From Technology Follower to Global Leader: The Evolution of China’s New Energy Vehicle Innovation Ecosystem Through Patent Cooperation Networks

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript's originality lies in combining well-known network metrics with LLM-assisted domain classification. The novelty lies in the workflow integration rather than the technique itself.
Scientific rigor is pending clarification regarding the validity of the classification, baselines, and robustness tests.
The bibliographic citations are relevant, but citations fundamental to the power-law tests and time-series network analysis should be added.

Comments and suggestions for the authors
An industry-specific patent classification methodology (ISPCM) was developed using a VNE knowledge graph and a linear learning model (Qwen) to construct the CNEVIP dataset. Subsequently, the structure, centrality, and cohesion of the dataset were measured to infer industry dynamics. Key findings include: (i) approximately 100,000 patents granted with an upward trend and a predominance of utility models; (ii) collaborations account for approximately 3%; (iii) an oligopolistic structure with state-owned enterprises as centers; (iv) the components segment as a center, with the complete vehicle segment relatively isolated; and (v) a possible decoupling between domestic and international firms.
The topic fits within the scope of WEVJ and is timely. However, the article requires significant revisions to ensure reproducibility, strengthen causal interpretation, and align the analysis with best practices in network science and patent analysis.

Important comments for acceptance

1. Reproducibility and availability of data/code (deadlines and completeness).

The letter L indicates that “All data and code… are shared on… Gitee… after publication of the article.” To ensure reproducibility, please provide a private link during the review process or an archived DOI (e.g., Zenodo) with: (a) the exact identifiers of the patents used; (b) the complete definitions of the ISPCM tags; (c) the instructions/rules and any LLM configurations; (d) scripts for network and figure reconstruction; and (e) a README file with environment details.

2. The ISPCM description and validation require much more detail.

The manuscript explains that the NEV knowledge graph (7 layers: top-down/intermediate/top-down) was integrated with Qwen for multi-label hierarchical classification, but it lacks: (i) specific label inventories with counts per level, (ii) example templates/rules for queries, (iii) model version, temperature/top-p settings, and inference batches, (iv) clarification on training versus exampleless learning, (v) benchmark comparisons (IPC/keyword) with accuracy/recall/F1 on a labeled validation set, and (vi) human validation protocol (sample size, labeling guidelines, inter-rater agreement). Please add this information so the community can assess the added value of ISPCM beyond the intuitive reduction in retrieved counts.

3. Clarify the scope of the dataset and selection bias.
The analysis is based on CNIPA patents. Explain how excluding non-CNIPA applications (e.g., EPO/USPTO/WIPO) affects the cross-border collaboration inference and the "decoupling" narrative; consider creating a robustness subset that includes links to PCT families or joint assignments abroad to avoid home jurisdiction bias.

4. Defining "collaboration" and building networks.
Collaboration is defined as multiple applicants per patent, analyzed as an undirected edge. This omits (i) inventor-level collaboration, (ii) applicant-inventor interrelationships, (iii) changes in licensing/assignments, and (iv) edge weighting (number of co-applications). Please (a) present a sensitivity analysis using inventor co-authorship networks and weighted applicant networks; (b) justify the undirected assumption; and (c) explain how the prevalence of utility models and corporate structures in China may reduce formal co-application rates (~3%).

5. “Small-world” and “scale-free” claims require statistical evidence.

You report high clustering (C≈0.711) and short path length (L≈3.738), and mention a scale-free structure. Please provide (i) the small-world condition in relation to paired random graphs, and (ii) formal power-law tests (e.g., Clauset-Shalizi-Newman goodness of fit with alternative heavy-tailed distributions) before concluding that the distribution is “scale-free.” Include graphs and p-values; otherwise, temper the claim that the distribution is “heavy-tailed.”

6. Time modeling and policy inference.

The three-phase evolution is plausible, but please add networks with time windows (mobile or political regime), report on assortativity, community persistence, and node turnover. When attributing changes to specific policies (e.g., “Ten cities, one thousand vehicles”), consider using graphs or event study-style references to support the causal language.

7. The segment-level findings require operational definitions.

The authors indicate that components act as a central node and that complete vehicles are relatively isolated. Please add segment-level metrics (density, modularity contribution, intersegment mix matrix) and define “isolation” quantitatively (e.g., lower external-to-internal slope ratio).

8. Figure and table clarity & consistency.
Please ensure: (i) all axis labels and legends are readable; (ii) consistent node/edge color semantics across figures; (iii) numeric tables include units and time spans; and (iv) captions fully describe data and construction (e.g., “multi-applicant, undirected, weighted by co-applications?”). Table 2 is useful—consider adding per-period versions.

