Impact of Inventor’s Cooperation Network on Ambidextrous Innovation in Chinese AI Enterprises

Abstract

With the intensification of artificial intelligence (AI) industry competition, innovation has become an important practical way for companies to achieve sustainable development. In this context, it is important to study the transmission paths affecting the development of ambidextrous innovation in AI firms from the perspectives of inventor cooperation networks and technological knowledge base variety. This study uses a sample of 399 of China’s AI organizations from 2008 to 2017. We explore the impact of an inventor cooperation network on ambidextrous innovation and dissect the mediating role of technological knowledge base variety between inventor cooperation network and ambidextrous innovation. We find that inventor cooperation network structural hole and centrality have an inverted U-shape relationship with a firm’s ambidextrous innovation. Additionally, inventor cooperation network centrality has a positive effect on technological knowledge base variety. In addition, there is an inverted U-shaped relationship between inventor cooperative network structure holes and technological knowledge base variety. Meanwhile, the research also finds that the unrelated variety of technological knowledge base mediates the relationship between the inventor cooperation network and ambidextrous innovation. The related variety of technological knowledge base mediates the relationship between the inventor cooperation network and exploitative innovation. However, the related variety of technological knowledge base cannot play a mediating role between inventor cooperation network and exploratory innovation.

1. Introduction

With the rise of a new generation of artificial intelligence (AI) wave, countries around the world have increased investment in the field of AI, in order to use the ability of AI technology to promote the construction and sustainable development of the digital economy. Additionally, for AI enterprises, as macro digital economic activities and meso digital industry development of micro subjects, its level of innovation and development is directly related to the success or failure of economic development mode conversion. Innovation is the key for AI companies to gain sustainable competitiveness and thus lead the continuous evolution and development of new economic forms, especially as China’s economy and society have now shifted to the stage of high-quality development. AI enterprises should focus on two types of innovation development, namely exploratory and exploitative innovation [1]. Ambidextrous innovation includes exploratory innovation and exploitative innovation. Among them, exploratory innovation refers to the process by which firms pursue new knowledge and new fields to meet the changing needs of the market. In contrast, exploitative innovation is a progressive process that builds on existing knowledge and contributes to the effectiveness of the methods and technologies owned by the company, thereby increasing existing competitiveness [2,3]. These two innovation capabilities are key channels for improving the ability of technology companies to achieve sustained success, but they also bring management paradoxes. Specifically, because these two types of innovation require different resources and knowledge, companies face the challenge of how to appropriately handle resource allocation [4]. Liao and Tsai (2019) argue that external knowledge can be an important source of ambidextrous innovation [5]. As individual firms find it difficult to gain competitive advantage in the current market, they must interact with different partners to gain knowledge and resources to thrive in the highly competitive market. Cooperative networks formed by interconnected firms are often seen as an important source mechanism for acquiring external knowledge and are playing an increasingly important role in explaining firms’ ambidextrous innovation [6].

However, the implementation of innovation is subject to human behaviors that involve internal communication cooperation, emotions [7,8,9], etc. Inventors are the backbone of corporate innovation actors. Ardito et al. (2016) pointed out that the key to increasing the R & D capability of an organization or industry is how the inventor acquires and reorganizes knowledge [10]. Inventors are the key players who are actually involved in the process of acquiring and applying new knowledge. Based on the two-dimensional combination of social capital, Rong et al. (2020) divided the role of inventors and found that “star inventors” have a higher social capital advantage and generate the highest innovation capability [11]. Sun and Cui (2021) further demonstrated empirically that there is a strong correlation between star inventors and innovation impact and novelty [12]. Liu and Yang (2020) theorized on inventor knowledge characteristic, analyzing the important role of the uniqueness and diversity of inventor’s knowledge on the cooperation network centrality [13]. However, companies pursuing the process of knowledge recombination need social interactions among inventors to work effectively to produce more valuable innovations [14].

Whether inside or outside the company, inventors use their social networks to collect resources and information to continually improve their network position [15] when a focal researcher is in a vantage position from which to create a sustainable competitive advantage for firms to further enjoy greater opportunities for knowledge flow [16]. There is a double-sided effect of inventor cooperation networks on innovation capability in existing studies. On the one hand, Zhang et al. (2017) believed that inventors with more centrality will have access to more information, more power and greater influence, which is beneficial to enterprise innovation output [17]. Fu et al. (2018) proposed that the structure holes in cooperative networks at the inventor level has a significant positive impact on the firm’s exploratory innovation [18]. On the other hand, Zang’s research (2018) found that occupying structural holes would hinder trust among inventors and increase opportunistic behavior in the network, which would have a negative relationship with exploitative innovation [19]. De Araujo et al. (2019) showed that high levels of duplicate connections may create technological lock-in effects among inventors, which is detrimental to improving innovation performance [20]. In summary, prior research has focused on the linear relationship between the inventor cooperation network and innovation capability and often ignores the possibility of a non-linear relationship between these two. Therefore, the relationship between inventor cooperation networks and a firms’ innovation capability needs to be further investigated.

Meanwhile, Paruchuri and Awate (2017) showed the importance of an inventor network in shaping the use of firm knowledge in innovation activities [21]. The use of organizational knowledge requires that firms have a certain knowledge base because the knowledge base provides for the application of knowledge and the creation of knowledge in the enterprise. According to the knowledge-based theory, the technological knowledge base variety is a critical component of innovative capabilities [22]. Knowledge base variety is a source of discovering new ideas and opportunities to assist firms in achieving economics of scope in innovation. It also enhances firms’ ability to interpret externally acquired information and promotes innovation performance. Tsai (2001) also pointed out that the distributional characteristics of the knowledge base can directly influence the technological innovation mode [23]. Technological innovation can be seen as the outcome of a recombination of existing knowledge elements or the new knowledge elements. Thus, the technological knowledge base variety is a collection of multiple technological areas required for the enterprise to carry out economic activities, and its variations determine the firm’s future innovation direction [24]. Krafft et al. (2011) argued that technological knowledge base variety can be divided into related and unrelated [25]. The former refers to the proportion of the enterprise’s resources allocated to the relevant technology area to facilitate the development of exploitative innovation. The latter refers to the ratio of resources allocated to completely new technological areas to support the development of exploratory innovation. However, it is not clear whether different types of diversification have distinct effects on enterprise ambidextrous innovation.

