The application of generative artificial intelligence (GenAI) has become a critical pathway for reshaping enterprise production management structures and enhancing competitiveness. Existing literature highlights its dual role in reducing R&D costs,automating workflows,and facilitating human-machine collaboration. While intelligent technologies are generally expected to drive productivity growth,there is a lack of systematic explanation of how generative artificial intelligence affects enterprise total factor productivity (TFP),with most mechanism analyses remaining at the level of technology-enablement hypotheses. The transmission path of "technology application → factor reorganization → efficiency leap" has yet to be empirically tested,and the "black box" between technological advantage and productivity improvement remains unexplored.
Therefore,this study selects Chinese A-share manufacturing companies from 2017 to 2022 as samples and conducts an in-depth investigation into the mechanism of generative artificial intelligence in this critical sector. The study employs BERT intelligent text analysis to construct measurement indicators for GenAI adoption and systematically examines its impact on total factor productivity and underlying mechanisms. Baseline regression reveals that GenAI application significantly improves TFP. The promotion effect essentially reflects comprehensive outcomes of expanded enterprise resource bases,improved key resource allocation,and enhanced resource value release capabilities. Mechanism analysis identifies three critical channels:at the strategic resource level,GenAI optimizes R&D configuration; at the operational resource level,it improves internal management efficiency; and at the production factor resource level,it enhances data element utilization. These three layers collectively drive TFP improvement.
Moderation analysis demonstrates that the relationship between GenAI and TFP is contingent on contextual conditions. Specifically,human capital upgrading within enterprise organizations and digital infrastructure improvement outside organizations exert positive moderating effects,amplifying AI's productivity benefits. Heterogeneity analysis reveals significant differences across enterprise characteristics and regional contexts. Regarding enterprise heterogeneity,while firms with different property rights all achieve productivity improvements when applying GenAI,the efficacy strongly depends on contextual features. The effects are more pronounced in non-labor-intensive,non-asset-intensive,and technology-intensive enterprises,whereas relatively limited in asset-intensive and labor-intensive firms. This indicates that GenAI functions as a knowledge-enabling technology that synergizes with flexible production structures and innovation activities,complementing high-skilled human capital and high-intensity innovation activities rather than simply replacing procedural labor. Regarding regional heterogeneity,significant effects in eastern and central regions confirm the supporting role of mature economic ecosystems,while the insignificant effect in the western region warns of potential "lag effects" in technology diffusion. Furthermore,when regions with high innovation capabilities apply GenAI,the impact on manufacturing TFP becomes more significant. Notably,regions with low levels of digital economic development exhibit a "filling effect",providing optimistic practical possibilities and theoretical inspirations for latecomers to achieve "technological leapfrogging" through strategic AI adoption.
These findings provide empirical support for understanding how "artificial intelligence +" promotes manufacturing productivity. They verify that GenAI enhances TFP while revealing the internal mechanisms and practical boundaries of its effects. Specifically,GenAI is not a universally applicable and automatically effective "universal technology",but rather a strategic resource that can fully release its efficacy only under appropriate resource bases and environmental conditions. The productivity gains depend on whether AI can form effective coupling with enterprises' internal knowledge resources,organizational resources,and data resources. This conclusion offers explanatory evidence and theoretical inspiration for understanding how GenAI enables the real economy,promotes transformation and upgrading of manufacturing industries,and cultivates new quality productive forces.

With the rapid advancement of artificial intelligence as a general-purpose technology, innovation processes are increasingly embedded in open, networked, and collaborative ecosystems. Compared with traditional proprietary R&D models, AI innovation is characterized by high knowledge intensity, technological complexity, cumulative learning, and strong interdependence among heterogeneous actors. Open-source communities have therefore become critical organizational infrastructures for AI development. However, although existing research has examined open innovation and digital collaboration, the mechanisms through which different participating actors influence innovation performance in AI-oriented open-source projects remain insufficiently clarified. This study investigates how demand-side user participation and supply-side developer engagement affect innovation performance in open-source AI projects and whether their interaction generates complementary or nonlinear effects.
Drawing on open innovation theory and the knowledge production function, this study develops an analytical model linking heterogeneous participant behavior to project-level innovation outcomes. Users enhance problem articulation and market relevance, while developers improve technical depth and iterative efficiency. The study posits that while user-developer interaction fosters synergistic knowledge recombination, excessive heterogeneity may increase coordination costs. Empirically, the study analyzes a panel of 3.98 million GitHub projects (2009–2025), identifying AI-related initiatives via domain classification. Innovation performance is measured by star growth, fork activity, and sustained contributions. Core explanatory variables quantify user engagement through issue submissions and feedback interactions, and developer participation through active contributors and commit frequency. The empirical strategy employs fixed-effects regression models to control for unobserved project heterogeneity, complemented by robustness checks incorporating lagged variables and instrumental variable approaches to mitigate potential endogeneity concerns.
The empirical results reveal several critical findings. First, both user participation and developer engagement significantly enhance innovation performance, but their functional roles differ markedly: user demand exhibits approximately threefold greater impact than developer supply in AI-specific projects (coefficients: 0.258 vs. 0.066), underscoring the dominance of demand-pull mechanisms in data-driven innovation. Second, the interaction between users and developers exhibits an inverted-U relationship: within moderate ranges, participation generates complementary effects, but when both reach high levels simultaneously, a significant negative interaction emerges, indicating a "crowding effect" where coordination costs outweigh synergistic benefits-a manifestation of the "openness paradox". Third, heterogeneity analysis demonstrates that these effects are attenuated by technical complexity (programming language count), which increases coordination frictions, and moderated by license restrictiveness, which enhances developer incentives but impedes user participation. AI open-source projects show stronger demand-pull and supply-driven effects than other sectors, with demand-pull being more pronounced. NLP and multimodal projects exhibit higher demand sensitivity than computer vision, while developer supply benefits multimodal most, reflecting data-driven and scenario-driven characteristics.