 

Comments on the Quality of English Language

The English could be improved to express the research more clearly.
Several typographical and formatting problems were noted, as well as some unclear phrases, for example, "and if it forms an oligopolistic structure," and inconsistent punctuation in the section lists.

 

Author Response

Thank you for your thorough and constructive feedback on our manuscript. We sincerely appreciate the time and expertise you have invested in reviewing our work. Below, we provide a point-by-point response to each of the comments, detailing the revisions and clarifications we have made to address your valuable suggestions.

 

Comments 1: [ Reproducibility and availability of data/code (deadlines and completeness). The letter L indicates that “All data and code… are shared on… Gitee… after publication of the article.” To ensure reproducibility, please provide a private link during the review process or an archived DOI (e.g., Zenodo) with: (a) the exact identifiers of the patents used; (b) the complete definitions of the ISPCM tags; (c) the instructions/rules and any LLM configurations; (d) scripts for network and figure reconstruction; and (e) a README file with environment details.]

Response 1: [The data/code that you required is in https://gitee.com/LyuXiaozhong/NEV openly for a month.]

 

Comments 2: [The ISPCM description and validation require much more detail. The manuscript explains that the NEV knowledge graph (7 layers: top-down/intermediate/top-down) was integrated with Qwen for multi-label hierarchical classification, but it lacks: (i) specific label inventories with counts per level, (ii) example templates/rules for queries, (iii) model version, temperature/top-p settings, and inference batches, (iv) clarification on training versus exampleless learning, (v) benchmark comparisons (IPC/keyword) with accuracy/recall/F1 on a labeled validation set, and (vi) human validation protocol (sample size, labeling guidelines, inter-rater agreement). Please add this information so the community can assess the added value of ISPCM beyond the intuitive reduction in retrieved counts.]

Response 2: [ We give a change in page 5 as followed:

”The NEV knowledge graph comprises seven hierarchical layers, as shown in Table 1. Owing to its large scale, the full graph is not included in this paper but is available on Gitee at https://gitee.com/LyuXiaozhong/NEV. The ISPCM model (using Qwen-14b-chat with default parameters after three-step training) was parallelized on a server equipped with eight NVIDIA A40 GPUs to efficiently extract and classify NEV-related patents from the CNIPA dataset, which resulted in the construction of the CNEVIP dataset. Comprehensive details regarding the ISPCM are provided in our granted patent (granted number: CN118798188B) [32]. All patents can be categorized across the various layers of China’s NEV industrial chain. Since this method supports multilabel classification, a single patent can be assigned multiple labels. Compared with conventional IPC- or keyword-based filtering approaches, the ISPCM significantly reduces noise and improves accuracy, thereby providing a more robust foundation for subsequent network analysis.

Table 1. The label counts per layer in the NEV knowledge graph.

layer	label counts
1	1
2	3
3	21
4	96
5	134
6	127
7	51
total	433

All patents can be categorized across the various layers of China’s NEV industrial chain. Since this method supports multilabel classification, a single patent can be assigned multiple labels. Compared with conventional IPC- or keyword-based filtering approaches, the ISPCM significantly reduces noise and improves accuracy, thereby providing a more robust foundation for subsequent network analysis. A set of 700 samples was independently labeled by three experts. By employing a majority rule, a classification was deemed correct if it was agreed upon by at least two experts. The final results show that the ISPCM achieved an accuracy of 92%, while that of the traditional method reached 61%.”]

 

Comments 3: [Clarify the scope of the dataset and selection bias. The analysis is based on CNIPA patents. Explain how excluding non-CNIPA applications (e.g., EPO/USPTO/WIPO) affects the cross-border collaboration inference and the "decoupling" narrative; consider creating a robustness subset that includes links to PCT families or joint assignments abroad to avoid home jurisdiction bias.]

Response 3: [Thank you for raising the important point regarding the scope of our dataset and the potential for selection bias. We agree that this is a valuable consideration. In direct response to your comment, we have revised the manuscript to explicitly acknowledge this limitation. A paragraph has been added to the "Discussion" section of the paper, which now states:

"This study relies exclusively on patent data from the CNIPA. Consequently, our analysis does not capture cross-border collaborations that resulted only in patent filings in other jurisdictions (e.g., EPO, USPTO, WIPO). This may introduce a 'home jurisdiction bias', indicating that our findings regarding international 'decoupling' are primarily reflective of the collaboration dynamics within the Chinese domestic innovation system. Future research incorporating international patent families would provide a more global per-spective."

