Potential Analysis of Technological Value in the Intelligent Connected Vehicles Field from the Patent Licensing Perspective

Abstract

Patent licensing is essential for sustainable technological diffusion, fostering innovation and strengthening industrial resilience. However, the determinants influencing patent licensing decisions remain underexplored. This study investigates these factors at both the enterprise and patent levels, emphasizing their role in promoting sustainable industrial innovation and knowledge transfer. Given the low proportion of licensed patents, this research proposes a measurement framework to identify thematically similar but unlicensed patents and applies a conditional logistic regression model to analyze the factors affecting licensing decisions. Using patent abstracts from the intelligent connected vehicles (ICVs) sector, topic modeling is conducted to classify technological themes, and Kullback–Leibler divergence is applied to measure differences between licensed and unlicensed patents. The results indicate that technological prestige and depth negatively influence licensing, whereas technological breadth, advancement, and stability have a positive effect. From a sustainability perspective, enterprises should optimize technology management to support responsible knowledge transfer and green innovation. Universities should enhance patent quality and innovation impact to contribute more effectively to sustainable development. Policymakers should refine patent licensing frameworks to foster an efficient, inclusive, and sustainable intellectual property ecosystem, thereby facilitating cross-sectoral technology diffusion, advancing eco-friendly industrial transformation, and promoting sustainable economic growth.

1. Introduction

As the leading prospective sustainable large-scale intelligent terminals for the forthcoming years, intelligent connected vehicles (ICVs) have emerged as a crucial arena for technological innovation and sustainable industrial transformation among global superpowers. To strengthen their competitive edge, multinational corporations have sequentially established core patented technologies within this domain, underscoring its role in driving eco-friendly and intelligent mobility solutions. It is posited that enterprises can enhance their foundational patent portfolio via acquisitions, mergers, licensing agreements, and other strategic initiatives, thereby leveraging intellectual property assets to support a greener and more resilient ICV industry value chain. Concurrently, facilitating cross-licensing arrangements between leading corporations and smaller enterprises can accelerate technological diffusion, encourage inclusive innovation, and foster a more sustainable industrial ecosystem. In the realm of intelligent and sustainable transportation, technological advancements are progressively reaching a state of maturity, whereas market expansion remains nascent. Hence, efficiently transforming innovative technological breakthroughs into environmentally and economically viable solutions is imperative. In the age of global open innovation, collaborations among enterprises, universities, and research institutions are burgeoning, with patent licensing emerging as a critical avenue for enhancing technological spillovers, knowledge sharing, and sustainability-driven advancements [1]. Through a systematic examination of licensed patents, insights can be gleaned into the patterns of sustainable technology dissemination and licensing dynamics across diverse nations and sectors [2].

From a patent characteristics standpoint, factors such as citation frequency, publication time, and protection scope can exert influence on the licensing process. Presently, there exists a body of research that delves into the correlation between citation frequency and the propensity for patent licensing. Notably, a positive relationship has been observed between patent licensing activities and citation frequency. A higher citation frequency typically signifies a more intricate and expansive patent portfolio, thereby piquing the interest of potential licensees and conferring a favorable position upon the licensor [3,4]. However, the association between citation and licensing is not strictly linear, as various measurement indicators or variables impacting citation rates, including enterprise size and invention scope, can also exhibit significant correlations with licensing outcomes [5]. Moreover, it is posited that specific temporal milestones bear relevance to patent licensing dynamics [6]. The quality and applicability of technologies serve as pivotal determinants driving licensing transactions, with technologies characterized by enhanced transferability and versatility being more predisposed to licensing agreements [7]. While the existing research predominantly explores the link between specific patent attributes and licensing activities, it is crucial to acknowledge that the decision-making process governing patent licensing is multifaceted and emanates from the interplay of diverse factors.

The patent examination duration has also drawn increasing academic attention in recent years. Several studies have examined how delays in the examination process affect patent value, technology transfer, and firm innovation dynamics. For example, Yan et al. [8] analyzed invention patents in China’s photovoltaics industry and identified various technical and applicant-related factors that influence examination time. Similarly, Yang et al. [9] explored how priority examination mechanisms in mission-oriented fields like climate change and COVID-19 affect both the speed and quality of patent grants. Furthermore, Graham et al. [10] introduced a comprehensive dataset (PatEx) for studying USPTO examination processes, improving empirical accessibility to such data post-2000. These studies highlight the importance of timely patent granting for maximizing economic and technological returns. Extended examination periods have been found to delay the practical deployment of innovations, hinder the pace of technology transfer, and potentially discourage firms from pursuing further innovation due to uncertainty in patent protection. However, compared to examination lag, licensing behavior—particularly in complex and emerging fields such as intelligent connected vehicles (ICVs)—remains underexplored. Our study thus complements this growing body of work by shifting the analytical focus from the upstream (application and grant) to the downstream (commercialization and licensing), addressing how patent-level and organizational attributes jointly determine licensing outcomes in the ICV sector.

In contrast to citation-based studies that focus on knowledge diffusion and technological originality [11], our study adopts a fundamentally different approach by examining licensing behavior as a downstream indicator of commercialization and economic utility. Citation metrics, while valuable, often reflect academic or technological recognition rather than actual market uptake. Licensing, on the other hand, captures the real-world application of patented technologies, particularly in emerging and application-oriented sectors like ICVs. Moreover, our study advances the literature methodologically by integrating topic modeling and conditional logistic regression to assess how latent technological themes and organizational factors jointly shape licensing decisions. In this sense, our approach complements citation-based perspectives by offering a market-facing and policy-relevant lens for understanding how patents generate value through transfer and reuse.