In summary, the following shortcomings of the existing studies remain. First, the existing literature mostly focuses on inventor identification and knowledge characteristics from the perspectives of an inventor’s social capital and knowledge elements. Fewer scholars have focused on intra-organizational network, especially on inventors who occupy different positions in the intra-organizational network. This may lead to a lack of research on cooperative networks. Moreover, prior research has adopted linear thinking to examine the relationship between cooperative networks and ambidextrous innovation, ignoring the potential costs of over-embeddedness. It is necessary to explore whether a non-linear relationship exists between inventor cooperation network position and ambidextrous innovation from the inventor cooperation network perspective. Second, the existing literature has recognized the importance of the technological knowledge base. However, it does not combine the inventor cooperative network with technological knowledge to examine its mechanism in different innovation activities, nor does it examine whether the different types of technological knowledge base variety are mediated in cooperation networks and ambidextrous innovation. As a result, it is impossible to effectively reveal the influence mechanism of inventor’s cooperative network on enterprise ambidextrous innovation.

To the extent that this paper integrates the resource base view and social network theory, it explores the direct effect of inventor cooperation network centrality and structural hole on ambidextrous innovation. Additionally, we elucidate the mediating role of different types of technological knowledge base variety. Then, the patent data of China’s AI industry from 2008 to 2017 are used for empirical testing. This study systematically shows the development mechanism of enterprise ambidextrous innovation from the perspective of knowledge flow and inventors and provides practical guidance for enterprises to reasonably choose the network position and combine their own technological knowledge base to develop their ambidextrous innovation capability.

This research contributes to our understanding of the role of network and knowledge in ambidextrous innovation. First, based on social network theory, we construct a theoretical relationship model between inventor cooperation network and ambidextrous innovation. This paper analyzes the influence mechanism of network centrality and structural hole on exploratory innovation and exploitative innovation and expands the research content of structural embedding. In particular, the research on the nonlinear relationship between structural hole and exploratory innovation has a certain theoretical significance to explain the paradox of the relationship between cooperation network and innovation. Second, based on the resource base view, the related and unrelated variety of the technological knowledge base is incorporated into the theoretical model to analyze its mediating role between inventor cooperation network and ambidextrous innovation, and the internal mechanism between them is analyzed deeply. This study deepens the theoretical research of inventor cooperative network and ambidextrous innovation. Third, the results of this paper can provide practical guidance for enterprises to reasonably choose the network position and combine their own technological knowledge base to develop ambidextrous innovation capability.

2. Theoretical Background and Hypothesis Development

2.1. Inventor Cooperation Network

Ardito et al. (2016) showed that inventors are the core of the corporate knowledge search and recombination process and are the main group of companies involved in invention activities [10]. According to the social network theory, the cooperative relationship between inventors constitutes a network of enterprise researchers, which can provide access to diverse knowledge and is the key to improving corporate competitive ability. Paruchuri et al. (2017) found that when firms build their own technological knowledge through cooperative networks, it is the inventors who build organizational knowledge through cooperative networks [21]. Kogut et al. (1992) pointed out that inventor cooperation networks can not only determine the quantity and quality of network resources available to enterprises [26], but they also enable the knowledge recombination process through a cooperative relationship among them and then affect the technological innovation performance of the firm. Figure 1 provides the network coupling between the firm level, inventor level and knowledge level.

2.2. Research Hypothesis

2.2.1. Inventor Cooperation Network Structure Hole and Ambidextrous Innovation

Structural hole depicts the non-redundant ties presented by other inventors in the cooperative network who are directly connected to the key inventor [27]. The nodes that occupy structural holes act as a bridge for the flow of information and new knowledge between networks. This means that inventors have access to all kinds of information and knowledge that play an important role in the enterprise innovation [28]. However, the actual innovation performance may vary with enterprises that emphasize different types of innovation because exploratory and exploitative innovation involve different types of resources and activities [29,30].

Firstly, exploratory innovation requires novel knowledge and resources. Structural holes can provide opportunities for non-redundant knowledge flow among inventors. Occupying structural holes, inventors will be exposed to new ideas and broader R & D views from diverse knowledge sources [31]. This will help inventors to overcome organizational inertia and propose new methods and new thinking frameworks for solving technological problems. Second, an inventor who occupies structural holes has information benefits and autonomy in decision making [32]. Information is different than knowledge elements. Information can be about the expertise and the previous successes and failures of the inventor. This information can alert inventors to emerging inventions [29]. The characteristic of exploratory innovation is the process of discovering new opportunities for enterprises. At the same time, the structural hole also gives the inventor autonomy. Inventors who occupy structural holes can explore new ideas without being limited by their partners, such as opinion leaders and herd mentality. These limitations can hinder the willingness and effectiveness of inventors to continue implementing innovations [33]. However, spanning too many structural holes can also inhibit enterprise exploratory innovation. On the one hand, when the inventors who occupy structural holes pursue novelty excessively, this will lead to the continuous change in their knowledge base in different and irrelevant dimensions, which may cause the final product to deviate from the market demand. On the other hand, inventors who span more structural holes have a large amount of non-redundant knowledge. In order to identify, assimilate, transform and exploit this knowledge, inventors must invest more resources and effort [34]. As there is a limit to the absorptive capacity of inventors, when this limit is exceeded, this leads to cognitive overload, which is not beneficial for enterprises’ exploratory innovation [35,36].

At the same time, the structural hole will also affect the exploitative innovation. According to the structural hole theory, the inventor who occupies the structural hole is able to obtain timeliness. Timeliness is an important feature for brokers to obtain information through their unconnected collaborators. Thus, brokers have quick access to various sources of information in the network, which contributes to the firm’s exploitative innovation. In addition, the inventor who occupies the structural hole enjoys control advantages. In particular, they are competitive in controlling the information flow and knowledge diffusion in the network [37]. Inventors across more structural holes can gain better insight into competitors’ business motives, effectively predict technology trends and capture market changes, and hence improve the use of existing technologies and methods in firms [38]. However, inventors spanning too many structural holes is negatively associated with exploitative innovation. Since exploitative innovation follows the original technological path, it is difficult for inventors to improve existing products with the diverse knowledge gained through the structural hole. Additionally, structural holes can create loosely coupled organizational structures that may prevent trust, increase the difficulty of knowledge transfer, and reduce the efficiency of the knowledge combination. In addition, its inability to provide fine-grained information will weaken firms’ exploitative innovation [28,39,40]. This leads to our hypotheses:

Hypothesis 1a (H1a).

The inventor cooperation network structure hole has an inverted U-shaped relationship with the firm’s exploratory innovation.

Hypothesis 1b (H1b).

The inventor cooperation network structure hole has an inverted U-shaped relationship with the firm’s exploitative innovation.