We believe this addition provides the necessary context and sets appropriate boundaries for interpreting our findings. Thank you for this constructive feedback, which has helped improve the clarity of our work.]

 

Comments 4: [Defining "collaboration" and building networks. Collaboration is defined as multiple applicants per patent, analyzed as an undirected edge. This omits (i) inventor-level collaboration, (ii) applicant-inventor interrelationships, (iii) changes in licensing/assignments, and (iv) edge weighting (number of co-applications). Please (a) present a sensitivity analysis using inventor co-authorship networks and weighted applicant networks; (b) justify the undirected assumption; and (c) explain how the prevalence of utility models and corporate structures in China may reduce formal co-application rates (~3%).]

Response 4: [Thank you for these extremely insightful and constructive comments. Your feedback regarding the definition and construction of "collaboration" networks is particularly pertinent and has provided crucial direction for refining our methodology. We fully acknowledge that our initial simplified approach overlooked several important dimensions of collaboration. Below we detail the revisions and explanations made in response to your suggestions:

Clarification and Acknowledgment of Limitations in Collaboration Definition:

We have supplemented the "Discussion" sections of the paper to explicitly state that:

The collaboration network constructed based on co-applicants serves as an effective yet incomplete indicator of formal inter-institutional collaboration. We acknowledge that this approach indeed fails to capture:

(i) Informal collaboration at the inventor level (measured through co-invention);

(ii) Complex relationships between applicants and inventors (e.g., service inventions under employment relationships);

(iii) Collaborations formed through subsequent licensing or assignments;

(iv) Variations in the intensity of collaboration between institutions.

Responses to Specific Requests:

(a) Sensitivity Analysis:

Weighted Applicant Network: We thank the reviewer for this suggestion. We wish to clarify that the applicant collaboration network analyzed in our study is, in fact, a weighted network, where edge weights are defined by the count of co-applications between two entities. The core trends and conclusions (e.g., the strengthening of domestic collaborative intensity) are based on the analysis of this weighted network, which inherently accounts for the variation in collaboration strength. This methodological choice directly addresses the concern about simple co-applicant counts.

Inventor Collaboration Network: We agree that this is a highly valuable perspective. However, due to the significant challenges in disambiguating inventor names in CNIPA data (e.g., homonyms, different names for the same person), constructing a clean co-occurrence network of inventors suitable for trend analysis was not feasible in this study. We have explicitly noted this limitation and identified it as a focus for future work—specifically, to compare institutional and inventor networks after rigorous name disambiguation.

(b) Justification for the Undirected Assumption: We have supplemented the methodology section with the rationale for using undirected edges:

Theoretical Basis: In innovation research, co-applying for a patent is generally regarded as a symmetric partnership. Both parties jointly own the property rights and hold equal legal status. The nature of this relationship is directionless.

Practical Consideration: Patent data itself does not indicate which party initiated the collaboration. Therefore, treating edges as undirected is the most common and reasonable approach, consistent with other major studies on collaboration networks.

(c) Explaining Low Co-application Rates in the Chinese Context:

This point is particularly crucial. We have added a paragraph to the Result section in page specifically addressing how characteristics of the Chinese patent system affect our findings:

“Notably, the high proportion of utility model patents and the prevalence of single corporate structures in China may indeed suppress formal co-application rates. Utility models typically involve incremental innovations, where the need for collaboration may be lower. More importantly, in China, technical collaboration between parent and sub-sidiary companies or among different entities within the same group is often managed through internal governance mechanisms and does not necessarily manifest as legal co-applications. Consequently, a substantial amount of de facto collaboration is not captured by our method. Therefore, we emphasize that the approximately 3% collabo-ration rate revealed in this study should be interpreted as a baseline measure of "formal, cross-institutional, equally shared" collaboration within the Chinese innovation system.”

Your comments have significantly enhanced the rigor and depth of our study. We believe that through the additional analyses, methodological justifications, and contextual discussions outlined above, the paper's treatment of the definition, measurement, and limitations of "collaboration" has become more transparent and robust. All these revisions have been incorporated into the revised manuscript. ]

 

Comments 5: [“Small-world” and “scale-free” claims require statistical evidence.You report high clustering (C≈0.711) and short path length (L≈3.738), and mention a scale-free structure. Please provide (i) the small-world condition in relation to paired random graphs, and (ii) formal power-law tests (e.g., Clauset-Shalizi-Newman goodness of fit with alternative heavy-tailed distributions) before concluding that the distribution is “scale-free.” Include graphs and p-values; otherwise, temper the claim that the distribution is “heavy-tailed.”]