When considering companies with patents, attributes such as the size of the licensor play a crucial role in patent licensing dynamics. Existing research offers diverse perspectives on the relationship between enterprise size and licensing activities. Shen et al. posit an inverted U-shaped correlation between licensor prestige and the likelihood of licensing agreements, indicating that firms with moderate prestige are more likely to engage in licensing activities and facilitate technology transfer [12]. Conversely, Li et al. demonstrated a U-shaped relationship between enterprise size and licensing, suggesting that both small and large enterprises are more inclined to participate in licensing, particularly in highly competitive markets [13]. Park’s analysis highlights that company characteristics, particularly size and associated factors, influence outbound licensing activities while having less impact on inbound licensing behaviors [14]. Furthermore, differences in economic development and levels of technological innovation across countries contribute to variations in licensing behavior. Motohashi explored a nonlinear association between enterprise size and licensing behaviors, focusing on how organizational structures shape licensing strategies in Japanese enterprises [15]. Feng’s research underscores a significant self-selection effect in the licensing behavior of Chinese manufacturing enterprises, demonstrating that firms with strong innovation capabilities are more likely to engage in patent licensing, thereby reinforcing their market competitiveness [16]. Additionally, smaller enterprises often face greater challenges in resolving patent licensing issues compared to larger European firms, which typically have well-established intellectual property management systems [17]. However, existing research findings may not be directly applicable to the intelligent connected vehicle (ICV) sector, due to its unique characteristics. As a key enabler of green mobility and intelligent transportation, the ICV industry integrates advanced digital and automotive technologies to enhance energy efficiency, reduce emissions, and optimize transportation networks. The rapid evolution of this industry creates new challenges and opportunities in patent licensing, particularly regarding the role of technological depth and breadth. These factors significantly influence licensing activities, yet they have received limited scholarly attention, highlighting the need for further research on the licensing mechanisms in the ICV sector from a sustainable development perspective.

Despite the existing body of literature exploring various factors that influence patent licensing, research focusing specifically on the ICV sector remains sparse. This study contributes to the literature in two significant ways: First, it offers the first systematic investigation of patent licensing mechanisms in the ICV sector, a domain that integrates digital and automotive technologies for sustainable mobility. Second, it adopts an interdisciplinary methodological framework that combines topic modeling and conditional logistic regression, enabling us to capture latent technological themes and assess their impact on licensing likelihood in a statistically robust manner. This integrated approach bridges computational text analysis with econometric modeling, providing new insights into how thematic similarity and organizational attributes interact to shape licensing outcomes.

In the realm of intelligent connected vehicles (ICVs), licensing activities are influenced by both patent-specific features and licensor attributes. However, the licensing dynamics within the ICV industry remain underexplored, particularly from the perspective of sustainability and industrial transformation. This study explicitly seeks to investigate how enterprise-level factors and patent-level characteristics jointly influence the likelihood of patent licensing in the ICV domain. To this end, we propose a structured research approach: patent abstracts related to ICV technologies from recent years are selected as the primary dataset, upon which topic modeling is conducted to identify thematic areas. The similarity between licensed and unlicensed patents is measured based on these themes to facilitate comparative analysis. By examining both licensed patents and their most topically similar unlicensed counterparts, this study aims to uncover patterns and determinants that shape licensing decisions. The findings are expected to contribute to a deeper understanding of the mechanisms governing patent licensing in the ICV industry and offer practical implications for promoting technology transfer and sustainable innovation.

2. Prior Review

In the field of innovation economics, patents are widely employed as indicators of inventive activity, yet their macro-economic pay-offs remain heterogeneous. Large-sample market-value evidence shows that citation-weighted patent stocks—especially those rich in ICT technologies—are strongly associated with firm-level growth in advanced economies [18]. At the macro level, the landscape survey by Mejia and Kajikawa [19] indicates that the same accumulation of ICT patents can yield diminishing or even negative returns in emerging contexts when absorptive capacity is limited. For China, the findings in that survey also reveal that subsidy-fueled increases in patent counts have come at the expense of quality, creating ‘policy noise’ and prompting strategic low-value filings [19]. Across Europe, the diffusion of ICT, rising research intensity, and an active venture-capital ecosystem are recognized as core drivers of sustainable growth, yet Bacchiocchi and Montobbio [20] showed that cross-regional knowledge flows are hampered by a pronounced home bias in patent citations, implying spatially uneven benefits. Regional transfer studies likewise uncover complex effects: while patent transactions spur substantive innovation in leading cities, they may crowd out sustainable innovation in lagging regions [20]. The Scandinavian panel work summarized by Higham et al. [21] links patent stocks and human–capital complementarities to long-run GDP gains, yet it also reports a surprising negative association between education spending and growth, underscoring the context-dependence of patent returns. Theoretical modeling of institutional change further suggests that stronger protection can accelerate an economy’s take-off from stagnation, though it may dampen steady-state growth if not balanced with diffusion incentives [22]. Finally, simulation evidence based on semantic novelty metrics points to contrasting distributional effects: robust patent protection tends to widen income gaps, whereas realistically scaled R&D subsidies help narrow them [23].

In sum, the recent literature confirms that patents are a pivotal—yet context-sensitive—instrument for innovation-led development. Their effectiveness hinges on the national development stage, institutional quality, and regional capabilities; hence, patent strategies must be embedded in a broader, well-functioning innovation ecosystem to advance both economic growth and social equity.

The investment in research and development (R&D) capabilities by licensees can be facilitated through engaging in patent licensing agreements [24]. Given the higher labor and time costs associated with internal R&D activities, licensees are often inclined to acquire knowledge from external sources rather than develop it internally [25]. The decision to enter into a patent licensing arrangement is typically contingent upon both the licensee’s and the licensor’s mutual interest in maximizing their respective benefits. The likelihood of technology licenses being transacted is influenced by a blend of enterprises’ characteristics and patent attributes [26].

In this study, we focus on several key factors at the enterprise level and the patent level that impact the licensing process. Enterprise-level factors include technology prestige, technology depth, and technology breadth. On the other hand, patent-level factors such as technological advancement, technological stability, and protection scope also exert influence on the licensing dynamics.

2.1. Licensor Prestige

Licensor prestige plays a pivotal role in shaping licensing outcomes. Prestigious inventors and universities often command higher transaction prices for licensed technologies due to enhanced perceived value and stronger bargaining positions. This effect is reinforced when the supplying university also holds high prestige but attenuated when the buying firm is itself prestigious, indicating prestige-based asymmetries in negotiation power [27].

However, prestige does not always correlate with commercialization success. Evidence from a leading U.S. university shows that while academic stars receive more support for patent filings, it is licensing experience—not prestige—that predicts real market impact, revealing a possible misallocation of TTO resources [28].