2.2.2. Inventor Cooperation Network Centrality and Ambidextrous Innovation

Inventor cooperation network centrality refers to the number of other network members that are directly connected to an inventor [41]. The greater the centrality of the inventor, the stronger the ambidextrous innovation ability of the enterprise [33]. On the one hand, inventors with high centrality are often in a central position and can obtain diversified information from the cooperation of different inventors, resulting in competitive advantage [42]. This information will also be recombinant into novel, useful knowledge and techniques that help inventors discover new technologies and potential market opportunities. Additionally, cooperation with other inventors in the network may reduce market and technology risk and promote exploratory innovation in the firm [32,43]. At the same time, inventors with higher centrality may obtain new information and seek new developments faster than inventors with lower centrality [44]. On the other hand, the central network position enables inventors to obtain more diversified resources and capabilities, thus benefiting enterprises. This centrality allows inventors to better understand market demand and create products that meet consumer desires. Moreover, inventors with higher centrality maintain frequent contact and interaction with other members within the network, which in turn enhances trust and reciprocity and facilitates the diffusion of tacit knowledge. Hsu et al. (2019) argued that trust and reciprocity stimulate network members to invest energy and effort in the exchange of tacit knowledge. Additionally, with the increasing level of interaction between the inventors with higher centrality and other members, companies will have sufficient information to seize new opportunities related to exploitative innovation in a timely manner [45].

However, when the centrality of inventors exceeds a certain degree, the relationship between inventors’ network centrality and ambidextrous innovation may be negative. First of all, inventors will exchange and share proprietary information and resources with each other to overcome the difficulties in the innovation process [46]. Therefore, there may be a risk of knowledge spillover when collaborating with other inventors. Additionally, these negative effects may be aggravated gradually with the increase in number of partners, which is not beneficial for firms’ ambidextrous innovation. Second, inventors with too much centrality may have attention rationing problems. Since one person’s time is limited, the centrally located inventor must maintain contact with a large number of collaborators. Thus, the time allocated to each collaborator is reduced, leading to a decrease in the strength of their ties. Inventors cannot effectively obtain the required knowledge and technology, let alone the tacit knowledge that would create a cognitive burden [47]. Moreover, inventors can develop a thinking fixation that hinders their own search for new knowledge and weakens the firms’ exploratory innovation [48]. This leads to our hypotheses:

Hypothesis 2a (H2a).

The inventor cooperation network centrality has an inverted U-shaped relationship with the firm’s exploratory innovation.

Hypothesis 2b (H2b).

The inventor cooperation network centrality has an inverted U-shaped relationship with the firm’s exploitative innovation.

2.2.3. Inventor Cooperation Network and Technological Knowledge Base Variety

Open innovation theory suggests that it is difficult for a single individual to gain competitive advantage in the current market, but that individual partners can achieve higher innovation performance from cooperative networks [49]. As cooperative networks consist of extensive knowledge flows between different partners, the knowledge gained from partners allows employees within the company to deepen their thinking and advance innovative concepts [50]. Hence, inventors can access resources and knowledge through cooperative networks and complement the enterprise’s knowledge base [51].

Previous studies have pointed out that the inventors’ cooperative network position can obviously promote the technological knowledge base variety. On the one hand, inventors at the center of the network can establish extensive contacts with other inventors and can quickly collect diversified information and knowledge [52] so as to strengthen their technological capabilities and improve the information richness. At the same time, inventors with higher centrality can identify and recombine external information and knowledge, improve technological novelty, and effectively expand the firms’ technological knowledge base variety [53]. On the other hand, the inventors in the central position have direct contact with other partners, which can facilitate the exchange and sharing of tacit knowledge in the company because it is possible for each company to acquire more knowledge through the cooperative network, thus facilitating the integration of complementary knowledge [54,55,56].

Inventors with rich structural holes acquire diversified technologies and knowledge through cooperative networks to improve the coupling between new and old knowledge. This can improve their level of understanding, better capture resources or complementary technologies needed for business development, and carry out the activities of technological knowledge base variety in enterprises [57]. Additionally, inventors with rich structural holes have more opportunities for creative knowledge recombination, enhancing the ability to communicate and share knowledge with partners. This provides access to a wider range of knowledge and resources and increases the intellectual capital of the company [58]. However, the heterogeneous knowledge acquired by inventors through cooperative networks requires them to have the appropriate knowledge processing capabilities [59]. As heterogeneous knowledge continues to grow, this will excessively disperse the limited efforts and resources of inventors. Even when heterogeneous knowledge exceeds a certain limit, inventors may have cognitive overload, which reduces the willingness of enterprises to implement the knowledge base variety [60]. This leads to our hypotheses:

Hypothesis 3a (H3a).

The inventor cooperation network centrality has a positive impact on the related variety of technological knowledge base.

Hypothesis 3b (H3b).

The inventor cooperation network centrality has a positive impact on the unrelated variety of technological knowledge base.

Hypothesis 4a (H4a).

The inventor cooperation network structure hole has an inverted U-shaped relationship with the related variety of technological knowledge base.

Hypothesis 4b (H4b).

The inventor cooperation network structure hole has an inverted U-shaped relationship with the unrelated variety of technological knowledge base.

2.2.4. Mediating Effect of Technological Knowledge Base Variety

According to the knowledge-based theory, the reuse of existing knowledge is considered a key aspect of exploitative innovation, while exploratory innovation requires the recombination of novel and diverse knowledge [61,62]. In this regard, inventors can expand the knowledge base of the enterprise by acquiring external knowledge from an advantageous position. This will complement the additional knowledge that may not come from internal knowledge heterogeneity and facilitate firms’ participation in ambidextrous innovation [63]. However, not all knowledge is conducive to ambidextrous innovation. Excessive external knowledge heterogeneity will lead to information overload. Excessive internal knowledge heterogeneity will increase complexity, difficulty in coordination, and R & D costs, thus weakening a firm’s innovation capability [64,65]. The ability to effectively use external knowledge to create value is largely dependent on the level of a prior knowledge of the company [1]. The more prior knowledge a firm possesses, the better its absorptive capacity and the stronger the firm’s ambidextrous innovation capability. Similarly, when enterprises identify and absorb new and diverse knowledge, the integration ability of enterprises also plays a key role in dealing with knowledge and increasing their innovation activities [66]. In other words, firms must have the ability to combine, process and apply the knowledge and technology acquired so that they can better utilize their own technological knowledge base for ambidextrous innovation. Therefore, in order to improve ambidextrous innovation in the rapidly changing business environment, firms need to seek new knowledge externally, combined with internal efforts to build their own technological knowledge base, so as to improve the firm’s innovation capabilities. Accordingly, this paper argues that technological knowledge base variety has a mediating effect between inventor cooperation networks and ambidextrous innovation.

In terms of knowledge spillover and information flow, inventors in the central position have greater advantages than marginal inventors [67]. These advantages can help enterprises to obtain complementary information and resources, and then enrich the diversity of technological knowledge, which is helpful to improve the firm’s ambidextrous innovation capabilities. On the other hand, inventors in the central position can establish extensive contacts with different partners. It is convenient for them to deepen domain knowledge and improve the quality of knowledge acquisition, as well as to obtain more new ideas from the external environment [22,52]. By integrating the old and new knowledge of the enterprise, this can expand the way the enterprise recombines knowledge and generates new solutions, which is conducive to the realization of ambidextrous innovation. This leads to our hypotheses:

Hypothesis 5a (H5a).

