency of manufacturing enterprises. Forero(2013) proposed that Internet technology development can improve a technical efficiency of a country.

The production function is a functional model that reflects the value-creation process by describing the relationship between input and output (Ackerberg, Caves & Frazer, 2015). Marshall(1920) constructed the first gross production function. He proposed that the commodity value could be created jointly by labor, capital, land, and entrepreneurs. However, he did not provide the specific form of the function. Subsequently, Cobb and Douglas(1928) introduced technical factors into the relationship between input and output and proposed the Cobb-Douglas (C-D) production function, which has become the most widely used production function in economics. The C-D production function was further developed into the CES (Qian & Wu, 2020), the VES (Gamlath & Lahiri, 2018), Stochastic Frontier (Banker, Natarajan & Zhang, 2019; Kumbhakar & Tsionas, 2006), and trans-log production functions (B. Lin & Raza, 2020). The main difference between these kinds of production functions lies in the different settings of the elasticity of substitution. The early C-D production function assumes that technological progress is neutral and the elasticity of substitution is equal to 1. However, many subsequent studies have shown that technological progress was not necessarily neutral(Hicks, 1932a), and biased technical progress was proposed (Acemoglu, 2002, 2007). Subsequently, Arrow, Chenery, Minhas & Solow, 1961 thought that the elasticity of substitution in actual production is not equal to 1. Furthermore, they proposed the CES production function, whose elasticity of substitution is a constant greater than 0.

The initial production function focused on the relationship between various kinds of physical capital (machines, buildings) and labor factors (Berndt & Christensen, 1973). However, some scholars have subsequently, studied the alternative relationship between agricultural production factors (Shumway, 1995) and the relationship between energy and other production factors (Popp, 2002). Recent studies have paid more attention to the alternative relationships among skilled labor, unskilled labor, and physical capital (Acemoglu & Restrepo, 2018, 2019) and have also introduced digital factors into the production function (Meng & Wang, 2021).The production function has developed from two factors to multi-factors, such as the nested CES production function, to meet the increase in factor types (Henningsen, Henningsen & van der Werf, 2019).

Existing research has analyzed the effect of digital technologies on innovation capacity, cost, and enterprise productivity. However, some deficiencies still exist. First, existing studies lack a unified theoretical framework to systematically study the digital transformation of enterprises and fail to uncover the black box of the process of digitally enabling enterprises. Second, regarding the measurement method of the digitalization degree, most researchers use the comprehensive indicator method and maturity model, which results in measurement results strongly dependent on the coverage, integrity, and comprehensiveness of the indicator data. However, owing to the lag and unavailability of statistical data, comprehensive and complete indicator data that can reflect the degree of digitalization cannot be obtained in many cases, which affects the accuracy of measurement results. Third, most studies on the impact of digitalization on enterprises usually use econometric models for analysis. The econometric model has high requirements for the quantity and distribution of data, and the data needs to meet several assumptions (Coester, Hofkes & Papyrakis, 2018). In the digital economy, massive data has been continuously emerging, creating higher requirements for data collection and processing and model function construction (Georgiadou, Angelopoulos & Drake, 2020), which inevitably highlights the limitations of the econometric model.

In response to the above problems, this study puts forward the production function of digital empowerment to construct a theoretical framework, and analyzes the digital enabling process for enterprises from the production factor (that is logical starting point of value creation) and the economic activity (that is value converter). Then, this study utilizes text mining technology and big data analysis to identify the key information reflecting the degree of digitalization from the massive data to measures the degree of digitalization of enterprises, and extracts the relationship model between the degree of digitalization of various enterprises and enterprise performance. The use of these digital technologies makes up for the incompleteness and lag of statistical data to a certain extent. Finally, the association rule analysis method with the characteristics of making full use of big data information and without assumptions is used to excavate the hidden association rules between various production factors and economic activities under different relationship models, which overcomes some shortcomings of econometric models.

Production factors are the logical starting point of value creation. It is necessary first to clarify the composition of the production factors involved in enterprise value creation to study the impact of digitalization on the value creation of enterprises. In the early stage, the production factors involved in value creation mainly included all kinds of material capital (machinery, construction), labor, and energy. With the development of science and technology, labor was further subdivided into skilled and unskilled labor. The change in the interrelation between labor and physical capital has become an essential factor affecting the value creation of enterprises (Acemoglu & Autor, 2011; Acemoglu & Restrepo, 2018, 2019). In the digital economy, digital information and knowledge have become the key production factors. The composition and interrelation of production factors involved in value creation have changed again. Digitalization has infiltrated traditional labor (Lt) and traditional capital (Kt) to form digital labor (Ld) and digital capital (Kd). They jointly create the value, Y = f (Lt, ALLd, Kt, AKKd). Digital labor and digital capital have been involved in value creation, changing the relationship among production factors by replacing or supplementing traditional production factors and endowing energy to production factors to stimulate their value creation potential.