This study contributes to the literature in several respects. First, it advances open innovation research by shifting the analytical focus from firm-level boundary management to ecosystem-level actor interaction within the context of AI technological paradigms. Second, by employing large-scale digital trace data, it improves empirical identification and external validity beyond small-sample case studies. Third, it extends the knowledge production framework to digital collaborative environments, demonstrating how heterogeneous inputs interact under conditions of technological complexity. Finally, the findings provide policy implications for digital innovation governance, suggesting that platform operators and policymakers should design institutional mechanisms that balance participation incentives, reduce coordination frictions, and enhance collaborative efficiency in AI ecosystems.

With the rapid advancement and pervasive integration of Artificial Intelligence (AI) across economic, political, military, and environmental domains, AI technologies are not only empowering societal transformation but also introducing novel risks to governance and human development. The governance and control of these risks have become a critical issue influencing social development. Against this backdrop, this study aims to critically examine the current cognitive paradigms used to understand the social risks arising from AI applications. Prevailing research is often trapped in the dichotomy between technological determinism and social constructivism, leading to fragmented and partial understandings. By introducing the lens of knowledge production theory, this study seeks to transcend these paradigmatic limitations. It endeavors to construct a more integrated and explanatory framework for comprehending the social risks of AI application at ontological, epistemological, and methodological levels, thereby contributing to a more robust foundation for risk identification, interpretation, and governance.
This study adopts a qualitative, theory-building approach grounded in the synthesis and critical analysis of extant literature. It systematically reviews and deconstructs the current body of knowledge on AI application social risks, identifying dominant paradigms, thematic evolution, and methodological trends through conceptual analysis. The core analytical framework is derived from knowledge production theory, which is used to interrogate the knowledge system concerning AI risks by examining elements such as knowledge producers, instruments of production (research methods), and productive relations (academic ecology). To ground the theoretical discussion and demonstrate the proposed framework's utility, the study conducts an in-depth secondary analysis of two representative case studies in the domain of autonomous driving, involving Tesla vehicles in Florida(2016) and California (2018-2019). These cases are analyzed through the newly constructed framework to illustrate the multidimensional and dynamic nature of AI application risks.
The analysis reveals serious systemic flaws in how the social risks of AI applications are understood from the perspective of knowledge production. Ontologically, research on this topic is fragmented: technological determinism foregrounds inherent technical flaws such as algorithmic bias and data errors, whereas social constructivism emphasizes algorithmic myths and ethical conflicts; this dualism results in a reductive and static view of risk that overlooks its emergent nature arising from the dynamic interplay among technology, institutions, and human cognition. Epistemologically, deep disciplinary silos separate computer science, management, law, and the humanities, leading to disconnected research themes, non-communicating discourses, and a lack of genuine interdisciplinary collaboration, further exacerbated by the monopolization of academic capital like computing power and data, path dependency in disciplinary practices, and evaluation systems rooted in Utility theory. Methodologically, quantitative studies often exhibit "computational worship" by neglecting social context, while qualitative studies frequently lack sufficient empirical grounding, and integrated mixed-methods approaches remain rare and poorly executed.
To address these issues, the study proposes a comprehensive framework for transforming the cognitive pathway toward understanding AI-related social risks. Ontological reconstruction calls for shifting from a substantialist to a relational ontology, reconceptualizing risks as emergent phenomena produced through the dynamic coupling of embodied technology (which reshapes human perception and action), embedded institutions (which provide constraints and possibilities), and situated social contexts (which confer meaning). Epistemological transformation involves fostering collaborative knowledge production through a reflexive turn in technical sciences,such as embedding value-sensitive design,engaged research in the humanities and social sciences like lab ethnography, and the development of interface governance paradigms in management science to evaluate the alignment between technology and socio-organizational systems. Methodological innovation entails building a dynamic governance framework that includes multidimensional monitoring, adaptive assessment using dialectical models encompassing technical, normative, and interpretive dimensions, agile response through a spectrum of intervention strategies, and organizational learning via double- and triple-loop mechanisms. Institutional reformation aims to revolutionize knowledge production relations by redistributing academic capital to break monopolies over computation and data, cultivating interdisciplinary hybrid cultures, and establishing experimental governance regimes such as regulatory sandboxes and scenario-based testing. A case analysis of autonomous driving accidents vividly demonstrates these dynamics, revealing how risks emerge from the coupling of environment, technology, and perception, how knowledge claims are contested among manufacturers, regulators, and users, and how methodological and institutional gaps impede effective risk governance.

In the digital economy era, platform ecosystems have emerged as pivotal arenas for entrepreneurial innovation, particularly for platform-dependent startups, emerging firms that heavily rely on digital platforms for traffic, users, channels, and infrastructure to conduct business and develop products. These startups face a unique dual-identity dilemma: they act as complementors within the platform ecosystem while striving for independent growth. This tension often leads to innovation barriers, such as resource coordination difficulties, cognitive convergence, and over-reliance on platform rules, which complicate new product development (NPD). Despite the growing importance of ecosystem embeddedness, the process by which these firms integrate into platform networks to share resources and collaborate, existing research has largely overlooked its dynamic impacts on NPD in this context. Prior studies tend to adopt static perspectives, focusing on platform leaders or strong complementors, while neglecting the challenges faced by resource-constrained, peripheral startups. Moreover, they fail to explore nonlinear mechanisms and the role of iterative processes in overcoming uncertainties.
This study addresses these gaps by investigating how ecosystem embeddedness influences NPD performance in platform-dependent startups, adopting an opportunity iteration perspective. Opportunity iteration refers to the dynamic, nonlinear evolution of entrepreneurial opportunities through continuous adjustment, optimization, and upgrading based on feedback loops, stakeholder interactions, and environmental changes. Drawing on an integrated theoretical framework that combines embeddedness theory, dynamic capabilities theory, entrepreneurial opportunity theory, and complex adaptive systems theory, the study proposes a nonlinear model of "ecosystem embeddedness-opportunity iteration-NPD performance" This framework posits that embeddedness enables resource sharing and multi-stakeholder collaboration but can turn counterproductive at excessive levels. The study hypothesizes an inverted U-shaped relationship between embeddedness and NPD performance, mediated by opportunity iteration, with entrepreneurial learning and entrepreneurial networks as moderators.