The related variety of technological knowledge base mediates the relationship between the inventor cooperation network centrality and exploratory innovation.

Hypothesis 5b (H5b).

The related variety of technological knowledge base mediates the relationship between the inventor cooperation network centrality and exploitative innovation.

Hypothesis 5c (H5c).

The unrelated variety of technological knowledge base mediates the relationship between the inventor cooperation network centrality and exploratory innovation.

Hypothesis 5d (H5d).

The unrelated variety of technological knowledge base mediates the relationship between the inventor cooperation network centrality and exploitative innovation.

Inventors who occupy structural holes provide firms with more opportunities for creative knowledge recombination and easy knowledge sharing and acquisition [58]. Additionally, inventors have advantages in acquiring architectural knowledge through collaborative network structural holes. This knowledge broadens the technological knowledge base of enterprises, which is not only conducive to improving the exploitative innovation of existing technologies and products, but also conducive to the exploratory innovation of discovering new technologies and creating new products [68]. In addition, inventors who occupy structural holes can quickly access various information sources in the cooperative network. It is more likely to obtain heterogeneous information in time, so as to understand the development trend of the industry and grasp the best opportunity for enterprise innovation. This leads to our hypotheses:

Hypothesis 6a (H6a).

The related variety of technological knowledge base mediates the relationship between the inventor cooperation network structure hole and exploratory innovation.

Hypothesis 6b (H6b).

The related variety of technological knowledge base mediates the relationship between the inventor cooperation network structure hole and exploitative innovation.

Hypothesis 6c (H6c).

The unrelated variety of technological knowledge base mediates the relationship between the inventor cooperation network structure hole and exploratory innovation.

Hypothesis 6d (H6d).

The unrelated variety of technological knowledge base mediates the relationship between the inventor cooperation network structure hole and exploitative innovation.

Through the above analysis, this paper constructs the theoretical model, as shown in Figure 2:

3. Methods

3.1. Data Source

In “Made in China 2025”, AI has been officially listed as a strategic emerging industry in China, which has the characteristics of having high innovation requirements, being knowledge and technology intensive, and so on [69]. Clearly, the mechanism of ambidextrous innovation in the field of AI is conducive to promoting the development of a new generation of information technology in China, and further realizing the strategic goal of “Scientific and Technological Powerful Nation”. To this end, this paper selects the field of AI in China as the research subjects of this study to test the proposed research model. The data of this paper come from the database of National Key Industry Patent Information Service Platform. By searching the AI patents in this database, we screened out the data of patents whose applicant (patentee) type is a company, forming the initial sample of data in this paper. Organizations containing only one inventor or one technological field were then excluded and screened for organizations that successfully filed patents from 2008–2012 and still maintained patent applications from 2013–2017. Next, the names of applications (patentees) in the filtered patent data were extracted and duplicate patentees’ names were eliminated, resulting in a sample of 399 companies.

3.2. Variables and Measures

3.2.1. Dependent Variables

Ambidextrous innovation is measured by the change in enterprise knowledge stock. Exploratory innovation (ERA) means that firms generate technological inventions containing new knowledge elements. Exploitative innovation (EIT) means that firms generate technological inventions that are familiar with all knowledge elements. Drawing on Katila et al. (2002), the results of the IPC subclass comparison were used to distinguish firms’ ambidextrous innovation [70]. A patent with an IPC subclass is defined as an exploitative innovation if the firm has an IPC subclass in the latter stage that was present in the previous stage. Conversely, it is an exploratory innovation.

3.2.2. Independent Variables

The inventor cooperation network centrality (INCE): Drawing from Fu et al. (2018), the mean value of the centrality degree of all inventors in the organization was taken to measure this indicator [18].

The inventor cooperation network structural hole (INSH): This paper uses the constraint coefficient to measure the structural holes occupied by inventors in the collaborative network. The computational is as follows:

$$INS{H}_i=2-{{\displaystyle \sum _j\left(P_{ij}+{\displaystyle \sum _{q\ne i\ne j}{P}_{iq}{P}_{qj}}\right)}}^2$$

(1)

As the maximum value of the constraint coefficient index is 1.1536, the “2-constraint coefficient” [4] is used to indicate the extent to which the nodes occupy the structural holes.

3.2.3. Mediator Variables

Drawing on Chen et al.’s study [71], this paper classifies technological knowledge base variety into the related variety of technological knowledge base (RTV) and the unrelated variety of technological knowledge base (UTV). The first three digits of the IPC are used to calculate UTV, while the first four digits of the IPC calculate the technological knowledge base variety, measuring technological knowledge base variety (TV) by the entropy method. The computational is as follows:

$$TV={\displaystyle \sum _{a=1}^n{p}_a\ln \left(\frac{1}{p_a}\right)}$$

(2)

where n denotes the number of patent subclasses, a represents patent subclasses, and $P_a$ represents the ratio of the number of patents containing IPC subclass a to the total number of patents. The formula for UTV is as follows:

$$UTV={\displaystyle \sum _{j=1}^m{p}_j\ln \left(\frac{1}{p_j}\right)}$$

(3)

In Equation (3), m denotes the number of patent classes, j denotes a patent class, and $P_j$ represents the number of patents containing IPC class j as a proportion of the total number of patents. On this basis, the formula for RTV is as follows:

$$RTV=TV-UTV$$

(4)

3.2.4. Control Variables

Organizational age (Age): Organizational age has a significant impact on its ambidextrous innovation activities. The longer the organization years, the more knowledge will be accumulated. Organizations will tend to rely on existing knowledge and not focus on exploratory innovation. An organizational age measured by the difference between the year when the firm first applied for an AI patent and the current year examined in the study.

R & D intensity (RD): R & D intensity indicates the organization’s effort to continually invest in innovation. With reference to relevant research [2], the number of patent applications filed by the organization in the previous five years is used as its proxy indicator.

The names of all the variables are shown in

Table 1. List of variables.

Variable Type	Variable Name	Variable Symbol
Dependent variables	Exploratory innovation	ERA
	Exploitative innovation	EIT
Independent variables	The inventor cooperation network structural hole	INSH
	The inventor cooperation network centrality	INCE
Mediator variables	The unrelated variety of technological knowledge base	UTV
	The related variety of technological knowledge base	RTV
Control variables	Organizational age	AGE
	R & D intensity	RD

.

