China’s manufacturing digital technologies propagate along 14 patent-backed main paths led by core capabilities like image recognition; high-value technologies and recombination potential — not similarity — chiefly shape those paths, though the effects of diversity and proximity differ across firms, universities and regions.

Mapping China’s digital transformation: a multilayer network analysis of technology diffusion in manufacturing

Jialu Ren, Yong Zhou, Yangyang Yang, Shengkai Wang · April 17, 2026 · Humanities and Social Sciences Communications

openalex correlational medium evidence 7/10 relevance DOI Source PDF

Using 2000–2024 Chinese manufacturing patent data and multilayer network analysis, the study finds that high technological value and recombination opportunities consistently promote the formation of main diffusion paths, while technological diversity and geographic/proximity effects vary across patent, organizational, and regional networks.

The diffusion of digital technologies has driven the transformation and upgrading of manufacturing industries. However, the main pathways and formation mechanisms of digital technology diffusion remain unclear. Using patent data of China’s manufacturing digital technologies from 2000–2024, this study constructs a multilayer network comprising patent citation networks, inter-organizational technology diffusion networks, and geographical technology diffusion networks. Through main path analysis and exponential random graph models (ERGM), we analyze the formation mechanisms of China’s digital technology diffusion pathways. The findings identify 14 main paths in patent citation networks, spanning from core technologies like image recognition to enabling applications. Inter-organizational paths center on key universities, while geographical networks exhibit a “core-periphery” structure. ERGM results reveal that technological collaboration value and combination opportunities consistently promote main path formation. However, technological diversity and proximity exhibit differentiated effects across network layers. In patent networks, neither diversity nor proximity shows significant impact, suggesting that patent citation-based diffusion prioritizes technological value and recombination potential over field diversity or similarity. In inter-organizational networks, only diversity promotes path formation, highlighting how knowledge recombination drives micro-level trajectories. Conversely, in geographical networks, both diversity and proximity inhibit path formation, indicating that macro-regional evolution requires specialized focus and complementary knowledge. Heterogeneity analysis further reveals distinct evolutionary logics: universities bridge distant domains through knowledge diversity, whereas market-driven enterprises heavily rely on high-value core technologies. Spatially, Western regions rely heavily on core technological hubs, Eastern regions are driven by knowledge recombination opportunities, and Central regions have not yet formed a dominant diffusion mechanism. This study uncovers digital diffusion dynamics and provides theoretical foundations for policymaking.

Summary

Main Finding

Using China manufacturing patent data (2000–2024) and a multilayer network approach (patent citation, inter-organizational diffusion, geographic diffusion), the study identifies 14 principal diffusion paths of digital technologies (from core building blocks like image recognition to enabling applications), shows that inter-organizational diffusion centers on key universities while geographic diffusion forms a core–periphery pattern, and demonstrates via ERGMs that technological collaboration value and recombination (structural‑hole) opportunities consistently promote formation of main paths. Technological diversity and proximity have layer‑dependent effects: they are not significant in patent citation networks, diversity promotes inter‑organizational main paths, and both diversity and proximity inhibit geographic main‑path formation. Heterogeneity analysis finds universities act as brokers bridging distant domains through knowledge diversity, whereas firms rely more on high‑value core technologies; spatially, diffusion evolves from hub‑driven (West) to recombination‑driven embedded networks (East). The paper draws policy-relevant implications for guiding digital/AI diffusion in manufacturing.

Key Points

Multilayer framing: diffusion is analyzed across three interconnected layers—patent citations (technology flows), inter-organizational ties (who shares/adapts tech), and geographic links (regional spread).
Main-path discovery: 14 main citation paths extracted with SPNP reveal trajectories from core digital technologies (e.g., image recognition) to downstream manufacturing applications.
Organizational role: universities occupy central brokerage positions in inter-organizational diffusion paths; firms cluster around high‑value core technologies.
Spatial structure: geographic diffusion shows a core–periphery pattern; eastern regions display denser recombination-driven networks, western regions show more unidirectional hub dependence.
Formation mechanisms (ERGM results):
- Positive drivers across layers: technological collaboration value (degree centrality) and recombination opportunities (structural-hole measures).
- Layer‑dependent effects:
  - Patent layer: structural/positional advantage dominates; diversity and proximity not significant.
  - Organizational layer: technological diversity promotes main-path formation (knowledge recombination matters).
  - Geographic layer: both diversity and proximity negatively associated with forming cross‑regional main paths (local complexity can hinder cross‑regional diffusion).
Heterogeneity: universities vs firms and east vs west regions follow different evolutionary logics in how digital technologies diffuse.