Moreover, prestige influences not just price but also contract structure. An inverted U-shaped relationship exists between university prestige and exclusive licensing likelihood, suggesting that moderately prestigious institutions are more strategically positioned for exclusive deals. This relationship is moderated by TTO capabilities and past collaboration with firms [12].

In inter-firm settings, organizational prestige signals reliability and reduces uncertainty. Reputable firms are more likely to license late-stage technologies, while firms with weaker reputations but advantageous networks rely on borrowed prestige to access early-stage innovations [29].

2.2. Technological Depth

Technological depth, which signifies an enterprise’s level of specialization in specific research fields, has been observed to have a negative correlation with patent licensing activities.

By adopting a technology specialization strategy, enterprises can strengthen their technological capabilities within established domains while minimizing marginal innovation costs, thereby fostering stable and sustained innovation-driven growth under mature technological paradigms [30]. In contrast, firms that have not yet established a foothold in specific technological fields may exhibit a greater propensity to license out their patents. Such licensing behavior often stems from limited internal capacity or capability to realize the full value of their innovations, prompting firms to rely on external partners to commercialize unused technological assets [31].

Moreover, as the co-evolution of intellectual property systems and technological advancement continues, enterprises increasingly regard the protection of intellectual property rights not only as a legal safeguard but as a strategic prerequisite for deepening innovation pathways and sustaining long-term competitive advantage [32].

2.3. Technological Breadth

Technological breadth—the range of distinct technological domains in which a firm or invention is embedded—has been shown to affect both patent value and firm-level innovative performance. Recent evidence indicates that technological breadth displays a U-shaped association with patent impact: inventions that draw on either narrowly focused or very wide sets of technological inputs outperform those with moderate breadth, with the upward curvature steeper for technological breadth than for scientific breadth [33]. At the firm level, a wider portfolio of technological domains enhances innovation output, suggesting that diversification across knowledge fields provides complementary resources that stimulate creativity and experimentation [34]. During periods of environmental turbulence, such as financial crises, firms exhibiting greater openness breadth—i.e., tapping a broad set of external knowledge sources—achieve superior growth, although the marginal benefits of additional breadth eventually diminish [35]. Collectively, these findings underscore that expansive technological breadth can elevate the attractiveness of a patent portfolio, enrich the knowledge exchange pool, and ultimately reinforce the incentives for patent licensing, while also cautioning that optimal breadth may be context-dependent and subject to decreasing returns.

2.4. Technological Advancement

Although scholars increasingly call for holistic and reproducible frameworks to gauge technological advancement, empirical efforts remain fragmented and indicator definitions inconsistent [36]. Recent machine learning research on graphene-based supercapacitor patents demonstrates that citation network metrics—such as in-/out-closeness and out-degree centrality—are the most salient predictors of transfer and licensing potential, highlighting the premium that technologically advanced inventions command in the market [37]. Complementary work in the mechanical products sector developed a multi-tier evaluation scheme that weights technical attributes via entropy, principal component, and factor analyses; patents ranking high in novelty, functional complexity, and design scope were shown to accelerate downstream innovation and commercialization [38]. At the meta level, a GPT-4-assisted systematic review identified 985 distinct patent quality indicators, with forward citations, family size, and claim count dominating the literature but also suffering from definitional inconsistencies that hinder comparability [36]. Collectively, these studies suggest that rigorous assessment of technological advancement should integrate citation intensity, network position, textual novelty, and structural scope, as patents excelling along these dimensions exhibit a markedly higher likelihood of licensing and technology transfer, thereby amplifying knowledge diffusion.

2.5. Technological Stability

The technological stability of patents refers to their capacity to withstand challenges or invalidation requests following the grant of the patent, thereby significantly enhancing the overall quality and reliability of the patent [39]. This attribute holds a positive influence on the likelihood of patents being licensed out to third parties.

In delineating technological stability, researchers have examined various indicators, such as lapses in the patent’s validity, instances of litigation involving the patent, and the occurrences of pledges related to the patent, among others [40]. With the increasing amount of patent litigation, a notable proportion of patents face invalidation challenges. It is evident that patents facing partial or total invalidity are deemed to be less stable [41]. The registration of pledges on patent rights serves as a crucial metric for characterizing the value of a patent, rendering it more appealing to potential licensees [42]. Additionally, patent reexamination, invalidation proceedings, and litigation reflect the legal stability. Notably, patents of different types may exhibit varying degrees of stability, underscoring the nuanced nature of this attribute across diverse patent categories [43].

2.6. Protection Scope

The protection scope of a patent defines the boundaries that delineate the legitimate rights and interests of the patent holder vis-à-vis the general public. This scope is predicated on the claims and written description contained within the patent document. Factors contributing to the calculation of a patent’s protection scope encompass the number of claims, the presence of dependent claims, the existence of related family patents, and the remaining validity period, among others.

In terms of protection content, the number of claims serves as a barometer of patent quality. A higher number of claims typically translates to a broader protection scope for the patent. The interplay between independent and dependent claims via citation relationships not only mitigates the risk of complete invalidation but also renders the protection more comprehensive and systematic [44]. Furthermore, the quantity of family patents and the remaining valid period reflect the temporal and spatial dimensions of the protection scope, respectively. Licensees are inclined to engage with patents offering both quality and security, making patents with expansive protection scopes more attractive for licensing purposes.

To better situate our empirical analysis, this study develops a conceptual framework that integrates firm-level strategic behavior and patent-level technological characteristics within the context of sustainable technology diffusion. Cohen & Levinthal [45] posit that technologies with greater breadth and advancement are more transferable because they offer higher compatibility with a variety of organizational and technological environments. These attributes enhance the potential for licensees to recognize, assimilate, and apply external knowledge, particularly in sectors like ICVs that demand modular and cross-disciplinary innovation. In contrast, Grindley & Teec [46] suggest that firms may deliberately retain high-prestige or deep-core technologies as competitive assets. Such patents are less likely to be licensed not due to low value, but because they serve exclusionary or defensive functions in corporate innovation strategies. This theoretical perspective helps explain the observed negative association between technological prestige and licensing likelihood in our results. Furthermore, Chesbrough [47] underscores the importance of external knowledge flows and collaborative R&D in accelerating sustainable innovation. Smaller firms and universities, with limited internal R&D capacity, often rely on licensing as a pathway to both commercialize their own innovations and access complementary technologies. This insight is essential for interpreting the heterogeneity in licensing behavior identified in our case analyses.

