A Two-Layer Deep Learning Approach for R&D Partner

Recommendation in the Self-Driving Vehicle Industry

Juite Wang and Ying-Pei Kao

Graduate Institute of Technology Management, National Chung Hsing University, 145 Xingda Rd, Taichung,

Taiwan, Republic of China

Keywords: Self-Driving Vehicles, Partner Recommendation, Patent Analytics, Social Network Analysis,

Deep Neural Networks, Graph Neural Networks.

Abstract: This study presents a two-stage deep learning framework for recommending strategic R&D partners in the

self-driving vehicle (SDV) industry. Leveraging 165,775 U.S. patent applications from 2015 to 2023, we

constructed a co-patent network and extracted node, edge, and topological features to represent organizational

attributes, collaboration intensity, and network structure. These features were integrated using a hybrid Graph

Neural Network (GNN) and Deep Neural Network (DNN) architecture to predict future collaborations. The

model achieved high predictive performance (accuracy = 96.65%, precision = 70.83%, recall = 66.92%, F1 =

68.82%, and AUPRC = 78.93%) and demonstrated its ability to identify both established and emerging

partners. Community detection revealed influential clusters anchored by firms like Toyota and Hyundai. Case

analyses showed that the model can recommend both historical and emerging R&D partners. Compared to

prior work, this study contributes a scalable, data-driven approach that incorporates deep structural and

semantic signals to improve partner selection accuracy. The framework advances patent analytics by linking

network-based learning with partner recommendations, offering practical implications for R&D planning in

complex technology-based industries.

1 INTRODUCTION

The automotive industry is undergoing a profound

transformation driven by self-driving vehicle (SDV)

technologies. These advancements build on artificial

intelligence (AI), big data, 5G, Internet of Things

(IoT), and computer vision. Leading automakers and

technology firms alike are investing in SDV

development, while also seeking external partners to

accelerate innovation. Strategic R&D collaboration

has thus become a critical issue in the competitive

landscape.

The complexity of SDV innovation requires

coordination across hardware, software, and

infrastructure layers. Companies must integrate

capabilities in sensing technologies, real-time

decision-making algorithms, high-definition

mapping, and connectivity. This interdisciplinary

nature not only amplifies technical challenges but

also elevates the strategic importance of effective

partner selection. As competition intensifies, firms

that can identify and secure partnerships with

technically aligned collaborators will gain substantial

advantages in innovation speed and technological

scope.

Despite this urgency, systematic methods for

identifying suitable R&D partners remain limited.

Traditional approaches often rely on prior

relationships, managerial intuition, or firm-level

reputational signals. While helpful, these heuristics

overlook valuable insights embedded in structured

data sources such as patent filings. Patent data reveal

not only technological strengths but also patterns of

past cooperation, citation influence, and topical

complementarities. By transforming patent co-

application information into a relational network, one

can better understand the evolving landscape of

collaboration and anticipate future alliance

opportunities.

SDV development requires integration of diverse

technical domains. No single firm can master all areas,

making partner selection crucial for acquiring

complementary capabilities and accelerating time-to-

market. While prior collaborations provide valuable

signals, identifying future partners based on technical

fit and network dynamics remains a challenging

Wang, J. and Kao, Y.-P.

A Two-Layer Deep Learning Approach for R&D Partner Recommendation in the Self-Driving Vehicle Industry.

DOI: 10.5220/0013683100004000

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2025) - Volume 1: KDIR, pages 60-66

problem. This challenge is particularly salient in the

SDV context, where the rapid pace of change demands

a forward-looking approach to alliance formation.

Patent data offer a valuable lens for analyzing

R&D collaboration. They reflect firms' technological

focus and inventive activity and, when co-assigned,

signal formalized joint efforts. Studies on R&D

partner selection using patent data can be broadly

categorized into three methodological approaches:

bibliometric analysis (Geum et al., 2013; Wang,

2012), link prediction (Yan & Guns, 2014), and

textual analysis (Kim et al., 2020; Wang et al., 2017).

Yet, significant research gaps remain. First, most

prior studies on partner selection have relied on static

indicators or network metrics. These methods fail to

capture the complex, non-linear interactions among

structural features and R&D collaboration potential.