Substitution, supplement, and other associated matching relationships among production factors are usually expressed by substitution elasticity, which refers to the ratio of the change rate of the factor input ratio to its relative price change rate(Hicks, 1932b). Therefore, the elasticity of substitution between the digital and traditional factors can be expressed as:

The CES production function effectively analyzes the substitution elasticity among production factors and its impact on value creation. This study explores how digitalization can influence the relationship between traditional and digital factors through empowerment, thus affecting the value creation of enterprises. Therefore, the production function of digital empowerment with multi-factor and multi-level nesting is constructed.

It is assumed that the production factors mainly include traditional labor (LtT) and capital (KtT) and digital labor (LdT) and capital (KdT). The production function of digital empowerment is expressed as:

The digital enabling level is mainly determined by the digitalization degree and the influence of digitalization on enterprises.

Because the transformation and upgrading of innovation, production systems, and supporting equipment compatible with digitalization need some time to complete, the popularization of digitalization in enterprises at the initial stage is relatively slow, giving digitalization a minor role of in enterprise promotion. However, eventually, digitalization will significantly promote the development of enterprises (Brynjolfsson, Rock & Syverson, 2017). Some scholars have proposed that the influence of digitalization on economic development presents a trend of “first convex then concave” (Lin, Chen, Chen & Chen, 2020). Considering that the logistic function satisfies both the characteristics of “convex first, then concave” and the existence of upper and lower limits (Chiyo, Mizukami & Yokota, 2020; Van Loon, Fokkema, Szabo & de Rooij, 2020), this study uses the logistic function to describe the impact of digitalization on the enterprise. Therefore, the digital influence coefficient (DT) can be expressed as follows:

Based on the above production function of digital empowerment, the substitution or complementary among production factors can be obtained, which is an essential factor affecting the digital development of enterprises. As a value converter, economic activities are also the key factors that determine the digital development of enterprises. The digital development of enterprises is reflected in the degree of digitalization and performance of enterprises. Only when the degree of digitalization and performance are both high the digital development of enterprises improves. Therefore, it is necessary to identify the relationship between the degree of digitalization and performance and excavate the association rules between production factors (including traditional and digital factors) and economic activities under different relationship models.

The relationship between the degree of digitalization and enterprise performance can be divided into four types according to the degree of digitalization and enterprise performance: high digitalization-high performance (Hd-Hp), high digitalization-low performance (Hd-Lp), low digitalization-high performance (Ld-Hp), and low digitalization-low performance (Ld-Lp).

The association rule mining technology extracts the association rule between production factors and economic activities under different relationship modes.

Association rule analysis is an algorithm mining association relationships between transactions hidden in massive data(Agrawal, Imielinski & Swami, 1993). It identifies the combination with high frequency from big data, analyzes the dependence among different factors, and explores the emergence rule of each factor.

Digital degree, enterprise performance, the input proportions of production factors, and economic activities constitute the item set I={TL0–10-TL90–100, DL0–10-DL90–100, TK0–10-TK90–100, DK0–10-DK90–100, CP0–10CP90–100, CS0–10CS90–100}. Where the input proportions of various production factors in the range of [0, 100] are divided into 10 grades from large to small, i.e., traditional labor (TL0–10-TL90–100), digital labor (DL0–10-DL90–100), traditional capital (TK0–10-TK90–100), and digital capital (DK0–10-DK90–100). Economic activities mainly include production and sales activities, in which production and sales activities are represented by the production cost and sales expense rates, respectively. Similarly, they are equally divided into 10 grades in the range of [0, 100], production cost rate (CP0–10CP90–100), and sales expense rate (CS0–10CS90–100). T={t1, t2, t3, …, tn} is the set of transactions, ti⊆I. If X→Y satisfies X⊂I, Y⊂I, and X∩Y≠Φ, then X→Y is the association rule in T.