3.3. Theoretical Model

In this study, the results of ambidextrous innovation data show the characteristics of discrete non-negative integers and over-dispersion; that is, the variance is much larger than the mean, so the negative binomial regression model is used. The basic econometric model is as follows:

(1)

Testing the total effect model.

$$E\left(ER{A}_{it}/EI{T}_{it}\left|X_{it}\right.\right)=\exp \left(c_0+c_1{X}_{it}+c_2{Z}_{it}+{\epsilon }_{1it}\right)$$

(5)

(2)

Testing the mediating effect model.

$$E\left(UT{V}_{it}/RT{V}_{it}\left|X_{it}\right.\right)=\exp \left(a_0+a_1{X}_{it}+a_2{Z}_{it}+{\epsilon }_{2it}\right)$$

(6)

$$E\left(ER{A}_{it}/EI{T}_{it}\left|M_{it}\right.\right)=\exp \left(b_0+b_1{M}_{it}+b_2{Z}_{it}+{\epsilon }_{3it}\right)$$

(7)

$$E\left(ER{A}_{it}/EI{T}_{it}\left|X_{it}\right.\right)=\exp \left(d_0+d_1{X}_{it}+d_2{M}_{it}+d_3{Z}_{it}+{\epsilon }_{4it}\right)$$

(8)

In the above equation, i represents the firm, t represents the year, and ${\epsilon }_{it}$ is the error term. $X_{it}$ is the independent variable, including the primary term of the structure hole and centrality of the inventor cooperation network, as well as the quadratic term. $M_{it}$ is the mediator variable. $Z_{it}$ is the control variable including the organizational year and R & D intensity.

4. Results

4.1. Inventor Cooperation Network Position Recognition

In this paper, the inventor information of patent data is extracted on the basis of crawling, cleaning and screening patent data. The structure hole and centrality index of the inventor network are calculated. Meanwhile, this paper makes the inventor information into a co-occurrence matrix and imports it into Gephi software to more easily identify the inventor cooperation network position. This also helps us to understand the relationship between inventor cooperation network location and ambidextrous innovation in the later sections. The results of partial inventor cooperation network visualization are shown in Figure 3, where the node is the inventor, and the ties are the collaboration in previous inventions. If two inventors are jointly involved in a patented invention, they have a network connection. The centrality values of Figure 3A–D are 9.993, 9.311, 3.455 and 2.308, respectively. We can find that in the network centrality, institution A is more centralized than the other three institutions, with characteristics such as close communication of technology and knowledge. While the structural hole index of each organization is 1.246, 1.297, 1.340 and 1.572, respectively, institution D has the fewest network constraints. It can be seen that the inventors’ network centrality and structural holes are separated, and there are differences in the network position.

4.2. Descriptive Statistics

Table 2. Descriptive statistics.

	ERA	EIT	INSH	INCE	UTV	RTV	AGE	RD
ERA	1.000
EIT	0.669 ***	1.000
INSH	−0.108 **	−0.088 *	1.000
INCE	0.254 ***	0.315 ***	−0.169 ***	1.000
UTV	0.486 ***	0.487 ***	−0.216 ***	0.293 ***	1.000
RTV	0.288 ***	0.435 ***	−0.085 *	0.163 ***	0.531 ***	1.000
AGE	0.418 ***	0.457 ***	−0.209 ***	0.173 ***	0.595 ***	0.475 ***	1.000
RD	0.455 ***	0.818 ***	−0.111 **	0.307 ***	0.539 ***	0.477 ***	0.575 ***	1.000
MEAN	7.71	8.35	1.45	6.89	0.96	0.19	4.93	10.84
SD	12.65	20.15	0.34	7.90	0.83	0.32	3.70	21.72
VIF			1.08	1.16	1.94	1.54	1.88	1.79

Note: * p < 0.1; ** p < 0.05; *** p < 0.01.

shows the descriptive statistics and multicollinearity results of each variable. According to Table 2, it can be seen that there is no significant difference between the mean values of the ambidextrous innovation. This shows that at present, AI enterprises not only pay attention to exploitative innovation to strengthen existing capabilities, but also pay attention to exploratory innovation to produce new capabilities. Therefore, this field is suitable to be used as a data source for this article. Table 2 also tests the VIF values. The VIF values of all variables were found to be below 2, which is much smaller than the threshold value of 5, indicating that there is no multicollinearity problem.

4.3. Regression Analysis and Effect Test

4.3.1. Total Effect Test

In this part, negative binomial regression is used for analysis, and the results of regression analysis are shown in

Table 3. The effect of inventor cooperation network on ambidextrous innovation.

Variable	ERA			EIT
	M1	M2	M3	M4	M5	M6
RD	0.017 **	0.013 ***	0.009 **	0.044 ***	0.036 ***	0.033 ***
	(2.470)	(2.800)	(2.300)	(4.810)	(5.190)	(4.150)
AGE	0.097 ***	0.071 ***	0.093 ***	0.063 **	0.047	0.065 **
	(3.340)	(2.780)	(3.610)	(2.090)	(1.640)	(2.090)
INCE			0.092 ***			0.081 ***
			(4.060)			(3.830)
INCE2			−0.001 ***			−0.001 ***
			(−3.470)			(−3.00)
INSH		10.726 ***			10.775 ***
		(5.390)			(5.110)
INSH2		−3.648 ***			−3.722 ***
		(−5.510)			(−5.350)
_CONS	1.167 ***	−6.204 ***	0.689 ***	0.724 ***	−6.563 ***	0.304
	(7.560)	(−4.530)	(5.100)	(4.170)	(−4.470)	(1.630)
Log likelihood	−1153.020	−1134.029	−1136.840	−1035.100	−1017.295	−1023.005
Wald chi2	78.890	164.620	150.900	66.900	205.260	129.890
Prob > chi2	0	0	0	0	0	0
R-sq	0.038	0.054	0.052	0.088	0.104	0.099

Note: t-values are in parentheses; ** p < 0.05; *** p < 0.01.

. M1 and M2 show that the coefficient of INSH is 10.726, which is significant at the 1% level. Meanwhile, the square coefficient of INSH is −3.648, which is significant at the 1% level. The results show that there is an inverted U-shaped relationship between the inventor cooperation network structure hole and exploratory innovation, and H1a passes the test. M1 and M3 show that the coefficient of INCE is 0.092, which is significant at the 1% level. Meanwhile, the square coefficient of INCE is −0.001, which is significant at the 1% level. The results show that there is an inverted U-shaped relationship between the inventor cooperation network centrality and exploratory innovation, and H2a passes the test. M4 and M5 show that the coefficient of INSH is 10.775, which is significant at the 1% level. Meanwhile, the square coefficient of INSH is −3.722, which is significant at the 1% level. The results show that there is an inverted U-shaped relationship between the inventor cooperation network structure hole and exploitative innovation, and H1b passes the test. M4 and M6 show that the coefficient of INCE is 0.081, which is significant at the 1% level. Meanwhile, the square coefficient of INCE is −0.001, which is significant at the 1% level. The results show that there is an inverted U-shaped relationship between the inventor cooperation network centrality and exploitative innovation, and H2b passes the test.