Integrating these perspectives, we propose a multi-level conceptual framework: enterprise-level attributes (e.g., size, R&D orientation) influence the strategic deployment of patent assets, while patent-level characteristics (e.g., breadth, advancement, protection scope) shape the perceived value and accessibility of technologies. By aligning this framework with sustainability goals, we clarify how licensing serves as a mechanism for both innovation diffusion and long-term industrial resilience in the ICV sector.

3. Methods

In this paper, the study is structured into five distinct processes—data retrieval, data preprocessing, feature extraction, similarity comparison, and factor analysis—as depicted in Figure 1. (1) Data Retrieval: Initially, all patents within the intelligent connected vehicles field were retrieved and downloaded from a patent database. (2) Data Preprocessing: Subsequently, in the data preprocessing stage, relevant patent data were extracted, while filtering out missing or erroneous data. The final output was a standardized and comprehensive dataset ready for further analysis. (3) Feature Extraction: In the feature extraction process, a Latent Dirichlet Allocation (LDA) model was applied to all patent texts to extract the text-topic distribution and topic-word distribution. This step aided in identifying key features and topics embedded within the patent documents. (4) Similarity Comparison: The next phase involved calculating the Kullback–Leibler divergence (KL distance) between licensed patents and unlicensed patents to determine their similarities. Based on this analysis, the top k similar patents for each licensed patent were identified. (5) Regression Analysis: In the final stage, the top k similar patents and licensed patents were utilized to develop a regression model. Through this model, correlations between enterprise-level factors, patent-level factors, and licensing activities were explored and quantified. The insights derived from this analysis provide valuable references and support for decision-making in enterprises, universities, and governmental organizations involved in intellectual property management and technology licensing endeavors. By following this structured methodology, this research aims to offer actionable insights and guidance to stakeholders seeking to navigate the complex landscape of patent analysis and licensing within the ICV domain.

3.1. Data and Factors

The main data used in this study were obtained from the IncoPat patent database, limited with priority years ranging from 1 January 2001 to 31 December 2021. There are a total of 17,187 samples related to ICVs, of which 71 are licensed.

The retrieval model for ICV patents was (TI = ((intelligen* OR smart OR intellectual* OR wisdom* OR (automat* driving) OR (Self driving)) AND (vehicle* OR car OR cars)) AND TIAB = (network* OR connect*) AND TIABC = (transport* OR traffic* OR perception* OR detect* OR position* OR navigat* OR identificat* OR recognit* OR discriminat* OR control* OR execut* OR architect* OR scene OR predict* OR warn* OR track* OR test OR tech*) AND AD = ([20010101 TO 20211231]).

The dependent variable selected in this paper is a 0–1 variable, with a value of 1 if the patent is licensed and 0 if not.

The main explanatory variables are divided into two types: licensor variables (licensor prestige, technological depth, and technological breadth) and patent variables (technological stability, technological advancement, and patent protection scope). Licensor prestige reflects the licensor’s position, ability, and reputation in the market, represented by the number of citations. The measurement of technological depth is the number of patents applied by an enterprise with the same IPC as the licensed patents. Technological breadth indicates the diversity of the enterprise’s research technology, measured by the number of IPCs for patents applied by an enterprise. The technological stability of a patent refers to its ability to resist invalidation requests after being granted, the measurement of which includes whether a utility model patent has expired, whether litigation has occurred, whether pledge preservation has occurred, whether a request for reexamination has been made, and whether invalidation declaration has been applied for. The measurement of technological advancement includes the number of R&D personnel, the number of citations of patents with the same family, whether transfer or licensing, etc. Patent protection scope is the boundary that divides the legitimate rights and interests of the patentee and the public, the measurement of which includes the number of claims and the patent layout.

3.2. Conditional Logistic Regression

Logistic regression is a general classification model using the logit model to study the correlation between binary dependent variables and a set of explanatory variables. Assume that the vector $x=(x_1,x_2,\dots ,x_t)$, with t independent variables; conditional probability $P\left(y=1|x\right)=p$ is the probability of observation relative to event x; and the classification function $g\left(x\right)=w_0+w_1{x}_1+\dots +w_t{x}_t$. The form of the logistic regression is as follows:

$$P\left(y=1|x\right)={\displaystyle \frac{1}{1+e^{-g\left(x\right)}}}$$

(1)

In Stata16 software, the clogit function fits what biostatisticians and epidemiologists call conditional logistic regression for matched case–control groups and what economists and other social scientists call fixed-effects logit for panel data. It is also applicable in the study of factors affecting patent licensing.

3.3. Topic Modeling

Topic modeling is an algorithm for semantic analysis and topic search in the field of natural language processing, representing the process of selecting words belonging to a topic from documents. LDA is a typical model for topic modeling. LDA is a three-layer Bayesian probability model, mainly learning two probability distributions: the polynomial distribution from document to topic, and the polynomial distribution from topic to word. The probability distribution between topics and words is the proportion of different words appearing under different topics, and there is no need to pay attention to the order relationship between words.

KL distance, also known as relative entropy, is used to measure the relative differences between two probability distributions in the same space. When two probability distributions are the same, the KL distance is 0. The larger the difference between two probability distributions, the greater the KL distance. KL distance is expressed as follows:

$$D(p||q)=\left(\sum _{x\in X}p(x)log{\displaystyle \frac{p\left(x\right)}{q\left(x\right)}}\right)$$

(2)

In this paper, we apply the LDA model to extract keywords representing text features from patent documents. By calculating the distance between different feature distributions using the KL formula, the similarity between unlicensed and licensed patents can be determined, thereby predicting the probability of patent licensing.

4. Results and Analysis

4.1. The Results of Topic Modeling

In this study, the LDA model was performed on the abstract of all patents.

Table 1. The process of word segmentation, dictionary building, and vectorization.