The mining process consists of two stages: the first is identifying all frequent item sets; we use the most classic Apriori algorithm (Agrawal et al., 1993) to search frequent item sets. The second is the mining association rules in these frequent item sets. Association rules are usually measured by support and confidence. Support refers to the probability of occurrence of an item set (X→Y) in the whole data set, representing the frequency of rules. The calculation formula is:

Generally, a support threshold is set; when the support of an item set exceeds this value, it will be considered worthy of attention. Such an item set is called a frequent item set.

Confidence is a conditional probability; under the condition that X occurs, the probability of Y derived from X indicates the strength of the rule. The confidence of X → y is defined as:

In addition to the support and confidence, we introduce the lift as the standard of rule selection. Lift refers to the ratio of the confidence of the rule to the probability of the occurrence of the consequent, reflecting the positive and negative correlation between the antecedent and consequent.

By adding a lift to the measurement framework of rules, we can remove the rules with less relevance from the rules mined by support and confidence. If the lift is greater than 1, and the higher the lift is, the higher the positive correlation is; if the lift is less than 1, and the lower the negative correlation is, the higher the negative correlation is, and if the lift is equal to 1, it means there is no correlation.

This study takes 988 A-share listed manufacturing enterprises in Shanghai and Shenzhen stock markets from 2019 to 2021 as the research object, and 862 listed companies are finally selected for the research excluding 126 samples with missing data. Since the impact of digitalization on value creation is characterized by enterprise heterogeneity, we firstly classify manufacturing enterprises into light textile enterprises, resource processing enterprises and mechanical and electronic enterprises based on the operation characteristics and product attributes of enterprises (W. Wang & Niu, 2019).

The variables involved in the production function of digital empowerment mainly include: output of enterprise (Y) is expressed by the total output of listed enterprise(H. Wang, Gao & Zhang, 2019). Digital labor (Ld) is represented by the total number of R & D personnel and operators of digital equipment, while traditional labor (Lt) is represented by the number of general personnel, that is, the difference between the total number of personnel and the number of digital labor. Digital capital (Kd) refers to total investment of R & D, digital devices and platforms, while traditional capital (Kt) refers to net fixed assets including plant construction, general equipment, raw materials. The digitalization degree (digitalT) is reflected by the digitalization behavior of the enterprise, which can be obtained by using text mining to identify the number of keywords related to digitalization, such as artificial intelligence(AI), big data, cloud computing, digitization, Information, intelligence, network, Internet +, Internet of Things, e-commerce, platform economy, sharing economy, online and offline, O2O, B2B, C2C, B2C, C2B. (Meng & Wang, 2020). The performance of enterprise can be expressed by the return on assets(Chang & Rhee, 2011; Hsu, Lien & Chen, 2013). All data are from CSMAR database.

The parameters of the generalized technical level C,C1, and C2 are set to 1 to simplify theoretical analysis and estimation (Acemoglu, 2002, 2003)

In the function of the digital influence coefficient (DT), {D¯,ξ,and h} are determined by the evolution track of digitalization enabling. The existing studies provide the basis for their values. Purdy et al. (2017) and Purdy, M and Daugherty (P, 2017) predicted the influence of digital technologies on labor productivity in the United States, Japan, China, and other countries. They found that by 2035, the digital technique would increase labor productivity in the United States and Japan by 11%–41% and in China by 27%. Based on the above research, this study uses the labor productivity increase brought by digitalization to express a digital enabling coefficient. It assumes that the digital enabling coefficients of resource processing, light textile, and mechanical and electronic enterprises in 2035 are 0.10, 0.2, and 0.3, respectively, which is consistent with the assumption that digital technology will increase China's labor productivity by 27% in 2035. The OECD (2015) pointed out that since 2000, the average growth rate of labor productivity of industries using digital frontier technology has been 3.5%. Based on this, this study sets the initial digital enabling coefficients of resource processing, light textile, and mechanical and electronic enterprises to 0.03, 0.035, and 0.04, respectively. Referring to the research of Chen, Lin & Chen, 2019, the change curvature of the digital enabling coefficient ξ is set to 0.055, and then h can be calculated. The relevant parameters of the digital enabling coefficient of the three types of enterprises are shown in Table 1.

Based on the above empirical results, this study proposes the digital transformation path of enterprises with different relationship models to promote the development of enterprises.