Response 5: [Thank you for this rigorous and crucial feedback regarding our assertions about the network structure. You are absolutely correct that our initial use of terms such as "small-world" and "scale-free" requires solid statistical testing for support, and we indeed fell short in this regard.

In accordance with your feedback, we have made the following important revisions to the relevant statements in the manuscript:

Regarding the "small-world" property:

We have removed conclusive statements directly claiming that the network exhibits "small-world" characteristics.

Instead, we have reframed this in the Results section as a descriptive finding: "We observed that this collaboration network simultaneously exhibits a high clustering coefficient (C ≈ 0.711) and a short average path length (L ≈ 3.738)." In the Discussion, we note that this structural feature may facilitate the local accumulation and rapid dissemination of information within the network, while avoiding directly equating it with the strictly defined small-world property.

Regarding the "scale-free" and "heavy-tailed" distribution:

We have removed the term "scale-free."

We acknowledge that formal power-law fitting tests (such as the Clauset-Shalizi-Newman method) are beyond the current scope of this study. Consequently, we have toned down the description to state: "the node degree distribution exhibits a heavy-tailed characteristic," and we point out that this reflects the heterogeneity of collaborative relationships within the network, namely the existence of a small number of hub nodes with a large number of connections. We have removed all inferences regarding "scale-free."

We believe that, although rigorous statistical tests are lacking, reporting these highly significant empirical observations (high clustering, short path length, heavy-tailed distribution) itself holds important indicative value for characterizing the structural features of the collaborative ecosystem in China's new energy vehicle industry. Your feedback has helped us avoid overinterpreting the data and made our conclusions more accurate and robust.

We have incorporated all the above revisions into the revised manuscript.]

 

Comments 6: [Time modeling and policy inference.The three-phase evolution is plausible, but please add networks with time windows (mobile or political regime), report on assortativity, community persistence, and node turnover. When attributing changes to specific policies (e.g., “Ten cities, one thousand vehicles”), consider using graphs or event study-style references to support the causal language.]

Response 6: [Thank you for your insightful comments regarding the temporal modeling and policy inferences. Your points concerning the rigor of dynamic network analysis and causal attribution are crucial for improving our study.

We have carefully considered your suggestions and have made the following key revisions to the manuscript to more accurately reflect the boundaries and contributions of our research:

1. Regarding Dynamic Network Analysis and the "Three-Phase" Periodization:

We fully agree that analyzing dynamic metrics such as assortativity and community persistence through time windows could provide richer evidence for the phase. However, the computations and validations required for such analyses are beyond the scope and timeframe of this revision.

Therefore, we have chosen to address this issue with greater prudence and transparency:

We have added a paragraph to the "Discussion" section of the paper, explicitly stating: "The three-phase evolution model proposed in this study is primarily based on the observation of time-series macro-level network metrics (e.g., number of nodes, number of links). While heuristic, future research could validate and refine this model by employing more granular dynamic network analysis methods, such as introducing overlapping time windows and calculating community persistence indices. Future studies also could employ temporal network analysis or exponential random graph models to better capture the dynamics of collaboration formation and network evolution."

By doing so, we both acknowledge the boundaries of our current approach and indicate a direction for future research.

2. Regarding Causal Inferences about Policy Impact (The Core and Key Revision of This Response):

This is an area where we can and must make immediate improvements. We completely agree that observational data cannot firmly establish strict causality.

We have systematically revised the entire text to remove or substantially weaken all direct causal claims regarding policies (e.g., the "Ten Cities, One Thousand Vehicles" program).

The revised formulations consistently emphasize temporal correlation rather than causality. For example, we have changed phrases like "this policy caused the expansion of the collaboration network" to "during the period following the implementation of this policy, we observed an expansion trend in the collaboration network" or "changes in network structure coincided temporally with the rollout of specific industrial policies."

Simultaneously, we have added a statement in the Discussion section clarifying that the observed correlations could be influenced by other co-evolving factors, such as technological maturation and market development, thereby avoiding overinterpretation.

In summary, your comments have prompted us to maintain the highest degree of caution in our inferences. We believe that by clearly defining the study's limitations and thoroughly tempering the causal language, the revised manuscript has achieved a substantial enhancement in academic rigor.]

 

Comments 7: [The segment-level findings require operational definitions.The authors indicate that components act as a central node and that complete vehicles are relatively isolated. Please add segment-level metrics (density, modularity contribution, intersegment mix matrix) and define “isolation” quantitatively (e.g., lower external-to-internal slope ratio).]