Text	The invention discloses a parallel acceleration track planning method for intelligent vehicle GPUs, which comprises the following steps: at the CPU end, acquiring a real-time perception barrier grid diagram, global reference path, intelligent vehicle real-time GPS information, and upper-layer decision instructions.
Words	‘invention’, ‘discloses’, ‘parallel’, ‘acceleration’, ‘track’, ‘planning’, ‘method’, ‘intelligent’, ‘vehicle’, ‘gpu’, ‘comprises’, ‘following’, ‘steps’, ‘cpu’, ‘end’, ‘acquiring’, ‘real’, ‘time’, ‘perception’, ‘barrier’, ‘grid’, ‘diagram’, ‘global’, ‘reference’, ‘path’, ‘gps’, ‘information’, ‘upper’, ‘laye’.
Bigram or Trigram	‘invention’, ‘discloses’, ‘parallel’, ‘acceleration’, ‘track’, ‘planning’, ‘method’, ‘intelligent’, ‘vehicle’, ‘gpu’, ‘comprises’, ‘following’, ‘steps’, ‘cpu’, ‘end’, ‘acquiring’, ‘real’, ‘time’, ‘perception’, ‘barrier’, ‘grid’, ‘diagram’, ‘global’, ‘reference’, ‘path’, ‘intelligent’, ‘vehicle’, ‘real’, ‘time’, ‘gps’.
Frequency	(0, 1), (1, 1), (2, 1), (3, 1), (4, 1), (5, 1), (6, 1), (7, 2), (8, 2), (9, 1), (10, 1), (11, 1), (12, 1), (13, 1), (14, 1), (15, 1), (16, 2), (17, 2), (18, 1), (19, 1), (20, 1), (21, 1), (22, 1), (23, 1), (24, 1), (25, 1), (26, 1), (27, 1), (28, 1), (29, 1), (30, 1), (31, 1), (32, 1), (33, 1), (34, 1), (35, 1), (36, 1), (37, 1), (38, 1), (39, 1).

demonstrates the workflow of word segmentation, vocabulary construction, and vectorization of text data, incorporating stop-word removal to enhance semantic feature representation. Then, an LDA model can be built on all texts. The results of topic modeling are shown in Figure 2. It can be seen that the distribution among all topics is relatively scattered. After multiple attempts, it was found the best effect was achieved when the topic number was set to 18. The samples of keywords and weights generated under five topics are shown in

Table 2. The keywords of topics.

Topic	Weights × Keywords
0	0.268* drive + 0.178* motor + 0.083* wheel + 0.074* mechanism + 0.057* oil + 0.039* automatic + 0.026* shaft + 0.018* move + 0.017* rotate + 0.014* electromagnetic
1	0.085* speed + 0.073* brake + 0.048* engine + 0.044* gear + 0.039* steer + 0.034* sensor + 0.031* transmission + 0.030* turn + 0.028* instal + 0.024* guide
2	0.264* display + 0.147* mobile + 0.093* phone + 0.092* screen + 0.083* voice + 0.051* mount + 0.034* cylinder + 0.029* sound + 0.029* panel + 0.028* touch
3	0.158* power + 0.152* electric + 0.112* charge + 0.105* battery + 0.070* supply + 0.061* voltage + 0.042* energy” + 0.042* current” + 0.031* storage + 0.019* arm
4	0.075* device + 0.074* control + 0.064* connect + 0.052* intelligent + 0.034* controller + 0.033* model + 0.031* utility + 0.020* switch + 0.015* high + 0.014* use

* p < 0.1.

.

Table 3. Probability distribution of topics.

Topic	Probability	Topic	Probability
0	0.051	9	0.356
1	0.068	10	0
2	0.022	11	0
3	0	12	0.022
4	0.087	13	0
5	0.110	14	0.011
6	0.061	15	0
7	0	16	0.157
8	0	17	0.033

shows the probability distribution of a paragraph of text in the topic model.

Table 4. The results of similarity calculation.

Unlicensed Patent	Licensed Patent	Similarity
CN214450702U	CN209351336U	0.799019565
CN214450702U	CN213892890U	0.803042765
CN214450702U	CN110469629A	0.803311529
CN113085897B	KR1020120031335A	0.431203754
CN113085897B	US20190090109A1	0.43357143
CN113085897B	US20190349732A1	0.433574971
CN113085897B	US10382921B2	0.433582299
CN113085897B	CN113895543A	0.442997278
CN113085897B	CN108045378A	0.460353069
CN113085897A	CN216817624U	0.375796941
CN113085897A	CN205405260U	0.391245224
CN113085897A	CN213581848U	0.395965484
CN113085897A	CN104408953A	0.397384294
CN113085897A	CN202975679U	0.489108898
CN113085897A	CN215922159U	0.490964043
CN113085897A	CN108974014A	0.498187615
CN113085897A	CN105743902B	0.514014792

shows the results of similarity calculation between unlicensed patents and licensed patents according to the topic distributions of the texts. From this, not only is the possibility of patent licensing determined, but also unlicensed patents that are most similar to licensed patents are found.

4.2. Regression Results and Analysis

In

Table 5. Descriptive statistics and variable correlations.

Variables	1	2	3	4	5	6	7	8	9	10
1.Licensor prestige	1
2.Licensor technological depth	0.050 *	1
3.Licensor technological breadth	0.097 ***	0.380 ***	1
4.Patent technological advancement	0.333 ***	0.074 **	0.098 ***	1
5.Patent technological stability	−0.070 **	0.103 ***	0.084 ***	0.005	1
6.Patent protection scope	−0.002	0.167 ***	0.140 ***	0.104 ***	0.648 ***	1
7. Licensor research age	0.288 ***	0.207 ***	0.439 ***	0.314 ***	−0.210 ***	−0.129 ***	1
8. Licensor ownership	−0.029	0.011	−0.114 ***	−0.023	−0.179 ***	−0.034	−0.170 ***	1
9.Patent disclosure duration	0.246 ***	−0.039	−0.044	0.344 ***	−0.301 ***	−0.245 ***	0.722 ***	−0.142 ***	1
10.Patent applicant number	0.004	−0.052 *	−0.068 **	−0.031	0.01	−0.079 ***	−0.098 ***	0.141 ***	−0.044	1
Obs	1210	1210	1210	1210	1210	1210	1210	1210	1210	1210
Mean	2.891	2.168	6.266	6.271	4.883	4.964	7.911	2.952	6.012	1.118
Std. dev.	5.893	2.852	8.391	3.235	2.957	2.245	4.335	1.225	3.734	0.625
Min	0	1	1	1	1	1	1	1	1	1
Max	92	34	49	10	10	10	21	6	20	11

* p < 0.1, ** p < 0.05, *** p < 0.01.