• (1)

  For light textile enterprises, strengthening the cultivation and introduction of digital labor and increasing digital capital investment is the crucial path to improving the degree of digitalization and enterprise performance. Compared with the optimal relationship model (Hd-Hp), the main task of enterprises belonging to the Hd-Lp model is to increase the input of digital capital in sales links, such as the procurement of digital sales equipment, and investment in construction, operation, and maintenance of digital sales platform. For enterprises under the Ld-Hp and Ld-Lp relationship model, the production cost rates are higher than that of the Hd-Hp model, and the sales expense rate is lower than that of the Hd-Hp model. These enterprises should focus on both production and sales, and increase the digital labor input in the production link (such as digital equipment operators, hardware installation, and maintenance personnel), also increase the input of digital sales skills. Additionally, investment in digital equipment, technology research and development, digital platform construction, and other aspects of production and sales should also be strengthened.

• (2)

  For resource-processing enterprises, it is necessary to increase the input of digital labor and capital in the production process, which has the greatest impact on its profitability. Compared with the Hd-Hp model, the production cost rate of the other three models is higher, and the proportions of digital labor and capital are relatively low. This indicates that increasing the investment and talents in the transformation of automation, informatization, digitalization, and intelligent in the production process can inhibit or reduce the dependence on traditional capital and low-skilled labor, to improve the digitalization degree and performance of enterprises.

• (3)

  For mechanical and electronic enterprises, the cultivation and introduction of high digital technology labors and the financial support for research and development and the procurement of advanced digital equipment are very important. Mechanical and electronic enterprises belonging to the Hd-Lp model have relatively advanced digital equipment and highly skilled labor; the high-end and scarce talents who master AI, big data, cloud computing, 5 G, and other professional digital technologies are decisive factors for the digitalization degree and performance, which need special attention. However, enterprises in the Ld-Hp and Ld-Lp models still have a low degree of digitalization, and their primary task is to increase the input of labor and capital to support the informatization and networking of the basic business processes in the production process.

This study proposes several insights, which have three-fold theoretical contributions for the research on digital transformation of different types of enterprises.

Firstly, the production function of digital empowerment is proposed to analyze the influence mechanism of digitalization on value creation. Although there are some researches analyzing the of digital technologies on innovation capacity, production and sales costs and production efficiency of enterprises, these researches rarely consider the digital empowerment of production factors, and embody this empowerment mechanism in the model. The production function of digital empowerment theoretically analyzes the influence mechanism of digitalization on enterprise value creation, which includes the digital enabling level on each factor and the elasticities of substitution among the factors after introducing the empowerment. Thus, this study is an extension of previous studies, and offers a theoretical framework that may benefit future studies of digital transformation of enterprise.

Secondly, the association rules among the production factors and economic activities of enterprises is excavated. Production factors are the logical starting point of value creation, and economic activities are the conversion of factors into value. Digital resources and traditional factors accelerate the flow and reorganization in economic activities, which will bring about subversive changes to the internal and external environment and structure of enterprises (Bejakovic & Mrnjavac, 2020; Mueller & Grindal, 2019). Therefore, from the factor-activity aspect, this paper can fundamentally explore the impact of digitalization on enterprise value creation. This is a new perspective of research on enterprise value creation under the background of digital economy.

Thirdly, this study uses data mining techniques such as text mining, big data analysis and association rule analysis to conduct empirical analysis, which breaks through the limitations of high data requirements and preset assumptions in econometric models. Therefore, it is a supplement of the previous research methods.

Several practical implications can be highlighted from this work, these insights may provide direction for the digital transformation of various enterprises in the future. First, enterprises with different scales, products and industrial types have different digital transformation paths due to different capital guarantee, technical support and business models. Despite some literatures have studied the effect of digitalization on many different types of enterprises, such as oil enterprises(Lu, Guo, Azimi & Huang, 2019), automobile enterprises (Forero, 2013; Mouelhi, 2009), food enterprises (Fernández-Uclés, Bernal-Jurado, Mozas-Moral & Medina-Viruel, 2019), most of them only analyze a certain type of enterprise, and there is a lack of comprehensive analysis on the differences of digital empowerment and digital transformation of different types of enterprises. We classify enterprises based on their industry characteristics and product attributes, and analyze the heterogeneous influence of digitization on different types of enterprises. Our research results will provide a reference for different types of enterprises to choose the appropriate path of digitalization transformation. Second, we analyze the association rules between production factors and economic activities of three types of enterprises in different digitalization- performance relationship model, which can help enterprises find the shortcoming in the process of digital transformation, and then guide different enterprises to reasonably allocate resources and organize various economic activities in the process of digital transformation. It provides decision support for the path design of improving enterprise performance through digital technology.