Response 7: [Thank you for pointing out the lack of precision in our descriptions of the various segments in our study. We fully agree that the "network scale" should be more accurately defined in operational terms.

In accordance with your suggestions, we have made the following important revisions to the manuscript:

Clarification and Quantification of Key Definitions:

We have removed the original imprecise descriptions of "hub" and "relative isolation" and instead adopted more accurate descriptions of network scale:

“Across the industrial chain, the component segment forms the largest network, the complete vehicle segment comprises the smallest, and the aftermarket is clustered around battery recycling..”

We believe that by adopting these direct and measurable operational definitions, our descriptions of the network scale across different segments now rest on a more accurate quantitative foundation. All these revisions have been incorporated into the "Methods" and "Results" sections of the paper.]

 

Comments 8: [Figure and table clarity & consistency. Please ensure: (i) all axis labels and legends are readable; (ii) consistent node/edge color semantics across figures; (iii) numeric tables include units and time spans; and (iv) captions fully describe data and construction (e.g., “multi-applicant, undirected, weighted by co-applications?”). Table 2 is useful—consider adding per-period versions.]

Response 8: [We have implemented all the requested revisions as follows:

Enhanced the readability of all figures and charts, ensuring all axis labels and legends are clearly legible.

Unified the color semantics for nodes and edges across all visualizations.

Improved data tables by adding measurement units and time range specifications.

Rewrote all figure and table captions to include complete data sources and construction methodologies.

The period-based statistical table has been added to the manuscript, showing the evolution of network metrics across different time periods.]

All modifications have been incorporated into the revised manuscript. We thank the reviewer for these valuable suggestions.

 

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

I would like to thank the authors for the opportunity to review this manuscript. The article addresses a highly relevant topic, as it seeks to combine expert knowledge and artificial intelligence techniques to improve patent classification. Overall, the manuscript presents a clear structure, fluent language, and an original proposal. However, the following points are suggested for improvement:

-It is recommended to include a broader and more up-to-date literature review that specifically addresses theories and approaches on technological innovation and innovation ecosystems, as well as international studies analysing the evolution of the new energy vehicles sector. At present, the review is limited to descriptive and regional studies, lacking a solid conceptual framework that would allow the Chinese case to be contrasted with global trends or with current academic literature.

-A critical discussion of the reviewed literature is missing, as is a dedicated conclusions section. This omission is significant, as it prevents the paper from providing an interpretive synthesis of the findings and reflecting on their academic, technological, and policy implications. It is therefore suggested to include these sections to strengthen the study’s contribution to academic knowledge and decision-making within the NEV sector.

Author Response

I would like to thank the authors for the opportunity to review this manuscript. The article addresses a highly relevant topic, as it seeks to combine expert knowledge and artificial intelligence techniques to improve patent classification. Overall, the manuscript presents a clear structure, fluent language, and an original proposal. However, the following points are suggested for improvement:

 

Comments 1: [It is recommended to include a broader and more up-to-date literature review that specifically addresses theories and approaches on technological innovation and innovation ecosystems, as well as international studies analysing the evolution of the new energy vehicles sector. At present, the review is limited to descriptive and regional studies, lacking a solid conceptual framework that would allow the Chinese case to be contrasted with global trends or with current academic literature.]

Response 1: [We thank the reviewer for this suggestion. We agree that placing the Chinese case within a broader global context is an important endeavor. In our initial literature review, we primarily focused on the methodological challenges of patent classification and the application of complex network theory in innovation studies, as these are the core contributions of our work.

We acknowledge that expanding the review to include a detailed comparative analysis of international NEV sector evolution would be valuable. However, conducting a truly parallel, network-based analysis of NEV innovation ecosystems in other countries (e.g., the US or Europe) is challenging due to significant differences in patent data accessibility, classification systems, and the structure of industrial chains. Such a comprehensive international comparative study is a substantial research project in its own right and falls outside the primary scope and page constraints of this paper, which is to introduce and validate the ISPCM and delineate the structure of China's NEV innovation ecosystem.

Nevertheless, to better frame our study, we have now strengthened the introduction and literature review by: More explicitly stating that our findings and proposed model are derived from and specific to the Chinese context,

However, we add two newest references that specifically addresses theories and approaches on technological innovation and innovation ecosystems into Section 2.