To test these hypotheses, the study employed an empirical approach using survey data from 405 platform-dependent startups in China, reliant on major digital platforms. Samples were collected via incubators and alumni networks in multiple economic regions, targeting mid-to-senior managers. Variables were measured using established Likert-scale items: ecosystem embeddedness (4 items); NPD performance (5 items); opportunity iteration (5 items); entrepreneurial learning (10 items); and entrepreneurial networks (2 items). Firm age and size served as controls. Reliability and validity were confirmed through standard statistical tests, with common method bias appropriately addressed.
The hierarchical regression results are summarized. First, ecosystem embeddedness exhibits a significant inverted U-shaped effect on NPD performance, supporting H1. Moderate embeddedness facilitates resource access and innovation synergies, but over-embeddedness induces path dependence and conflicts. Second, embeddedness similarly affects opportunity iteration in an inverted U-shape, while opportunity iteration positively impacts NPD, confirming H2a and H2b. Opportunity iteration mediates this relationship nonlinearly (H2c), with Bootstrap tests showing significant direct (0.2185) and indirect (0.1740) effects (95% CI excluding zero). Third, entrepreneurial learning moderates the embeddedness-opportunity iteration link, steepening the inverted U-curve, validating H3 by amplifying positive effects through knowledge internalization. However, entrepreneurial networks do not significantly moderate embeddedness-NPD, rejecting H4, possibly due to network paradoxes like redundancy in platform-constrained contexts. Robustness checks affirmed the results.
This research extends embeddedness theory to digital platforms by highlighting its nonlinear "double-edged sword" nature in resource-constrained startups, integrating complex adaptive systems to explain self-organization and emergence. By introducing opportunity iteration as a nonlinear mediator, it advances entrepreneurial opportunity theory beyond static views, linking it with dynamic capabilities to elucidate how iterations bridge embeddedness to innovation under uncertainty. The mixed moderating effects refine boundary conditions, challenging network theory's universality and emphasizing contextual heterogeneity. Therefore, firms should monitor embeddedness levels, foster iterative systems, and prioritize learning programs. Platform leaders can offer modular interfaces, while policymakers might subsidize multi-platform strategies to mitigate dependencies.

Research universities serve as a core engine of scientific and technological innovation and constitute a primary source of high-value patents and technology transfer. In China, however, a persistent challenge remains: while invention patent applications from universities continue to grow, the rate of successful industrialization lags significantly behind expectations and international benchmarks. This gap underscores an urgent need to shift the patent management paradigm from a purely volume-driven model toward a framework that prioritizes quality and latent commercial value. National policy initiatives have explicitly called for the effective screening of high-value patents from vast university portfolios. Yet, accurately and efficiently identifying such assets within extensive patent databases remains a formidable methodological undertaking. This study addresses this gap by developing and validating a data-driven identification model tailored to the context of research universities, thereby seeking to offer both theoretical refinement and practical support for more targeted patent management and technology transfer strategies.
This study proposes a hybrid modeling approach that integrates Genetic Algorithm (GA), Rough Set Theory (RST), and a Backpropagation Neural Network (BPNN). First, an initial multi-dimensional indicator system comprising 15 features was established, encompassing technical dimensions (e.g., inventor team size, CPC classification count, forward citations, scientific linkage), legal dimensions (e.g., patent validity status, first claim word count, litigation frequency, examination pendency), and economic dimensions (e.g., transaction frequency, patent family size, industry coverage breadth, technology-efficacy phrases, university-industry collaboration intensity). To mitigate feature redundancy and enhance model parsimony, GA was coupled with RST for attribute reduction, which yielded a compact subset of eight core indicators: inventor count, CPC count, forward citations, patent validity, first claim word count, examination pendency, industry coverage breadth, and technology-efficacy count.
Subsequently, a GA-BP neural network was constructed wherein GA assists in optimizing the initial weights and thresholds of the BPNN to facilitate convergence. The model was trained and validated using a balanced experimental dataset consisting of patents recognized with prestigious national awards (treated as positive samples) and a corresponding set of un-awarded patents drawn from identical technical domains. Following validation, the model was applied to a sample of invention patents from six representative universities, comprising both research and non-research universities. On the validation set, the GA-BP model demonstrated consistent improvements in classification performance relative to standard BP and PCA-BP benchmark models. While the absolute gains are modest, the pattern of enhanced predictive stability across multiple evaluation metrics suggests that the integration of GA for parameter initialization offers a viable means of improving model generalization. Furthermore, robustness checks under varying noise conditions indicated that the GA-BP framework maintained a more stable performance profile, hinting at its potential utility in handling real-world data fluctuations.
Application of the model to the six university samples allowed the identification of a substantial cohort of high-value patents. A comparative analysis revealed tentative distinctions in innovation patterns. Patents from research universities tended to exhibit characteristics associated with moderate-sized collaborative teams, relatively higher upper-bound technological impact, more efficient examination processes, and broader industry relevance. In contrast, patents from non-research universities more frequently displayed attributes such as more extensive first claim drafting and richer technology-efficacy descriptions, suggesting a greater reliance on textual elaboration and application-oriented narratives to signal value.
This study extends the patent valuation discourse by incorporating indicators such as examination pendency and technology-efficacy count, providing empirical evidence for the utility of CPC classifications in capturing application-oriented value. By juxtaposing research and non-research universities, the study illuminates heterogeneous pathways to value creation, suggesting that institutions should tailor strategies to their specific ecosystems. Specifically, while research universities might prioritize interdisciplinary collaboration and procedural optimization, other institutions may find greater value in meticulous claim drafting and articulating practical benefits. Ultimately, this model offers a decision-support mechanism to assist technology transfer professionals and policymakers in triaging patent portfolios, aiming to facilitate the efficient translation of academic research into societal and economic benefits.

The coupling coordination between digitalization and green development is essentially the dialectical unity of productivity innovation, represented by digital technology, and the transformation of development models oriented towards green development. Digitalization refers to the process of utilizing digital technologies to endow individuals, organizations, or systems with new capabilities and opportunities, thereby achieving efficiency improvements, innovative development, and value creation. The concept of green development takes the harmonious coexistence of humanity and nature as its core, advocating for the minimization of natural resource consumption and ecological damage in economic activities to achieve sustainable economic, social, and ecological development. The coupling coordination between digitalization and green development is manifested in a dynamic collaborative process of "empowerment-feedback". This interaction transcends the industrial perspective of a "digital economy-green economy" dichotomy, upgrading to a cross-system collaboration of "system-technology-culture".