4.3.2. Mediation Effect Test

To test the mediating role of technological knowledge base variety between the inventor cooperation network and ambidextrous innovation, the mediating effect test proposed by Wen et al. [72] is adopted in this paper. First, it examines the effect of the inventor cooperation network on the related variety of technological knowledge base. According to M1 in

Table 4. The mediating effects of related variety of technological knowledge base.

Variable	RTV		ERA			EIT
	M1	M2	M3	M4	M5	M6	M7	M8
RD	0.005 **	0.005 ***	0.014 ***	0.008 **	0.012 ***	0.038 ***	0.028 ***	0.034 ***
	(2.550)	(2.840)	(−2.780)	(2.100)	(2.670)	(5.190)	(3.730)	(4.890)
AGE	0.130 ***	0.108 ***	0.092 ***	0.091 ***	0.068 **	0.041	0.049	0.032
	(7.330)	(5.780)	(3.350)	(3.370)	(2.530)	(1.450)	(1.560)	(1.110)
INCE	0.014 **			0.091 ***			0.080 ***
	(2.400)			(4.090)			(3.910)
INCE2				−0.001 ***			−0.001 ***
				(−3.47)			(−2.94)
INSH		9.652 ***			10.647 ***			10.394 ***
		(3.180)			(5.430)			(4.960)
INSH2		−3.216 ***			−3.625 ***			−3.603 ***
		(−3.240)			(−5.550)			(−5.200)
RTV			1.653 ***	0.162	0.193	2.623 ***	0.645 **	0.468 *
			(2.580)	(0.760)	(0.880)	(3.970)	(2.160)	(1.650)
RTV2			−1.389 ***			−2.014 ***
			(−2.900)			(−3.860)
_CONS	−2.667 ***	−9.400 ***	1.094 ***	0.687 ***	−6.147 ***	0.639 ***	0.305	−6.276 ***
	(−17.060)	(−4.180)	(8.240)	(5.090)	(−4.540)	(3.660)	(1.610)	(−4.300)
Log likelihood	−170.468	−168.030	−1147.930	−1136.625	−1133.721	−1024.620	−1020.434	−1015.873
Wald chi2	159.700	147.770	112.040	153.760	166.940	99.990	145.890	225.710
Prob > chi2	0	0	0	0	0	0	0	0
R-sq	0.111	0.124	0.043	0.052	0.055	0.097	0.101	0.105

Note: t-values are in parentheses; * p < 0.1; ** p < 0.05; *** p < 0.01.

, the coefficient of INCE is 0.014, which is significant at the 5% level. The results show that the inventor cooperation network centrality has a positive impact on the related variety of the technological knowledge base, and H3a passes the test. M2 shows that the coefficient of INSH is 9.652, which is significant at the 1% level. Meanwhile, the square coefficient of INSH is −3.216, which is significant at the 1% level. The results show that there is an inverted U-shaped relationship between the inventor cooperation network structural hole and the related variety of technological knowledge base, and H4a passes the test. Consequently, the test results combined with the direct effects satisfy the valid premise of the existence of mediating effects.

M3, M4 and M5 in Table 4 are to test the mediating effect of a related variety of technological knowledge base between the inventor cooperation network and exploratory innovation. M4 shows that the coefficient of INCE is 0.091, which is significant at the 1% level. Meanwhile, the square coefficient of INCE is −0.001, which is significant at the 1% level. The coefficient of RTV is 0.162 but not significant. Combined with the M3 test results in Table 3, it shows that the related variety of technological knowledge base does not plays a mediator role between the inventor cooperation network centrality and exploratory innovation, and H5a failed the test. M5 shows that the coefficient of INSH is 10.647, which is significant at the 1% level. Meanwhile, the square coefficient of INSH is −3.625, which is significant at the 1% level. The coefficient of RTV is 0.193 but not significant. Combined with the M2 test results in Table 3, this shows that the related variety of technological knowledge base does not play a mediator role between the inventor cooperation network structure hole and exploratory innovation, and H6a failed the test.

M6, M7 and M8 in Table 4 are used to test the mediating effect of a related variety of technological knowledge base between the inventor cooperation network and exploitative innovation. M7 shows that the coefficient of INCE is 0.080, which is significant at the 1% level. Meanwhile, the square coefficient of INCE is −0.001, which is significant at the 1% level. The coefficient of RTV is 0.645, and it is significant at the 5% level. Combined with the M6 test results in Table 3, the primary coefficients of INCE is reduced and still significant, indicating that the related variety of technological knowledge base plays a partially mediating role between the inventor cooperation network centrality and exploitative innovation. H5b passed the test. M8 shows that the coefficient of INSH is 10.394, which is significant at the 1% level. Meanwhile, the square coefficient of INSH is −3.603, which is significant at the 1% level. The coefficient of RTV is 0.468, and it is significant at 10% level. Combined with the M5 test results in Table 3, the absolute values of primary and quadratic coefficients of INSH are reduced and still significant, indicating that the related variety of technological knowledge base plays a partially mediating role between the inventor cooperation network structural hole and exploitative innovation. H6b passed the test.

Secondly, this study examines the influence of the inventor cooperation network on the unrelated variety of technological knowledge base. According to M1 in

Table 5. The mediating effects of unrelated variety of technological knowledge base.

Variable	UTV		ERA			EIT
	M1	M2	M3	M4	M5	M6	M7	M8
RD	0.003 **	0.004 ***	0.005 *	0.002	0.005 **	0.029 ***	0.024 ***	0.027 ***
	(2.210)	(3.010)	(1.780)	(0.860)	(2.020)	(4.810)	(4.170)	(5.240)
AGE	0.093 ***	0.068 ***	0.050 *	0.051 **	0.040	−0.002	0.010	−0.001
	(10.000)	(7.400)	(1.840)	(1.990)	(1.620)	(−0.060)	(0.320)	(−0.050)
INCE	0.016 ***			0.068 ***			0.051 ***
	(4.590)			(3.610)			(3.110)
INCE2				−0.001 ***			−0.001 **
				(−3.000)			(−2.080)
INSH		8.067 ***			7.868 ***			7.948 ***
		(6.980)			(4.290)			(3.870)
INSH2		−2.840 ***			−2.661 ***			−2.735 ***
		(−7.420)			(−4.320)			(−3.960)
RTV
RTV2
UTV			0.633 ***	0.543 ***	0.500 ***	0.721 ***	0.626 ***	0.572 ***
			(5.090)	(5.050)	(4.200)	(5.680)	(5.380)	(4.390)
_CONS	−0.753 ***	−5.985 ***	0.839 ***	0.528 ***	−4.538 ***	0.424 **	0.181	−4.925 ***
	(−10.320)	(−7.030)	(6.360)	(3.850)	(−3.570)	(2.490)	(0.930)	(−3.530)
Log likelihood	−430.016	−417.5485	−1131.382	−1120.984	−1121.194	−1014.734	−1008.464	−1005.036
Wald chi2	236.770	267.290	156.130	221.320	210.330	166.850	246.380	288.020
Prob > chi2	0	0	0	0	0	0	0	0
R-sq	0.102	0.128	0.057	0.065	0.065	0.106	0.112	0.115

Note: t-values are in parentheses; * p < 0.1; ** p < 0.05; *** p < 0.01.