, the descriptive statistics and correlation results for six explanatory variables and five control variables are presented. The main explanatory variables are divided into two types: licensor variables and patent variables. Licensor variables include licensor prestige, technological depth, and technological breadth. Patent variables include technological stability, technological advancement, and protection scope. Licensor research age, licensor ownership, patent age, patent disclosure duration, and patent applicant number are control variables. Descriptive statistical results include the standard deviation, statistics, mean, maximum, and minimum values of each variable, reflecting the stability of the data. The correlation results show that there is no serious multicollinearity between variables.

Due to the existence of highly correlated relationships between explanatory variables, the model estimation may be distorted or difficult to estimate accurately. Therefore, collinearity analysis should be performed on the explanatory variables. As shown in

Table 6. The result of multicollinearity analysis.

Variable	VIF
Licensor prestige	1.19
Licensor technological depth	1.19
Licensor technological breadth	2.19
Patent technological advancement	1.28
Patent technological stability	1.98
Patent protection scope	1.93
Licensor research age	3.96
Licensor ownership	1.17
Patent disclosure duration	3.42
Patent applicant number	1.06
Mean VIF	1.94

, there is no multicollinearity between variables. The variance inflation factor (VIF) can be used to diagnose multicollinearity between variables. The VIF value of each variable and the average VIF value in the results are both less than 10, indicating that they do not have collinearity.

As shown in

Table 7. The result of the outlier test.

Variable	Mean	Std. Dev.	Min	Max
Cook	0.0012	0.0084	8.13 × 10−12	0.2461

, the Cook distance is used to determine whether a strong influence point is an outlier point of the explanatory variable. It is believed that when D < 0.5, it is not an outlier point, but when D > 0.5, it is an outlier point. The result shows that the maximum Cook distance is 0.2461 < 0.5, indicating the absence of significant outliers.

Table 8. The results of conditional logistic regression.

Variables	Patent Licensing
	Model 1	Model 2	Model 3	Model 4
Licensor prestige	−0.1568 **	−0.1203 **	−0.0961 **	−0.3003 ***
	(−0.0659)	(−0.0496)	(−0.0406)	(−0.1095)
Licensor technological depth	−0.6716 ***	−0.5918 ***	−0.4432 ***	−0.415 **
	(−0.2547)	(−0.1921)	(−0.1513)	(−0.2016)
Licensor technological breadth	0.1047 **	0.028	−0.0064	0.1064 *
	(−0.0517)	(−0.0354)	(−0.0244)	(−0.056)
Patent technological advancement	3.3744 ***	3.1701 ***	2.8818 ***	3.6838 ***
	(−0.6514)	(−0.4634)	(−0.3604)	(−0.7852)
Patent technological stability	0.5081 ***	0.4117 ***	0.3309 ***	0.6103 ***
	(−0.1444)	(−0.1018)	(−0.0804)	(−0.1659)
Patent protection scope	−0.2173	−0.0956	−0.0982	−0.4609 **
	(−0.1498)	(−0.1124)	(−0.1012)	(−0.1997)
Licensor research age	−0.0783	0.0098	0.0603	−0.1289
	(−0.0944)	(−0.0789)	(−0.0635)	(−0.1091)
Licensor ownership	−0.1431	−0.436 **	−0.4048 ***	−0.2534
	(−0.2336)	(−0.1845)	(−0.1459)	(−0.2744)
Patent disclosure duration	0.2795 **	0.1041	0.0502	0.2351 *
	(−0.1313)	(−0.0883)	(−0.0732)	(−0.1295)
Patent applicant number	0.622	0.4292	−0.1661	0.111
	(−0.4544)	(−0.4058)	(−0.412)	(−0.7935)
Observations	710	1420	2130	530
Log likelihood	−36.80159487	−58.67131148	−87.96841871	−26.024
LR chi2 (10)	252.5067 ***	307.298 ***	306.1014 ***	191.6736 ***
Pseudo R2	0.7743	0.7237	0.635	0.7864

* p < 0.1, ** p < 0.05, *** p < 0.01.

shows the results of conditional logistic regression, using licensed patents and their similar patents in the field of intelligent connected vehicles. Among them, Model 1, Model 2, and Model 3 use data from 1 January 2001 to 31 December 2021. Similar patents were calculated using the topic modeling method in Section 3.2. The regression of Model 1 uses data from 71 licensed patents and the top 10 similar patents for each licensed patent, while Model 2 and Model 3 use data from licensed patents and their top 20 and 30 similar patents, respectively. The second column shows the regression results of Model 1. It can be seen that technological prestige, technological depth, technological breadth, technological advancement, and technological stability have a significant impact on patent licensing. Technological prestige and technological depth have significantly negative correlations with patent licensing, while technological breadth, technological advancement, and technological stability are significantly positively related to patent licensing. Surprisingly, the regression results show a negative correlation between patent protection scope and patent licensing, which is not significant. This result may stem from several factors. First, enterprises tend to retain patents with broad protection as strategic assets to maintain a competitive edge, reducing the incentive to license. Second, a wide-scope patent can impose higher implementation costs and risks on licensees, making them more hesitant. Finally, in the fast-evolving field of intelligent connected vehicles, patents with broad protection may face more technological uncertainties, leading patent holders to avoid licensing. The third and fourth columns in this table represent the conditional logistic regression results of Model 2 and Model 3, respectively. In comparison, the experimental results are not significantly different from the results of Model 1, proving the robustness of the model proposed in this paper. To demonstrate that the experimental data have little impact on the experimental results, 53 licensed patents from 1 January 2001 to 31 December 2021, along with their top 20 similar patents, are used in Model 4. The significance of its explanatory variables is similar to the results of Model 1. The influence of technological depth, technological breadth, and technological advancement on patent licensing is relatively stable, while that of patent protection scope on patent licensing is gradually significant.