This study investigates the coupling coordination mechanism between regional digitalization and green development. To empirically verify this coordination, explore its coupling model under the concept of ecological civilization, and clarify the pivotal role of digital empowerment in the ecological governance of the Yangtze River Economic Belt, this study comprehensively employs methods including the coupling coordination model, spatial autocorrelation model, obstacle degree model, and geographical detector to construct an evaluation indicator system and analyze the coupling coordination degree, spatiotemporal differentiation characteristics, and driving mechanisms from 2012 to 2023.
The results show that the coupling coordination degree of digitalization and green development in the Yangtze River Economic Belt has increased year by year, with obvious regional differentiation; there is a significant positive spatial correlation and it exhibits spatiotemporal characteristics of spatial differentiation; the driving factors include ecological protection, environmental restoration, resource utilization, and digital technology application, with certain differences across different years; the driving factors are divided into three categories along the time dimension: rising first then falling, falling first then rising, and continuously declining. In the spatial dimension, the intensity of each driving factor′s role shows obvious spatial differentiation, among which industrial structure, urbanization rate, and higher education constitute the core driving factors, while technology circulation represents a common factor with insufficient driving force. To improve the digitalization and green development capability of the Yangtze River Economic Belt, it is necessary to accelerate digital transformation and development to achieve coordinated acceleration of digitalization and green development; break through the constraints of key driving factors to balance digitalization and green development; classify and optimize the effectiveness of driving factors to accurately target and address spatial disparities in driving forces; and promote regional coordinated development to narrow the spatial differentiation of coupling coordination degrees.
The innovations of this paper are as follows: First, by incorporating core elements of artificial intelligence including data, computing power, and algorithms, and expanding the connotations of digitalization and green development, this paper explores their development mechanisms and reveals the inherent logic of their coupling coordination. Second, it constructs an evaluation index system for digitalization and green development, and uses panel data from 11 provinces and cities in the Yangtze River Economic Belt from 2012 to 2023 to conduct a coupling coordination analysis. Third, it reveals the spatiotemporal characteristics of the coupling coordination degree and performs a spatial autocorrelation test. Fourth, it analyzes the factors restricting the coupling coordination degree via the obstacle degree model. Fifth, it identifies the influencing driving factors of the coupling coordination degree using the geographical detector.

With the rapid development of the digital economy, digital industries have become a crucial engine for promoting urban innovation. As an important organizational form of digital-real economy integration, digital industrial clusters play an important role in enhancing urban innovation vitality and advancing regional economic development. However, existing studies have mainly examined the impact of digital industrial agglomeration on urban innovation from a static perspective. Insufficient attention has been paid to the dynamic evolutionary characteristics of agglomeration, and the boundary role of urban knowledge allocation capacity in shaping the innovation effect of dynamic agglomeration remains underexplored.
To address this research gap, this study draws on the perspective of dynamic externalities and incorporates dynamic agglomeration of digital industries, urban knowledge allocation capacity, and urban innovation into a unified analytical framework. This study constructs an unbalanced panel dataset of 3 237 city-year observations across 222 prefecture-level Chinese cities from 2006 to 2022. Urban innovation is proxied by the natural logarithm of one plus the sum of city-level invention and utility model patent applications. To measure digital industrial dynamic agglomeration, this study first identifies digital industries using the National Bureau of Statistics classification and then employs a dartboard approach with 500 Monte Carlo iterations to compute a standardized Z-index capturing deviations from random spatial allocation. The Bai-Perron structural break test is applied to detect regime shifts in each city’s agglomeration trajectory, and segmented trend regressions are estimated to quantify directional momentum, with the resulting slope coefficient serving as the dynamic agglomeration proxy. Urban knowledge allocation capacity is operationalized via betweenness centrality in intercity patent transfer networks (knowledge flow configuration) and patent collaboration networks (knowledge synergy configuration), reflecting a city’s bridging role in cross-regional knowledge circulation. City-level economic development, industrial structure, science-education expenditure, and digital infrastructure are included as controls.
The study finds that there is a significant inverted U-shaped relationship between the dynamic agglomeration of digital industries and urban innovation. Moderate dynamic agglomeration can effectively promote urban innovation by generating resource adsorption effects, network restructuring effects, and path renewal effects. However, excessive dynamic agglomeration may inhibit urban innovation because it increases the costs of knowledge transformation, structural adjustment, and factor allocation. This conclusion remains robust after a series of robustness checks. Furthermore, the moderating effect analysis shows that urban knowledge allocation capacity significantly strengthens the inverted U-shaped relationship between the dynamic agglomeration of digital industries and urban innovation. Specifically, cities with stronger knowledge allocation capacity exhibit a steeper inverted U-shaped curve, and the turning point shifts to the left. This indicates that stronger knowledge allocation capacity amplifies both the positive effect of dynamic agglomeration in the low-to-moderate stage and the negative effect in the excessive agglomeration stage. Heterogeneity analysis further reveals that the inverted U-shaped effect is more pronounced in central cities and cities with well-developed digital infrastructure, suggesting that the innovation consequences of dynamic agglomeration vary across cities with different resource endowments and digital foundations.
This study makes several marginal contributions. First, it moves beyond the static research paradigm of digital industrial agglomeration by introducing the perspective of cluster dynamics, thereby enriching research on the relationship between digital industrial agglomeration and urban innovation from the lens of dynamic agglomeration. Second, it develops a measurement index for the dynamic agglomeration of digital industries, providing a methodological reference for capturing the evolutionary characteristics of industrial agglomeration in the digital era. Third, it reveals the moderating mechanism of urban knowledge allocation capacity and the heterogeneous effects of urban attributes, thus clarifying the environment-strategy matching logic underlying the innovation effects of digital industrial dynamic agglomeration. Finally, the findings provide theoretical support and policy implications for optimizing the spatial layout of digital industries, promoting the differentiated development of digital industrial clusters, and enhancing urban innovation and high-quality development in the context of the digital economy.