, the coefficient of INCE is 0.016, which is significant at the 1% level. The results show that the inventor cooperation network centrality has a positive impact on the unrelated variety of technological knowledge base, and H3b passes the test. M2 shows that the coefficient of INSH is 8.067, which is significant at the 1% level. Meanwhile, the square coefficient of INSH is −2.840, which is significant at the 1% level. The results show that there is an inverted U-shaped relationship between the inventor cooperation network structural hole and the unrelated variety of technological knowledge base, and H4b passes the test. Consequently, the test results combined with the direct effects satisfy the valid premise of the existence of mediating effects.

M3, M4 and M5 in Table 5 are used to test the mediating effect of the unrelated variety of technological knowledge base between the inventor cooperation network and exploratory innovation. M4 shows that the coefficient of INCE is 0.068, which is significant at the 1% level. Meanwhile, the square coefficient of INCE is −0.001, which is significant at the 1% level. The coefficient of UTV is 0.543, and it is significant at the 1% level. Combined with the M3 test results in Table 3, the primary coefficients of INCE is reduced and still significant, indicating that the unrelated variety of technological knowledge base plays a partially mediating role between the inventor cooperation network centrality and exploratory innovation. H5c passed the test. M5 shows that the coefficient of INSH is 7.868, which is significant at the 1% level. Meanwhile, the square coefficient of INSH is −2.661, which is significant at the 1% level. The coefficient of UTV is 0.500, and it is significant at the 1% level. Combined with the M2 test results in Table 3, the absolute values of primary and quadratic coefficients of INSH are reduced and still significant, indicating that the unrelated variety of technological knowledge base plays a partially mediating role between the inventor cooperation network structural hole and exploratory innovation. H6c passed the test.

M6, M7 and M8 in Table 5 are used to test the mediating effect of unrelated variety of technological knowledge base between inventor cooperative network and exploitative innovation. M7 shows that the coefficient of INCE is 0.051, which is significant at the 1% level. Meanwhile, the square coefficient of INCE is −0.001, which is significant at the 5% level. The coefficient of UTV is 0.626, and it is significant at the 1% level. Combined with the M6 test results in Table 3, the primary coefficients of INCE is reduced and still significant, indicating that the unrelated variety of technological knowledge base plays a partially mediating role between the inventor cooperation network centrality and exploitative innovation. H5d passed the test. M8 shows that the coefficient of INSH is 7.948, which is significant at the 1% level. Meanwhile, the square coefficient of INSH is −2.735, which is significant at the 1% level. The coefficient of UTV is 0.572, and it is significant at the 1% level. Combined with the M5 test results in Table 3, the absolute values of primary and quadratic coefficients of INSH are reduced and still significant, indicating that the unrelated variety of technological knowledge base plays a partially mediating role between the inventor cooperation network structural hole and exploitative innovation. H6d passed the test.

4.4. Robustness Tests

In order to ensure the reliability of the above inverted U relationship and according to the test method of inverted U-shaped relations proposed by Hanns et al. (2016), this paper makes a further test and analysis of the inverted U-shaped relation [73]. Hanns et al. (2016) believe that the inverted U-shaped relationship between variables needs to meet three conditions [73]. First, the square coefficient of the independent variable must be negative and significant. Second, when the independent variable takes the maximum value (minimum value), the slope must be negative (positive). Finally, the turning point of the inverted U-shape must lie within the range of the independent variable. From M3 in Table 3, it can be seen that the square coefficient of the inventor cooperation network centrality is negative and significant, which satisfies the first condition. Then, the curve slope equation of the inventor cooperation network centrality to the enterprise exploratory innovation is calculated as S = −0.002X + 0.092 as the value of the inventor cooperation network ranges from 0 to 84.711. Therefore, when the inventor cooperation network centrality takes its maximum value, S is negative. S is positive when the inventor cooperation network centrality takes its minimum value. It satisfies the second condition. The turning point of X is $-{\beta }_1/\left(2{\beta }_2\right)=46$, which is in the range of the inventor cooperation network centrality. Therefore, the three conditions of the inverted U-shaped relationship are all satisfied, and H2a passes the test again. So H1a, H1b, H2a, H2b, H4a and H4b of this paper were all verified again. In addition, the Poisson model was used to verify the reliability of the results, and the direction of all regression results is significant and consistent with the original regression. This shows that the empirical results of this paper have good robustness.

5. Discussion

This study’s purpose was to understand how the inventor’s cooperation network adds to the development of ambidextrous innovation via technological knowledge base variety. To this purpose, this study linked the inventor’s cooperation network to technological knowledge base variety and ambidextrous innovation. Moreover, the mediation effect of technological knowledge base variety between the inventor’s cooperation network and ambidextrous innovation was also tested. We found some interesting phenomena.

Firstly, there is an inverted U-shaped relationship between structural holes and centrality of inventor cooperation networks and ambidextrous innovation. The improvement in the centrality and structural hole of the inventor cooperation network does not necessarily lead to the enhancement of the firm’s ambidextrous innovation. When it exceeds the appropriate level, the increase in the centrality and structural holes of the inventors’ cooperative network will reduce the ambidextrous innovation ability. Stakeholders believe that the negative effects of problems such as information overload and reduced recombination potential brought about by the inventor’s cooperative network outweigh the positive effects of the cooperation network. It is concluded that there is an inverted U-shaped curve relationship between inventor cooperative network and ambidextrous innovation.

Secondly, inventor cooperation network centrality has a positive effect on technological knowledge base variety. There is an inverted U-shaped relationship between inventor collaborative network structure holes and technological knowledge base variety. In the network, inventors at the central location acquire heterogeneous knowledge from partners, which not only meets the resource needs of enterprises, but also provides opportunities to understand different information. This enhances the understanding of knowledge, know-how and skills, which in turn increases the firm’s technological base variety. The inventor occupying the structural hole position can efficiently access the disconnected inventors in the network to obtain high-quality information and non-redundant resources from them. There are also more opportunities for creative knowledge recombination and enhancing the ability to communicate and sharing knowledge with colleagues to increase the firm’s knowledge base variety. However, when the knowledge base between the inventor and the remote partner is not highly overlapping, it may cost a lot of resources and energy to integrate the two knowledge bases to recombine innovation. This increases the uncertainty and risk of failure of enterprise knowledge integration, which is not conducive to the development of technological knowledge base variety.