23. Huang, Y.; Sun, Q.; Chen, Z.; Wenzhong Gao, D.; Bach Pedersen, T.; Guldstrand Larsen, K.; Li, Y. Dynamic Modeling and Analysis for Electricity-Gas Systems With Electric-Driven Compressors. IEEE Trans Smart Grid 2025, 16, 2144–2155, doi:10.1109/TSG.2025.3527221.
24. Wu, D.; Zhao, Q.; Xia, X.; Liu, C.; Xu, Y.; Li, Y. Road Adhesion Information Estimation of Connected Auton-omous Vehicle Based on Digital Twin and Vision Senor Fusion. IEEE Internet Things J 2025, 1–1, doi:10.1109/JIOT.2025.3621457.]

 

 

Comments 2: [A critical discussion of the reviewed literature is missing, as is a dedicated conclusions section. This omission is significant, as it prevents the paper from providing an interpretive synthesis of the findings and reflecting on their academic, technological, and policy implications. It is therefore suggested to include these sections to strengthen the study’s contribution to academic knowledge and decision-making within the NEV sector.]

Response 2: [We sincerely thank the reviewer for this critical suggestion. We fully agree that a critical synthesis of the literature and a dedicated conclusions section are essential for interpreting the findings and articulating the study's contributions.

In direct response, we have made the following significant revisions to the manuscript:

Added a Critical Summary Paragraph at the end of the Literature Review (Section 2):

We have added a new paragraph that critically discusses the limitations of prior studies and explicitly positions our research within the existing academic conversation. The new text summarizes how our study addresses key gaps, specifically through the development of the ISPCM and the multi-dimensional (temporal, industrial, spatial) analysis of the patent collaboration network.

“In summary, while existing research has made significant progress in analyzing the new energy vehicle innovation network using patent data, it still suffers from two core limitations.

First, at the data level, a heavy reliance on IPC codes or keyword searches for patent identification leads to insufficient accuracy in cross-disciplinary fields such as NEVs, making it difficult to support fine-grained industrial chain analysis.

Second, from an analytical perspective, most studies either focus on macrolevel trend descriptions or are confined to static analyses of network topology, lacking an integrated framework that combines temporal evolution, the industrial chain structure, and the dynamics of collaboration networks.

This study aims to address these challenges by introducing the ISPCM to construct a more accurate CNEVIP dataset. Building on this foundation, we systematically characterize the evolutionary trajectory and structural mechanisms of China's NEV patent collaboration network across temporal, industrial, and actor dimensions, providing new analytical perspectives and empirical evidence for understanding China's rapid rise in this field. ”

Change the new "Discussion" section provides a concise synthesis of the main findings, systematically outlines the study's academic contributions (to patent analysis methodology and innovation ecosystem theory) and policy implications (for fostering collaboration and mitigating decoupling risks within the NEV sector), and suggests clear directions for future research.

We believe these revisions have significantly strengthened the manuscript's narrative flow and scholarly impact. Thank you again for these constructive recommendations.

“This study introduces the ISPCM for China's NEV sector, integrating expert knowledge with LLMs to enhance the accuracy and relevance of patent screening. Applying this method, we systematically identified and analyzed NEV patents filed be-tween 2001 and 2022, constructing for the first time a comprehensive patent collaboration network that examines temporal evolution, the industrial chain structure, and applicant nationality. The analysis provides novel insights into the structural mechanisms driving China's global leadership in NEVs, offering significant theoretical contributions to innovation ecosystem research and substantive policy implications for sustainable industrial development. The key findings are as follows:

NEV patent filings in China have grown rapidly and continuously, evolving through three stages: initial development (2001–2008), accelerated growth (2009–2017), and maturity (2018–2022). Policy initiatives, such as the "Ten Cities, Thousand Vehicles" program, were implemented alongside market expansion during the observed period. The observed correlations could be influenced by other co-evolving factors, such as technological maturation and market development. Analysis of patent filings reveals that domestic applicants dominate throughout, with invention patents being more prevalent during the technology accumulation phase, while utility model patents be-come more common during the subsequent industrialization stage. This pattern in patent types is consistent with a transition from foundational R&D toward more application-oriented, iterative innovation.

 

...

In terms of the divergent innovation patterns across the industrial chain segments, the component segment exhibits a dual structure that combines state-led collaboration and market-driven R&D. The complete vehicle segment persists as a tightly knit "exclusive club," a structure defined by limited collaboration and intense internal competition. The aftermarket segment (e.g., battery recycling and reuse) forms specialized innovation clusters that are led by firms such as Brunp Recycling and GEM. Notably, the influence of the SGCC does not extend deeply into vehicle manufacturing, thereby revealing challenges in achieving full value chain integration.