Amid a persistently sluggish global economy and escalating external uncertainties, China's economy faces multiple pressures. Accelerating the cultivation of urban new quality productive forces is great significance. Digital technologies have emerged as crucial engines for fostering new quality productive forces by reshaping innovation spaces and enabling networked collaboration. Yet digital innovation activities are inherently volatile, uncertain, complex, and ambiguous, rendering them vulnerable to external disruptions. Faced with such shocks, some cities demonstrate strong adaptive capacities, while others struggle to recover and fail to drive new quality productive forces, a divergence attributable to variations in digital innovation resilience. Consequently, unlocking the value of digital innovation resilience for urban new quality productive forces warrants urgent scholarly attention.
In this context, this paper uses the panel data of 229 prefecture-level and above cities in China from 2014 to 2023, incorporates urban new quality productive forces, digital innovation resilience, entrepreneurial activity, digital divide and government digital attention into a unified research framework, and explores the internal mechanism through which digital innovation resilience affects urban new quality productive forces. Specifically, it examines the mediating role of entrepreneurial activity and the moderating roles of digital divide and government digital attention, thereby providing a comprehensive understanding of the transmission pathway and contextual boundaries of this relationship.
The main conclusions are as follows:First, digital innovation resilience has a significant positive impact on urban new quality productive forces, which is further confirmed by robustness tests. Heterogeneity analyses suggest that the positive impact is significant in cities with a high level of digital economy development, service-oriented cities, as well as large and medium-sized cities, whereas it is insignificant in cities with a low level of digital economy, production-oriented cities and small-sized cities. Second, entrepreneurial activity plays a partial mediating role in the impact of digital innovation resilience on urban new quality productive forces. Third,digital divide plays a negative moderating role in the impact of digital innovation resilience on urban new quality productive forces. The moderating role of government digital attention in the impact of digital innovation resilience on urban new quality productive forces is insignificant.
This study proposes the following policy implications. First, unified data middle platforms should be established and R&D data resources integrated to underpin the development of urban new quality productive forces driven by digital innovation resilience. Meanwhile, accelerating the construction of digital technology pilot bases and facilitating the penetration of digital technologies into traditional industries will help boost urban new quality productive forces. Furthermore, differentiated strategies for enhancing digital innovation resilience should be tailored to local digital economy development, industrial structure, and urban scale. Second, relevant departments need to optimize the business environment and enhance entrepreneurial vitality. In addition, efforts should be made to strengthen intellectual property protection and improve institutional guarantee for innovation and entrepreneurship, thereby smoothing the pathway for digital innovation resilience to drive urban new quality productive forces. Third, bridging the digital divide requires increased investment in new digital infrastructure to lay a hardware foundation. Open access to digital tools should be promoted to encourage widespread adoption by innovative entities. Moreover, intercity cooperation of digital innovation should be strengthened to inject strong impetus into the development of urban new quality productive forces driven by digital innovation resilience.
This study offers three theoretical contributions. First, existing studies have focused on the relationship between digital innovation and new quality productive forces, and lack analyses on the impact of digital innovation resilience on urban new quality productive forces. Based on resilience theory, this study verifies the positive impact of digital innovation resilience on urban new quality productive forces and deepens the research on the impact factor of new quality productive forces. Second, based on technological innovation theory, this study incorporates entrepreneurial activity into the research framework, verifies the mediating role of entrepreneurial activity in the impact of digital innovation resilience on urban new quality productive forces, and clarifies the transmission path through which digital innovation resilience affects urban new quality productive forces. Third, based on regional innovation theory and signaling theory, this study discusses the moderating roles of digital divide and government digital attention in the impact of digital innovation resilience on urban new quality productive forces, and expands the boundary conditions under which digital innovation resilience affects urban new quality productive forces.

As a key pillar of the national economy, the manufacturing industry faces significant challenges in energy consumption and pollutant emissions during its development. While green innovation serves as a crucial means to drive manufacturers toward greener and lower-carbon development, most manufacturing enterprises in China still rely on resource-intensive production models, and some firms engage only in perfunctory innovation to secure government subsidies and tax incentives, undermining genuine sustainable innovation capabilities and environmental performance.
In recent years, the proportion and quality of climate risk disclosure by manufacturing enterprises in China have shown an increasing trend year by year. Climate risk perception helps optimize internal governance structures, enhance corporate green resilience, and improve trust among investors and consumers.Yet, existing research relies heavily on broad samples of listed firms, paying insufficient attention to the manufacturing sector. Mechanism analyses have focused narrowly on environmental management, disclosure practices, and ESG performance, yielding fragmented insights. Critical gaps remain:how executive environmental awareness, financing constraints, and green investment interact, and whether they form a serial mediation chain remains underexplored.Whether government R&D subsidies and media oversight can facilitate the innovative driving effect of corporate climate risk perception also warrants further exploration.
Drawing on resource-based view and cognitive-behavioral theory, this study examines panel data from Chinese A-share manufacturing listed companies from 2010 to 2023 to empirically analyze the impact mechanism of climate risk perception on green innovation in manufacturing enterprises. The findings are as follows:(1) Climate risk perception can significantly promote green innovation in manufacturing enterprises, with notable effects on both invention-based and utility-based green innovation, a conclusion that remains valid after a series of robustness tests including the instrumental variable method and propensity score matching method; (2) Heterogeneity tests reveal that the impact of climate risk perception on green innovation in manufacturing enterprises exhibits heterogeneity in terms of ownership nature, pollution levels, industry attributes, and regional location, with more pronounced effects in state-owned enterprises, non-heavy-pollution enterprises, advanced manufacturing, and enterprises in eastern regions; (3) Mechanism tests indicate that climate risk perception primarily enhances green innovation levels in manufacturing enterprises through three pathways:strengthening executives' environmental awareness, alleviating financing constraints, and promoting green investment, with financing constraints having the strongest mediating effect, followed by green investment, while the role of executives' environmental awareness is relatively weaker. Moreover, a chain mediation effect exists among executives' environmental awareness, financing constraints, and green investment, with both government R&D subsidies and media supervision positively moderating the positive impact of climate risk perception on green innovation in manufacturing enterprises.