Furthermore, four typical enterprises—Beijing Baidu, Shenzhen BYD, Shenzhen Tencent, and Wuhan Cabit—were selected for qualitative analysis to analyze their patent layout for intelligent connected vehicles. Their patent licensing information is shown in

Table 9. The licensing information of Baidu, BYD, Tencent, and Cabit.

Enterprise	Tencent	BYD	Baidu	Cabit
NoP	7	26	12	2
NoLP	2	0	0	2
NoNLP	5	26	12	0
NoC	13	70	53	3

. Among them, NoP is the total number of patents, NoLP is the number of licensed patents, NoNLP is the number of unlicensed patents, and NoC is the number of citations. Baidu, BYD, and Tencent are all large-scale enterprises. Baidu is an AI company with a strong Internet foundation. Its business scope is not limited to intelligent logistics, smart homes, automatic driving, search engines, map navigation, etc. Baidu has a high influence and a deep foundation, with a large proportion of self-developed patented technologies and a high level of autonomy and controllability. Therefore, it has 12 patents in the ICV field, all of which are unlicensed, and they have been cited 53 times. BYD is a high-tech enterprise mainly engaged in automotive manufacturing, new-energy batteries, and other businesses, committed to exploring key technologies in the automotive field. It tends to form a patent layout based on key core technologies, using intellectual property to protect innovation, and maintaining an edge in industry competition. Tencent not only operates social platforms, online games, fintech, cloud computing, and other businesses but also has applications in areas such as cybersecurity, digital content, and intelligent hardware. Due to its extensive research field and lack of focus on the ICV field, a small portion of its patents have been licensed. Cabit is a technology company that focuses on the R&D of automotive interconnection technology and application products. Its scale is relatively small, but it has multiple core patented technologies, so it licenses out all patents to cooperate with other companies and develop together.

Baidu: As a large enterprise with artificial intelligence at its core and a strong Internet foundation, Baidu’s business widely covers multiple fields, such as intelligent logistics, autonomous driving, and search engines. This diversified business model determines that it requires a high degree of autonomy and control in technological research and development to ensure its competitive advantage in various business segments. Therefore, Baidu’s patents in the field of intelligent connected vehicles are all unlicensed patents. By independently researching and developing core technologies, it builds technological barriers. The high citation count also indicates that its patented technologies not only support the development of its own business but also serve as an important cornerstone for the technological development of the industry, which is in line with its business model positioning it as a technology exporter and industry leader.

BYD: Focusing on automotive manufacturing and new-energy battery businesses, BYD’s core business model lies in creating differentiated products through technological innovation and occupying a dominant position in the automotive market. In the patent layout of intelligent connected vehicles, BYD forms a patent portfolio around key fields such as automotive manufacturing and energy technologies, using intellectual property to protect its innovation achievements and enhance the technological content and added value of its products. This patent licensing strategy is consistent with the business model of deeply cultivating the automotive industry and pursuing technological leadership, helping to consolidate its competitive position in the automotive industry.

Tencent: Its business covers multiple fields, such as social platforms, online games, and fintech, and intelligent connected vehicles are not its core business direction. Its patent licensing strategy in this field is relatively flexible, with only a small number of patents being licensed. This is because Tencent’s investment and focus on intelligent connected vehicles are relatively low. Through selective patent licensing, Tencent can maintain its technological participation in this field while avoiding excessive resource investment, allowing it to focus more on its core business. This strategy is in line with Tencent’s diversified and flexible business model, aiming to balance the development needs of different businesses.

Cabit: As a technology company focusing on the research and development of automotive interconnection technologies, Cabit has a relatively small scale but possesses multiple core patented technologies. Based on its own resources and capability characteristics, Cabit adopts a strategy of licensing out all of its patents and collaborating with other enterprises. This model can fully leverage its technological advantages, and by relying on the resources and market channels of its partners, it can achieve the commercial application and promotion of its technologies and make up for its own limitations in scale and resources. This strategy matches its business model as a technology provider that achieves development through cooperation.

5. Discussion

In this section, the practical implications for enterprises, universities, and the government are discussed, and this study further examines how to optimize the patent licensing framework in the ICV sector to promote green innovation.

For enterprises, those with different levels of technological reputation should adopt development strategies that align with their technological capabilities and resource endowments. Enterprises with high technological reputation possess greater resources to diversify their technological portfolios, explore emerging innovations, and enhance the commercialization of technological achievements. In contrast, enterprises with lower technological reputation should prioritize the accumulation and continuous improvement of technical knowledge, leveraging patent licensing as a strategic tool for technology transfer and commercialization while strengthening their market position.

For universities, increasing investment in patent agency services and improving the standardization and innovation of patent documentation can significantly enhance patent quality and authorization rates. Referring to similar licensed patents during the patent drafting process may improve retrievability and licensing potential, facilitating the efficient transformation of research outcomes into real-world applications.

For the government, promoting the transformation of scientific and technological achievements requires not only refining laws, regulations, and policy incentives but also streamlining the patent licensing process, optimizing patent transactions, enhancing market resource allocation, and strengthening intellectual property protection. A well-functioning patent licensing framework can foster a more dynamic and innovation-driven environment, accelerating technology diffusion and sustainable industrial development.

Building on the preceding analysis, the empirical patterns observed across patent characteristics and firm strategies point to dynamics that are central to sustainable industrial transformation in the ICV sector. The greater licensing likelihood associated with patents exhibiting technological breadth, advancement, and stability reflects a preference for scalable, implementation-ready innovations that can be widely adopted across subsystems and applications. These traits are often aligned with environmentally responsible technologies—such as energy-efficient components and intelligent traffic systems—that support carbon reduction goals.

This preference suggests that green technology diffusion may be facilitated less by core or prestige innovations, which are often retained for competitive advantage, and more by integrable, versatile solutions. As a result, licensing becomes a key mechanism not only for spreading innovation but also for enabling environmental impact at scale.

Such diffusion pathways also reinforce long-term economic resilience. By reducing redundancy and enabling technological reuse, patent licensing contributes to efficient resource utilization and fosters collaborative innovation. The active engagement of SMEs and universities—often through incremental or complementary technologies—supports inclusive growth and accelerates the sustainability transition. The case analysis in this study illustrates how differing firm strategies (e.g., Baidu’s retention vs. Cabit’s full licensing) shape the distribution of innovation capacity across the industry.