Domestic and foreign applicants operate largely in parallel, with domestic net-works forming a policy-shaped mega-ecosystem with quantitative dominance and foreign networks forming exclusive “elite clubs” focused on high-value invention patents. The number of cross-ecosystem ties is minimal (with only 24 collaborative links), which indicates limited deep technological exchange and a potential decoupling risk. This study relies exclusively on patent data from the CNIPA. Consequently, our analysis does not capture cross-border collaborations that resulted only in patent filings in other jurisdictions (e.g., EPO, USPTO, WIPO). This may introduce a 'home jurisdiction bias', indicating that our findings regarding international 'decoupling' are primarily reflective of the collaboration dynamics within the Chinese domestic innovation system. Future research incorporating international patent families would provide a more global perspective.

...”]

We are grateful for the reviewer's insightful comments, which have significantly strengthened the academic rigor and narrative clarity of our manuscript. The revised version now includes a more focused literature review, a critical discussion of prior work, and a dedicated conclusions section that clearly outlines the theoretical and practical contributions of our study. We believe these enhancements have substantially improved the overall quality and impact of the paper.

 

 

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

This paper introduces an industry-specific patent classification methodology that integrates expert knowledge with large language models to over some problems in the new energy vehicle mentioned in the previously published patents.

Although this work summarizes the work presented in the patents published between 2001-2022, its contribution to the literature, researchers, and users is limited. It is important that the authors include those patents published after 2022.

 

The authors should address the following comments:

1. Include those patents published after 2022 up to date.
2. There is a typo mistake in Line 203, the second “E” should be replaced by “N”.
3. “D” has two different meanings in Equations 1 and 3.
4. Make it clear in Line 243 that the largest eigenvalue of A.
5. In Figure 2, explain the variables mentioned in the first column.

Author Response

We sincerely thank the reviewer for their careful reading and constructive feedback. We have addressed all comments point-by-point as detailed below.

 

Comments 1: [Include those patents published after 2022 up to date.]

Response 1: [We appreciate this suggestion. The primary reason for concluding our dataset with patents filed by the end of 2022 is to ensure data completeness and analytical validity. As cited in our manuscript (Line 273, Ref. 38), the average patent authorization time in China is approximately 2.9 years. Including patents filed after 2022 would introduce a significant number of pending applications, creating a substantial lag bias and distorting the analysis of collaboration trends and authorization outcomes. Therefore, using the 2001-2022 period provides a more accurate and complete picture of the evolved innovation landscape. We have also incorporated this explanation into the manuscript.]

 

Comments 2: [There is a typo mistake in Line 203, the second “E” should be replaced by “N”.]

Response 1: [We sincerely thank the reviewer for spotting this typo. The correction has been made as suggested. The text now correctly reads: "...where E is the actual number of edges and N is the number of nodes in formula (1) and (2)..."]

 

Comments 3: [“D” has two different meanings in Equations 1 and 3.]

Response 3: [We thank the reviewer for highlighting this inconsistency. To resolve the ambiguity, we have redefined the symbol for the diameter in Equation 3. It is now represented by the letter "R" (which can be associated with "Radius" of the network, a concept related to diameter). The updated equation is presented as: ”Network diameter (R) indicates ...”]

 

Comments 4: [Make it clear in Line 243 that the largest eigenvalue of A.]

Response 4: [We have clarified the description in Line 243 (now Line 249) as requested. The sentence now explicitly states: "...and  is the largest real eigenvalue of A."]

 

Comments 5: [In Figure 2, explain the variables mentioned in the first column.]

Response 5: [We have updated the caption for Figure 2 to provide a clear explanation of the variables in the first column. The revised caption now includes: "Figure 2. Statistical overview of patents in the Chinese NEV industry. The first column lists the different patent types analyzed: 'All' represents the total number of authorized NEV patents, 'Co. Patents and Co. Rate(%)' indicates the number and rate of authorized patents with multiple ap-plicants, followed by counts for authorized invention (Type B), design (Type S), and utility model (Type U) patents."]

 

All changes have been incorporated into the revised manuscript. We believe these revisions have significantly improved the clarity and precision of our work.