The innovations of this study are as follows:In terms of research subjects, unlike prior studies using broad firm samples, this study concentrates on manufacturing enterprises with the manufacturing sector further subdivided into advanced manufacturing and traditional manufacturing.This disaggregation reveals previously masked heterogeneity in how climate risk perception shapes green innovation, thereby expanding the research scope of the relationship between climate risk perception and green innovation. Regarding mechanism identification, a chain mediation model of "environmental cognition-resource acquisition-green investment" was constructed based on cognitive behavioral theory to reveal the influence mechanism of climate risk perception on the green innovation of manufacturing enterprises. At the theoretical level, it clarifies the cascading mediation mechanism capturing how climate risk perception elevates environmental awareness, facilitates resource access, and catalyzes green investment. At the empirical level, the study employs the Bootstrap chain mediation effect test method to validate the effectiveness of the multi-stage chain transmission path. In terms of research context, from the perspectives of formal and informal institutions, this study investigates institutional environmental factors, such as government R&D subsidies and media supervision, to explore their moderating effects, providing empirical evidence and clear pathways for maximizing the innovative driving effects of climate risk perception.

In the context of intensifying global technological competition and industrial transformation, China's manufacturing sector faces a long-standing scale-strength mismatch. While existing research has examined how digital technologies affect innovation performance and how network embeddedness facilitates knowledge acquisition, scant attention has been devoted to whether and how digital innovation networks enable manufacturing firms to transcend existing technological trajectories and expand into new non-digital domains. This study addresses this gap by investigating how embeddedness in digital innovation networks affects the expansion of firms' real-economy technological boundaries, along with its underlying mechanisms and boundary conditions.
Grounded in the space of flows theory and general-purpose technology theory, the study argues that digital innovation networks, as a typical form of the space of flows, enable cross-domain knowledge transfer, resource reallocation, and digital-real technology fusion. These networks overcome geographical constraints and facilitate low-cost, real-time interactions, thereby helping manufacturing firms alleviate the resource and risk constraints associated with exploring unfamiliar technological territories. Specifically, the study proposes three mediating pathways: knowledge base broadening, resource allocation optimization, and digital-real technology integration. Empirically, the study constructs firm-level digital innovation networks using digital technology patent co-application data of A-share listed manufacturing firms in China from 2010 to 2024. The final sample comprises 20 386 firm-year observations. The key independent variable is degree centrality in the digital innovation network, capturing the extent of a firm's direct collaborative ties. The dependent variable, real-economy technological innovation boundary, is measured by the number of new IPC subgroups in non-digital patent applications that the firm has never entered before. Furthermore the study employs a panel fixed-effects model with firm, city, and year fixed effects, and address endogeneity using instrumental variables and a PSM-DID approach.
The empirical results reveal several key findings. First, digital innovation network embeddedness significantly expands manufacturing firms' real-economy technological innovation boundaries, and this effect remains robust after a series of tests including placebo tests, alternative variable measurements, exclusion of competitive hypotheses, and different clustering levels. Second, the mediation analyses confirm the three proposed mechanisms: network embeddedness enhances knowledge search breadth and knowledge diversity, reduces over-investment and excess labor, and promotes digital-real technology fusion. All indirect effects are significant according to Sobel and bootstrap tests. Third, the effect is heterogeneous: it is stronger for firms with higher technological generality, for those with lower artificial intelligence application levels, and for firms located in cities where the government pays greater attention to the digital economy. Fourth, further analysis shows that network structural features and network resource endowments positively moderate the main effect. Fifth, the boundary-expansion effect is not merely about "breadth" but also "quality and efficiency": network embeddedness improves the quality, efficiency, and breakthrough capability in core technologies of newly entered real-economy fields.
This study advances the literature in four key dimensions.In theoretical terms, this study pioneers an integrative framework bridging space of flows theory with GPT theory through the lens of "digital-real cross-domain empowerment", illuminating how digital collaboration networks catalyze technological breakthroughs in the physical economy. From a methodological perspective, it introduces precision measurement of real-economy technological boundaries by identifying non-digital patent domains and tracking firms' inaugural entries into new IPC subgroups, thereby isolating innovation dynamics from digital technology confounds. On the empirical front, the work furnishes comprehensive evidence on mediating mechanisms, boundary conditions, and real-economy outcomes, yielding actionable implications for corporate strategists and policymakers alike. The findings advocate for manufacturing firms' proactive positioning within digital innovation networks, particularly via structural hole exploitation and the cultivation of global, academic ties, while calling for governmental efforts to prioritize digital policies, strengthen data governance and IP protection, and implement tailored firm-support schemes.

Megaprojects are inherently characterized by high levels of volatility,uncertainty,complexity,and ambiguity (VUCA),which demand that team members engage in improvisational innovation to respond effectively to emergent conditions during project execution. This form of innovation involves not only generating novel ideas under severe time pressure but also translating those ideas into actionable solutions. However,idea generation and idea implementation follow distinct and often conflicting logics,i.e. exploration versus exploitation. This inherent tension raises a central question:How can team members in megaprojects be enabled to effectively engage in both activities? Although prior research has largely followed a "separation" logic by using temporal sequencing or structural isolation to avoid tensions between the two activities,this perspective has limited applicability to the complex context of megaprojects,where the two activities are interwoven and iterative. Drawing on ambidexterity theory and cognitive-affective theory,this study investigates how the dynamic interplay of ambidextrous leadership shapes project team members' engagement in these dual innovation activities.
This study operationalizes grounded theory through a multi-stage analytical protocol. The empirical foundation comprises semi-structured interviews with 39 members across 12 megaproject teams,selected via purposive sampling. The analytical trajectory involved:(1) open coding,inductively producing 50 initial subcategories; (2) axial coding,refining these into 25 core categories; and (3) selective coding,linking categories to theoretical constructs. The resultant process model explicates how ambidextrous leadership synergistically influences project team members' improvisational innovation.