Data & Methods

Data: Patent data covering China’s manufacturing digital technologies, 2000–2024 (paper is an unedited manuscript; details on exact patent sources/cleaning referenced in full text).
Multilayer network construction:
- Layer 1: Patent citation network (nodes = patents/technologies; edges = citations).
- Layer 2: Inter-organizational diffusion network (nodes = organizations; edges = technology transfer/relationships inferred from patent assignees, co‑patenting, etc.).
- Layer 3: Geographic diffusion network (nodes = regions/provinces; edges = patent flows or interregional linkages).
Main-path extraction: Search Path Node Pair (SPNP) algorithm applied to patent citation network to identify dominant knowledge trajectories (14 main paths).
Statistical modeling: Exponential Random Graph Models (ERGMs) used to test how node/edge-level attributes and network structural measures shape formation of main-path edges in each layer.
Key explanatory variables: technological knowledge diversity (breadth of IPC/CPC classes), technological cooperation value (degree centrality), recombination opportunities (structural‑hole/bridging measures), technological proximity (similarity/overlap measures).
Hypotheses tested: H1–H6 linking diversity, cooperation, recombination, and proximity to main-path formation across layers (some supported, some layer-dependent).

Implications for AI Economics

Diffusion dynamics of AI and other digital technologies are shaped more by collaboration value and recombination potential than by sheer diversity or proximity—especially at the patent/technology level. For AI economics, this suggests policies that raise the visibility and connectivity of high‑value AI building blocks (e.g., core models, toolkits, benchmarked components) will accelerate downstream diffusion.
Role of knowledge brokers/universities: Universities act as cross‑domain brokers that enable AI to bridge distant application domains. Supporting university research, open interfaces, and cross‑disciplinary labs amplifies AI’s transformative reach across manufacturing.
Firm strategy: Market‑driven firms tend to diffuse via consolidation around high‑value core AI technologies. Firms should invest in core component capabilities and partnerships that increase their centrality in technological networks to gain diffusion advantages.
Regional policy: Regional specialization (complementarity) may be more effective than attempting broad local diversification. Because geographic diversity and proximity can inhibit cross‑regional main-path formation, regional policies should (a) build complementary specializations and clear interfaces for interregional recombination, and (b) strengthen hub connections (especially for lagging western regions) to avoid one‑way dependency.
Fostering recombination: Structural‑hole positions and recombination capacity strongly predict main-path formation—thus policies that lower barriers to cross‑domain recombination (data sharing protocols, standard APIs, modular commons, collaborative testbeds) will support faster, broader AI adoption in manufacturing.
Tailored interventions: Given heterogeneous logics (universities vs firms; east vs west), a one‑size policy is suboptimal. Interventions should be tailored: support basic research and cross‑disciplinary exchange in university hubs; incentivize firms to adopt and integrate high‑value AI components; and design regionally specific ecosystem strategies emphasizing either hub linkage or recombination embedding.
Modeling and evaluation: For economic models of AI diffusion, multilayer network representations (technology, organizational, spatial) capture dynamics missed by single‑layer approaches—improving predictions of adoption, spillovers, and industrial upgrading.

Caveats: the manuscript is an early (unedited) version and relies on patent citations as proxies for knowledge flows (common but imperfect). Results are conditional on patent coverage, network construction choices, and ERGM specifications; further validation with complementary data (e.g., firm surveys, product adoption, employment/production metrics) would strengthen causal claims.