This study provides actionable insights into improving the patent licensing framework in the intelligent connected vehicle (ICV) sector, especially in the context of promoting green innovation. The empirical results from conditional logistic regression (Table 8) indicate that certain patent-level technological characteristics—such as technological breadth, advancement, and stability—significantly enhance the likelihood of licensing. In contrast, patents with high technological prestige or depth are less likely to be licensed, suggesting that highly strategic or core technologies may be deliberately withheld. These findings imply that policy instruments should encourage the licensing of patents that are versatile, mature, and innovation-enabling, such as through differentiated tax incentives or green technology licensing schemes.

Interestingly, the scope of patent protection does not positively contribute to licensing and may even hinder it. This finding suggests that overly broad patent claims might increase the implementation cost and legal ambiguity for licensees, especially in rapidly evolving sectors like ICVs. This highlights the need for a balanced approach between intellectual property protection and technology diffusion, particularly for sustainability-related innovations. Regulatory bodies should reconsider how protection scope is assessed and whether targeted adjustments can reduce barriers to green technology access.

Although the variable of topic similarity between patents was not included as an independent factor in the regression model, it played a key role in constructing matched comparisons between licensed and unlicensed patents. This approach enabled a more meaningful analysis of licensing decisions by controlling for technical content. While not a direct determinant in this study, the concept of topical relevance has important policy implications: It supports the idea of building AI-powered patent recommendation or matching platforms that can help firms identify potential partners based on technological alignment. Such tools can facilitate more efficient knowledge transfer and reduce informational barriers in the licensing process, especially for emerging green technologies.

Finally, the case analysis of Baidu, BYD, Tencent, and Cabit illustrates heterogeneity in firm-level licensing behavior. Large firms with strong R&D capacity tend to retain core technologies, while smaller firms are more willing to license their patents to collaborate and grow. This suggests that differentiated licensing policies—such as public–private licensing consortia or green patent pools—may be necessary to foster inclusive innovation and enable the broader adoption of clean and intelligent technologies in the ICV sector.

This study contributes to the theoretical understanding of patent licensing by integrating patent-level technological characteristics and firm-level strategic behavior in a sector undergoing rapid transformation. First, the finding that patents with high technological prestige and depth are less likely to be licensed challenges the conventional assumption that the most advanced or valuable technologies are the most transferable. This supports theoretical perspectives on knowledge hoarding and strategic patenting, where firms may use intellectual property not primarily for diffusion but for competitive exclusion.

Second, the consistent influence of attributes like technological breadth and advancement reinforces the importance of transferability and absorptive capacity in licensing theory. It aligns with the idea that technologies with broader application potential are more likely to diffuse through markets and ecosystems. Third, although topic similarity was not modeled directly, its use in constructing matched samples suggests that cognitive or thematic proximity plays a foundational role in how firms assess licensing opportunities—pointing to a new, underexplored dimension of patent transaction theory.

Finally, by linking empirical licensing patterns to firm-specific strategies and business models, this study advances a contextualized view of technology transfer, where licensing decisions are not purely market-driven but are shaped by organizational goals, innovation trajectories, and sectoral sustainability pressures. These insights enrich existing theoretical frameworks and open pathways for future research on green innovation, sector-specific licensing behavior, and data-informed IP policy design.

6. Conclusions

This study explores the comprehensive factors influencing patent licensing at both the enterprise and patent levels within the intelligent connected vehicle (ICV) industry, a sector that plays a critical role in the transition toward sustainable mobility and smart transportation systems. The research methodology involved selecting ICV-related patent abstracts, conducting topic modeling analysis on licensed patents, calculating the similarity between unlicensed and licensed patents based on topic distribution, and applying a conditional logistic regression model to identify key determinants of patent licensing trends. The findings validate the effectiveness and robustness of the proposed approach in assessing technology transfer and innovation diffusion within the ICV sector, contributing to the development of sustainable and knowledge-driven industrial ecosystems.

However, it is important to recognize that the proposed method has been validated only within the ICV domain. Given that patent licensing dynamics vary across industries and technological landscapes, the findings may not be directly generalizable to other fields. Future research should explore comparative studies across multiple industries, particularly in sectors contributing to green innovation, renewable energy, and climate-resilient technologies, to assess the broader applicability and adaptability of this method in supporting sustainability-driven technology transfer.

Conducting cross-sectoral analyses would provide valuable insights into how industry-specific patent characteristics influence licensing decisions, allowing for the refinement of sustainable technology commercialization strategies. By identifying variations in patent licensing determinants across different technological fields, researchers can enhance the understanding of sector-specific innovation diffusion mechanisms and contribute to the creation of more effective policies that facilitate sustainable industrial development.

However, this study has certain limitations. First, the sample size was constrained by the availability of publicly disclosed licensed patent data, which may not fully capture all licensing activities in the sector. Second, the analysis relied on secondary data sources, including patent databases and citation records, which may introduce limitations in data accuracy or completeness. These constraints could affect the generalizability of the findings. Future research could expand the dataset, incorporate primary data (e.g., surveys or interviews), and conduct cross-sectoral comparisons to validate and extend the results.

In addition to expanding the data sources and conducting cross-sectoral comparisons, future research could also explore the effects of different patent licensing models—such as exclusive versus non-exclusive licensing, cross-licensing, and open licensing—on technology transfer efficiency and innovation outcomes. Investigating how these licensing strategies interact with firm size, technological maturity, and market conditions in the ICV sector would provide deeper insights into optimizing IP strategies for sustainable industrial development.

Patent licensing serves as a key mechanism for sustainable knowledge transfer and technological dissemination, fostering the development of low-carbon industries and environmentally responsible innovation. The proposed method can also be applied to analyze other forms of knowledge transfer beyond patent licensing, offering broader implications for understanding technology diffusion, open innovation, and green industrial transformation. By strengthening intellectual property strategies that support sustainability, this research contributes to enhancing collaborative innovation, promoting circular economy practices, and accelerating the adoption of clean and intelligent technologies.