To explain how teams navigate the paradox without temporal or structural buffers,the study integrates ambidexterity theory with cognitive-affective theory,and proposes an analytical framework that links ambidextrous leadership,individual cognition and emotion,network reconfiguration,and improvisational innovation. This framework deconstructs the intrinsic synergistic mechanisms through which ambidextrous leadership facilitates team members' improvisational innovation in megaprojects. First,during idea generation,ambidextrous leadership dynamically shapes team members' cognitive patterns and emotional states through the strategic integration of "opening" and "closing" behaviors,thereby guiding them to construct differentiated social networks and ultimately achieving synergy between novelty and usefulness. Second,during idea implementation,ambidextrous leadership guides team members to reconfigure their social networks by shaping differentiated goal orientations and work states,thereby addressing the paradoxical tension between innovation quality and innovation speed. Third,throughout the entire improvisational innovation process,ambidextrous leadership resolves the tension between idea generation and implementation through a sequential pathway that integrates behavioral strategy,role cognitive adjustment,network reconfiguration,and innovation synergy.
This study offers several theoretical contributions. (1) Moving beyond the "separation" paradigm,it reveals how ambidextrous leadership resolves the paradox of creative idea generation and implementation in megaprojects,providing a novel theoretical explanation for achieving synergy when the two activities are difficult to separate. (2) From a holistic process perspective,this study elucidates the differentiated synergistic pathways of ambidextrous leadership at distinct stages of improvisational innovation and demonstrates how a single leadership behavior may disrupt this synergy,further uncovering how ambidextrous leadership addresses multidimensional paradoxes across the innovation process. (3) By integrating cognitive-affective theory and social network perspective,this study proposes a "cognition-emotion driven network reconfiguration" process model underlying the influence of ambidextrous leadership on improvisational innovation.
This study also provides key managerial insights. (1) Megaproject leaders should recognize the cyclical interplay between idea generation and implementation during improvisational innovation and flexibly alternate between opening and closing behavioral strategies according to different situations,thereby fostering the synergistic development of both processes. (2) Leaders should attend to the inherent contradictions in team members' innovation processes and adopt flexible behavioral strategies to reconcile the paradoxes of novelty versus usefulness in idea generation and quality versus speed in idea implementation. (3) Leaders should be cautious about the potential harm that a single behavioral strategy may impose on team members' cognition and emotions.

Technology mergers and acquisitions (M&A) are widely used by firms to acquire external knowledge and enhance innovation. However, the failure rate of technology M&A remains high, at approximately 70% in both the U.S. and China. Existing academic research on technology M&A innovation mostly focuses on the integration of technical talent, organizational structure, and resource allocation within acquired firms, or explores the optimization of M&A decision-making from the perspective of resource complementarity and technical matching between the two parties, yet largely overlooks the vital impact of M&A activities on the key inventors of the acquiring firms themselves. These key inventors serve as the core carriers of the acquiring firm′s original technical accumulation and the main force for knowledge absorption, integration, and secondary innovation following an M&A; changes in their creativity during the post-M&A stage directly determine the efficiency of benefit transformation and the sustainability of corporate innovation.
This study takes the evolution of creativity among key inventors in acquiring firms after technology M&A as its core research object, constructing a dynamic multi-level creativity evaluation model to realize scientific and accurate assessment of their creativity evolution, thereby addressing the limitations of existing single-dimensional and static evaluation methods. By combing the theories of social network, innovation economics and talent creativity, this study builds a multi-dimensional and interrelated evaluation framework from three core levels: collaboration network dynamics, individual inventor characteristics, and M&A context features, which systematically covers the internal and external factors affecting post-M&A creativity.
In terms of research methods, this study adopts a two-stage PCA-BP neural network model to solve the problems of high dimensionality and multicollinearity of initial indicators, as well as the non-linear correlation between influencing factors and creativity changes: firstly, Principal Component Analysis (PCA) is used to reduce the dimension of 12 initial evaluation indicators, filter out redundant information and extract core components; secondly, the extracted principal components are input into the Back-Propagation (BP) neural network for training and fitting, so as to build an efficient and accurate non-linear evaluation model.
For empirical testing, this study selects 219 key inventors active between 1997 and 2024 as the research sample, collecting 21 616 valid patent records with key inventors identified by patent quantity, citation frequency and claim counts to ensure sample representativeness. The sample is divided into 170 training sets and 49 testing sets to verify the model′s effectiveness. The test results show that the PCA-BP model has a mean square error (MSE) of only 0.000 58 and a prediction accuracy rate of 97%, with its predicted creativity trajectory highly consistent with the actual change trend, demonstrating strong generalization ability and robustness. Comparative analysis further proves that the dynamic multi-level index system and PCA-BP model are significantly superior to traditional static indicators, single BP neural network and random forest model in evaluation accuracy, convergence speed and anti-interference ability.
This study realizes innovation in research perspective and method in the field of technology M&A innovation: in theory, it shifts the research focus from acquired firms to the core inventors of acquiring firms, putting forward a systematic “network dynamics-individual characteristics-M&A context” analysis framework and enriching the theoretical research on post-M&A talent creativity and innovation management; methodologically, it integrates PCA and BP neural network, providing a feasible technical path for non-linear evaluation in small-sample and high-dimensional scenarios. In practice, the constructed model can serve as a targeted diagnostic tool for enterprises to monitor the creativity changes of key inventors after M&A, helping enterprises timely identify creativity decline risks, carry out targeted intervention, optimize core human capital management, and further improve the innovation performance and success rate of technology M&A, thus providing important theoretical support and practical reference for M&A decision-making and post-merger integration management.

The commercialization of state-owned scientific and technological achievements has become a key issue in China′s efforts to advance innovation-driven development, improve the efficiency of public R&D investment, and accelerate the formation of new quality productive forces. In practice, however, research institutions and universities commonly face persistent obstacles characterized by reluctance to commercialize, fear of commercialization, and difficulty in bringing achievements into application. The root of the problem lies not merely in procedural inefficiency or inadequate market coordination, but in a structural imbalance between rights and responsibilities. Although the Law of the People′s Republic of China on Scientific and Technological Progress grants research institutions autonomous disposal rights over scientific and technological achievements, the current state-owned asset management system still constrains commercialization through rigid requirements such as appraisal, approval, filing, and accountability. As a result, a reverse binding mechanism has emerged: research institutions are formally authorized to make autonomous decisions, yet once commercialization outcomes become uncertain, they still bear substantial administrative and legal risks.