Assessment

Paper Typecorrelational Evidence Strengthmedium — Uses comprehensive patent data (2000–2024) and appropriate network methods (main-path analysis, ERGM) to produce consistent associations across three network layers and heterogeneity checks, which lends credibility to the patterns; however, the analysis is observational with potential confounding, endogeneity (e.g., reverse causation between centrality and 'value'), and measurement limits from patent-based proxies, so causal claims remain tentative. Methods Rigorhigh — Applies advanced, field-appropriate methods (multilayer network construction, main-path extraction, ERGM for network formation, and heterogeneity analysis across organization types and regions); these methods are well suited to the research question and provide rigorous tests of associations, though they rely on strong modeling assumptions and observable covariates only. SamplePatent records of China's manufacturing digital technologies, 2000–2024, assembled into three layers: (1) patent citation network (identifies 14 main patent citation paths spanning core technologies such as image recognition and enabling applications), (2) inter-organizational diffusion network (links among universities, firms, other organizations), and (3) geographic diffusion network (regional links across Eastern, Central, Western China); analysis includes actor-type and regional heterogeneity. Themesinnovation adoption IdentificationObservational network analysis: constructs multilayer networks (patent citation, inter-organizational diffusion, geographic diffusion) and uses main-path analysis to identify backbone trajectories, then fits exponential random graph models (ERGM) to estimate associations between node/edge attributes (technological value, diversity, proximity, recombination opportunities) and the presence of main-path links; no experimental or quasi-experimental causal identification strategy. GeneralizabilityLimited to China — institutional, policy, and industrial structures may differ elsewhere, Based on patents — ignores unpatented diffusion, trade secrets, open-source, and informal knowledge flows, Focused on manufacturing digital technologies — findings may not extend to services or non-manufacturing sectors, Patent citations are an imperfect proxy for knowledge flow and may reflect strategic citing or examiner-added citations, Observational design and ERGM assumptions limit causal generalization to different time periods or policy environments

Claims (15)

Claim	Direction	Confidence	Outcome	Details
Using patent data of China’s manufacturing digital technologies from 2000–2024, this study constructs a multilayer network comprising patent citation networks, inter-organizational technology diffusion networks, and geographical technology diffusion networks. Other	positive	high	construction of multilayer diffusion network	0.5
The patent citation network analysis identifies 14 main paths spanning from core technologies like image recognition to enabling applications. Adoption Rate	positive	high	main paths in patent citation network (technology diffusion pathways)	0.5
Inter-organizational diffusion paths center on key universities. Adoption Rate	positive	high	centrality of universities in inter-organizational diffusion paths	0.3
Geographical technology diffusion networks exhibit a 'core–periphery' structure. Market Structure	positive	high	network structure (core–periphery) of geographical diffusion	0.3
ERGM results show that technological collaboration value consistently promotes the formation of main diffusion paths across network layers. Adoption Rate	positive	high	probability/formation of main diffusion paths	0.3
ERGM results show that combination opportunities (knowledge recombination potential) consistently promote the formation of main diffusion paths across network layers. Adoption Rate	positive	high	probability/formation of main diffusion paths	0.3
In the patent citation network, neither technological diversity nor technological proximity shows a significant impact on main path formation. Adoption Rate	null_result	high	effect of diversity and proximity on main path formation (patent layer)	0.3
In the inter-organizational network, only technological diversity (not proximity) promotes main path formation, indicating knowledge recombination drives micro-level trajectories. Adoption Rate	positive	high	effect of diversity on main path formation (inter-organizational layer)	0.3
In the geographical network, both technological diversity and technological proximity inhibit main path formation, implying macro-regional evolution requires specialized focus and complementary knowledge. Adoption Rate	negative	high	effect of diversity and proximity on main path formation (geographical layer)	0.3
Heterogeneity analysis: universities bridge distant domains through knowledge diversity. Adoption Rate	positive	medium	universities' role in bridging domains via knowledge diversity	0.18
Heterogeneity analysis: market-driven enterprises heavily rely on high-value core technologies. Adoption Rate	positive	medium	enterprises' reliance on core high-value technologies	0.18
Spatial heterogeneity: Western regions rely heavily on core technological hubs. Adoption Rate	positive	medium	regional dependence on core technological hubs (Western regions)	0.18
Spatial heterogeneity: Eastern regions are driven by knowledge recombination opportunities. Adoption Rate	positive	medium	drivers of diffusion in Eastern regions (knowledge recombination)	0.18
Spatial heterogeneity: Central regions have not yet formed a dominant diffusion mechanism. Adoption Rate	null_result	medium	presence/absence of dominant diffusion mechanism (Central regions)	0.09
This study uncovers digital diffusion dynamics and provides theoretical foundations for policymaking. Governance And Regulation	positive	high	theoretical and policy relevance of findings	0.05