Thirdly, the unrelated variety of technological knowledge base mediates the relationship between the inventor cooperation network and ambidextrous innovation. The related variety of technological knowledge base mediates the relationship between the inventor cooperation network and exploitative innovation. However, the related variety of technological knowledge base cannot play a mediating role between inventor cooperation network and exploratory innovation. The possible reason is that exploratory innovation emphasizes the acquisition of new knowledge to increase the innovation potential of enterprises, while the related variety of technological knowledge base involves knowledge within the same discipline. Although it will accumulate novel and diverse knowledge, it can play a role in the emerging stage of exploratory innovation in business development. Yet, when the exploratory innovation of enterprises reaches a certain level, the new knowledge provided by the relevant diversity of technological knowledge base is far from meeting the needs of exploratory innovation. In addition, collaborative networks reflect an important process in which inventors continuously exchange new knowledge with each other, provide complementary knowledge, and improve the potential of their knowledge combinations. This may provide new knowledge that is far greater than the knowledge that can be provided by the related variety of technological knowledge base. Thus, the related variety of technological knowledge base fails to play a mediating effect in the inventor collaborative network and exploratory innovation.

5.1. Managerial Implications

First, in the current competitive market environment, it is necessary for AI enterprises to rely on the collaborative network to improve their ambidextrous innovation capability. On the one hand, enterprises should encourage inventors to cooperate with inventors who have no or less contact with each other so that enterprises can obtain differentiated information and shape their own unique sustainable competitive advantages. However, enterprises should also pay attention to the increase in heterogeneous knowledge and connections, which takes a certain amount of time to integrate and transfer to the firm’s ambidextrous innovation. Firms should maintain an appropriate level of structural holes to avoid the negative effects of “over-embedding”. On the other hand, there is a threshold for the inventor cooperation network centrality. When strengthening the relationship between inventors and partners, AI enterprises should not only ensure that enterprises can obtain relevant new information and technologies through the information advantage of the central position, enhance sustainable competitiveness and generate comparative advantages, but at the same time, the excessive construction of cooperative relationships should be avoided as this causes knowledge redundancy, which will eventually lead to a decrease in the innovation capability of the enterprise. Therefore, it is necessary for AI enterprises to cultivate the corresponding network management ability, improve the effective use of resources and strengthen the achievements of innovation.

Secondly, AI enterprises should focus on the matching of inventors’ cooperative networks with the diversity model of technological knowledge, and pay attention to the positive role of collaborative network embedding. There are different factors in the inherent mechanisms of the influence of inventors’ cooperative network centrality and structural holes on ambidextrous innovation. While choosing the appropriate network position according to their own innovation behavior, enterprises also need to dynamically adjust the construction of technological knowledge base variety. On the one hand, when enterprises focus on exploratory innovation, enterprises need to pay attention to the exploration and excavation of new technology fields in order to effectively enhance the novelty of its recombinant technology or knowledge, resulting in high-value innovation. On the other hand, when the enterprise focuses on the exploitative innovation, it can develop its technological knowledge base in many ways and make full use of its knowledge to enhance the exploitative innovation.

5.2. Theoretical Implications

Firstly, this paper contributes to expanding the theory of ambidextrous innovation by revealing inventor cooperation networks as a key factor for firms to implement ambidextrous innovation activities. Prior research has mainly studied the impact of firm collaboration networks (macro) and knowledge networks (micro) on ambidextrous innovation but rarely explored the relationship between meso collaboration networks—the cooperation between inventors and ambidextrous innovation [31]. This paper complements the relevant research from the perspective of inventor cooperation networks. It also verifies the inverted U-shaped relationship between inventor cooperation networks and ambidextrous innovation and provides new insights into whether firms can improve innovation outputs through cooperation networks, especially the inverted U-shaped relationship between structural holes and exploratory innovation. Therefore, this paper not only enriches the research framework of the relationship between the existing collaboration network and enterprise ambidextrous innovation, but also strengthens and complements the new discovery of structural network embedding in the social network literature.

Secondly, this paper takes the technological knowledge base variety as a key mediating factor, which enriches the research on the influence of cooperation network on ambidextrous innovation. Most of the existing literature focuses on the direct effects and boundary conditions of cooperation networks on ambidextrous innovation [6], with little research discussing its potential mechanism. This study introduces the relevant theory of knowledge base into the research of ambidextrous innovation path and reveals the key path of inventor cooperation network position in driving enterprise ambidextrous innovation. It both extends our understanding of the importance of distinguishing between related and unrelated variety of technological knowledge bases in partnerships and enriches the literature on knowledge management within the partnership framework.

Thirdly, this paper uses Chinese AI industry data for hypothesis testing. This reflects the dynamic change in the enterprise’s technological knowledge base, showing the development process of inventors from external knowledge acquisition to the cultivation of technological knowledge base variety to ambidextrous innovation capabilities. So, the research results can highlight the path of knowledge flow and innovation and once again prove that knowledge is the most important factor endowment of enterprises. This study also enriches the research results on the relationship between inventor cooperation network and ambidextrous innovation in the Chinese context.

6. Conclusions

The basic objective of the study was to investigate the role of the inventor’s cooperation network and technological knowledge base variety in the improvement of the ambidextrous innovation of China’s AI enterprises. Furthermore, the current study also highlighted the mediation effect of technological knowledge base variety between the inventor’s cooperation network and ambidextrous innovation. This study’s findings revealed that the inventor’s cooperation network has an inverted U-shaped relationship with the ambidextrous innovation. The findings verified that the unrelated variety of technological knowledge base plays a mediating role in the linkage between the inventor’ cooperation network and ambidextrous innovation. The related variety of technological knowledge base mediates the relationship between the inventor cooperation network and exploitative innovation. However, the related variety of technological knowledge base plays no mediating role between the inventor’s cooperation network and exploratory innovation.

This study had a few limitations that might provide future research directions. First, this study is limited to samples of the Chinese AI industry, and its applicability to other industries needs to be verified. Future research can expand the range of samples and increase the universality of the results. Second, this study focuses on the mediating role of technological knowledge base variety in the inventor cooperation network position and ambidextrous innovation. Future research can divide the knowledge acquired by inventors through collaborative network position into external knowledge and internal knowledge and further explore other mediating or moderating factors that influence this relationship.