 

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

This paper stydied the Evolution of China’s New Energy Vehicle Innovation Ecosystem through Patent Cooperation Networks. This article comprehensively examines the relevant developments and is of certain reference value. However, there are still some issues that need to be revised as follows:
1. This paper uses complex network analysis and patent cooperation networks to reveal the mechanism of China's rise in the field of new energy vehicles. Could the author further explore the development context of patent contents and analyze the difficulties and issues that need to be addressed urgently in the research and development process of new energy vehicles from a technical perspective?
2. Please provide the reference source for the key formula.
3.Some of the pictures are not clear. Please check the entire text to avoid such issues.
4. The analysis results shown in the figure would be even better if data for the years 2023 and 2024 could be included.
5. Please compare with the latest relevant papers such as IEEE Transactions on Smart Grid, doi: 10.1109/TSG.2025.3527221, IEEE Internet of Things Journal, doi: 10.1109/JIOT.2025.3621457  and so on.

Author Response

Thank you very much for your time and effort in reviewing our manuscript and for providing these insightful and constructive comments. We have carefully considered all the points raised and have revised the manuscript accordingly. Our point-by-point responses to your comments are detailed below. All changes in the manuscript have been highlighted in yellow for your convenience.

 

Comments 1: [This paper uses complex network analysis and patent cooperation networks to reveal the mechanism of China's rise in the field of new energy vehicles. Could the author further explore the development context of patent contents and analyze the difficulties and issues that need to be addressed urgently in the research and development process of new energy vehicles from a technical perspective?]

Response 1: [We sincerely thank the reviewer for this valuable and profound suggestion. Conducting an in-depth analysis of patent content to reveal specific R&D challenges is indeed a highly valuable research direction. However, the core analytical framework of this manuscript is focused on the structural evolution of the "collaboration network." Integrating a deep, technical content analysis at this stage would shift the paper's primary focus and exceed its intended scope. We fully acknowledge the importance of this suggestion and have therefore added a clear statement in the "Discussion" section (Section 5) that "Beyond classification, future research could apply natural language processing to enable the semantic mining of patent texts, such as the identification of emerging technical themes, the tracing of technology trajectories, and the detection of innovation gaps, and integrate these insights into a network analysis to improve strategic fore-casting." We believe this will form a significant and logical continuation of the present study.]

 

Comments 2: [Please provide the reference source for the key formula.]

Response 2: [We apologize for this oversight. The source for the eigenvector centrality formula (Equation 9) has now been properly cited. The reference is: Newman, M.E.J. The Structure and Function of Complex Networks. SIAM Review 2003, 45, 167–256, doi:10.1137/S003614450342480. This reference has been added to the reference list.]

 

Comments 3: [Some of the pictures are not clear. Please check the entire text to avoid such issues.]

Response 3: [Thank you for pointing this out. We have re-exported all figures in the manuscript at a higher resolution and standardized the fonts and line weights to ensure they remain clear and legible at different zoom levels. We have verified the clarity of all figures in the final version.]

 

Comments 4: [The analysis results shown in the figure would be even better if data for the years 2023 and 2024 could be included.]

Response 4: [We agree that including the most recent data would be ideal. The primary reason for concluding our dataset with patents filed by the end of 2022 is to ensure data completeness and analytical validity. As cited in our manuscript (Line 273, Ref. 38), the average patent authorization time in China is approximately 2.9 years. Including patents filed after 2022 would introduce a significant number of pending applications, creating a substantial lag bias and distorting the analysis of collaboration trends and authorization outcomes. Therefore, using the 2001-2022 period provides a more accurate and complete picture of the evolved innovation landscape. We also add this explain to the manuscript.]

 

Comments 5: [*Please compare with the latest relevant papers such as IEEE Transactions on Smart Grid, doi: 10.1109/TSG.2025.3527221, IEEE Internet of Things Journal, doi: 10.1109/JIOT.2025.3621457 and so on.*]

Response 5: [We thank the reviewer for suggesting these highly relevant and recent publications. We have now studied these papers and integrated a discussion of them into our manuscript. Specifically, we have added a comparative analysis in the "Literature Review" (Section 2). ”We highlight how our study, which focuses on the structural collaboration network, provides a complementary perspective to technology-focused approaches (e.g., energy scheduling in smart grids [23] and IoT perception [24]). Doing so enriches the overall understanding of the NEV innovation ecosystem.”]

 

Once again, we extend our deepest gratitude for your thorough review and valuable guidance, which have significantly improved the quality of our work.

 

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

Based on the changes made by the authors, I believe that the article meets the minimum requirements for publication.

No more comments

Comments on the Quality of English Language

The English could be improved to express the research more clearly.
Several typographical and formatting problems were noted, as well as some unclear phrases, for example, "and if it forms an oligopolistic structure," and inconsistent punctuation in the section lists.

 

Reviewer 3 Report

Comments and Suggestions for Authors

All the comments have been incorporated.