The study identifies the root cause of the current predicament as the misidentification of state-owned scientific and technological achievements as ordinary state-owned assets, which subjects them to a static governance logic focused on physical control and book-value preservation. This approach fails to recognize their distinctive nature as knowledge-based, time-sensitive assets whose value depends on circulation speed and market adaptability, causing them to lose market windows and depreciate implicitly under lengthy approval procedures. This misalignment in legal characterization further generates three consequences. First, it gives rise to normative conflict between legal empowerment and administrative regulation. Higher-level laws encourage autonomous disposition, whereas lower-level rules retain strict approval and supervisory mechanisms, thereby weakening the practical effect of statutory authorization. Second, it creates operational blockages throughout the entire commercialization chain, including the hollowing-out of disposition rights, the conflict between dynamic value realization and static value-preservation requirements, and the excessive concentration of liability on individual decision-makers. Third, it leads to systemic failure in the broader commercialization ecosystem, manifested in distorted incentive mechanisms, reinforced bureaucratic governance, and the treatment of scientific and technological achievements as rigidly controlled asset objects rather than as carriers of innovation value.
By adopting a combination of normative legal analysis, institutional analysis, and policy observation,the study examines the value conflict between innovation-promoting legislation and risk-control-oriented state-owned asset regulation, and analyzes representative local reform pilots in different regions of China. These pilot practices are categorized into three types of policy instruments: rule-adjustment instruments, property-right incentive instruments, and market-enabling instruments. On this basis, the study evaluates both their reform effects and their institutional limits, and further proposes a systematic rule-of-law path for rebalancing rights and responsibilities in the commercialization of state-owned scientific and technological achievements.
To address this structural dilemma, this study proposes a four-dimensional legal framework centered on the concept of "time-limited franchised state-owned assets". This concept is a further refinement of the notion of "exceptional assets". Its core implication is that, although state-owned scientific and technological achievements should remain under state ownership because they are generated through public investment, part of the general rules governing state-owned assets should be expressly exempted by law in light of their intangibility, value volatility, and innovation-promoting policy function. At the legislative level, their special legal status should be confirmed through exclusion clauses in general state-owned asset legislation and through specialized legal norms. At the enforcement level, a regulatory model of "negative list + statutory exemption + ex post filing" should be established to shift from prior approval to bottom-line supervision. At the judicial level, a defense standard centered on due diligence should be constructed so as to move liability determination from result-based attribution to conduct-based review. At the supporting institutional level, reforms in property-right registration, revenue distribution, and exit mechanisms should be advanced in a coordinated manner so as to facilitate circulation, strengthen incentives, and tolerate reasonable failure.

The proposal of China's 'Dual-Carbon' goals has positioned market-oriented environmental rights trading as a pivotal policy instrument for advancing energy conservation and emission reduction. Energy use rights trading and carbon emission rights trading, as two core mechanisms, respectively facilitate optimal resource and environmental allocation through total energy consumption control at the source and aggregate carbon emission management at the output stage, thereby promoting green and low-carbon economic transition. However, given the severity of carbon reduction challenges and the entrenched carbon lock-in arising from the co-evolution of high-carbon technological systems and institutional arrangements, a critical question remains: Can environmental rights trading policies effectively overcome carbon lock-in constraints and fundamentally catalyze green transformation?
　　Existing research has documented significant positive effects of energy use rights and carbon emission rights trading on energy consumption, technological innovation, and industrial structure, yet predominantly examines these policies in isolation. While the role of environmental regulation in carbon unlocking has garnered increasing attention, most studies focus on single driving factors. Consequently, whether the combination of energy use rights and carbon emission rights trading generates synergistic effects exceeding individual policy impacts, and through what mechanisms such synergy operates, remains insufficiently understood.
Against the policy backdrop of dual-pilot regions, this study constructs a "source-process-output" analytical framework to examine how these two policies jointly influence carbon lock-in. Leveraging the quasi-natural experimental setting of overlapping pilot regions, the study systematically analyzes the synergistic effects and underlying mechanisms across input, process, and output dimensions. This study selects panel data of 300 Chinese cities from 2010 to 2023 as samples, and uses the difference-in-differences method to conduct an in-depth analysis of the impact of the two types of policy instruments on carbon lock-in. The dataset is primarily derived from China City Statistical Yearbooks and official government reports. To mitigate the potential influence of outliers, all continuous variables are winsorized at the 1% and 99% levels. This rigorous data preprocessing enhances the robustness of our empirical findings regarding the decoupling effects of energy and carbon trading schemes.
The research results show that,first, both the energy use rights and carbon emission rights trading schemes have significantly reduced the carbon lock-in level of pilot cities; Second, the analysis of the impact mechanism finds that energy use rights and carbon emission rights trading jointly break carbon lock-in by reducing fossil energy dependence, technological path dependence and high-carbon industry embeddedness; Third, further analysis finds that while breaking carbon lock-in, the energy use rights and carbon emission rights trading schemes do not compromise economic performance,and also help promote high-quality economic development.
Compared with the existing literature, the contributions of this study include the following three points: First, it expands the research boundary of existing environmental rights trading policies from the perspective of policy synergy. Second, it constructs a systematic unlocking framework of carbon lock-in of 'source control - end-of-pipe governance', and considers the synergistic effect of energy use rights and carbon emission rights trading in breaking carbon lock-in from a holistic perspective, providing a systematic driving perspective and empirical evidence for breaking carbon lock-in. Third, it comprehensively reveals how energy use rights and carbon emission rights trading break carbon lock-in from the whole-process perspective of 'input-process-output', namely fossil energy dependence, technological path dependence, and high-carbon industry embeddedness. This study provides decision-making references for refining energy use rights and carbon emission rights policies and formulating scientific and effective carbon unlocking measures, as well as accumulating new theoretical and empirical evidence for systematically promoting the green and low-carbon transformation of industries.