Second, few studies integrate deep learning with

patent analytics, despite growing evidence that neural

networks can improve pattern recognition in

structured and semi-structured data.

This study aims to develop a two-stage deep

learning model to predict potential R&D partnerships

in the self-driving vehicle industry. Patent data from

the United States Patent and Trademark Office

(USPTO) were collected. Co-applicant patents were

used to build an R&D collaboration network.

Bibliometric indicators, link prediction measures, and

social network analysis were used to generate

features. A two-stage deep learning model,

comprising a GNN and a DNN, was trained to predict

collaboration links.

According to the obtained results, the model

achieved acceptable predictive performance (F1 score

= 68.82, AUPRC = 78.93) by leveraging node, edge,

and topological features extracted from a co-patent

network. Network analysis revealed influential

clusters centered on firms like Toyota and Hyundai,

and the partner recommendation system successfully

identified both established and emerging

collaborators, offering practical insights for strategic

alliance planning.

This study contributes to the literature by (1)

introducing a deep learning-based framework that

unifies node, edge, and topological features; (2)

applying GNN and DNN models to large-scale patent

co-application data; and (3) demonstrating how the

system can generate interpretable partner

recommendations grounded in empirical

collaboration signals.

Section 2 reviews the literature on SDVs and

partner selection. Section 3 outlines the data and

methodological framework. Section 4 presents the

empirical results. Section 5 concludes with future

directions.

2 LITERATURE REVIEW

2.1 Self-Driving Vehicles

SDVs, also referred to as autonomous vehicles,

represent a convergence of several advanced

technologies, including machine learning, sensor

fusion, computer vision, and real-time decision-

making algorithms (Gwak et al., 2019). Research on

SDVs has examined their potential to reduce

accidents, improve traffic flow, and enhance mobility

for underserved populations (Ryan, 2020).

Technological classifications based on SAE levels

have been widely adopted, and attention has

increasingly turned toward regulatory, infrastructural,

and ethical considerations.

In addition to engineering and legal aspects,

researchers have explored the business and ecosystem

implications of SDVs. These include the emergence

of mobility-as-a-service (MaaS), the integration of

SDVs into smart cities, and the disruption of

incumbent value chains (Vosooghi et al., 2019).

However, many of these studies treat technological

development as a monolithic process, overlooking the

collaborative innovation dynamics that drive SDV

advancements.

2.2 Partner Selection and Patent

Analytics

In innovation studies, partner selection has long been

recognized as a determinant of alliance success (Shah

& Swaminathan, 2008). Firms enter R&D

collaborations to access complementary capabilities,

share costs, and reduce uncertainties in technological

exploration (Makri et al., 2010). Effective partner

selection thus requires evaluating not only

organizational fit but also technical and relational

compatibility.

Studies on R&D partner selection using patent

data can be broadly categorized into three

methodological approaches: bibliometric analysis,

link prediction, and textual analysis. Bibliometric

methods use structured metadata such as citations,

classifications, and assignee information to assess

firms’ technological strength and collaborative

potential (Geum et al., 2013; Wang, 2012), though

they are limited by citation delays and outdated

classification systems (Kim et al., 2020). Link

prediction approaches (Yan & Guns, 2014) leverage

A Two-Layer Deep Learning Approach for R&D Partner Recommendation in the Self-Driving Vehicle Industry

co-patenting or co-authorship networks to forecast

partnerships based on network topology, but they

often overlook non-structural factors that influence

collaboration decisions (Song et al., 2016). Textual

analysis applies NLP techniques to extract semantic

insights from patent texts, using methods like Subject-

Action-Object (SAO) structures or embedding models

to identify technological similarity (Kim et al., 2020;

Wang et al., 2017), though earlier approaches lacked

deep contextual understanding.

To address the limitations of single-method

approaches, a growing number of studies have adopted

hybrid models that combine multiple analytical

perspectives. For instance, Yoon and Song (2014)

integrated text mining and morphological analysis with

bibliometric indicators to map technological

opportunities and assess partner suitability. Park et al.

(2015) employed latent semantic analysis (LSA) in

conjunction with bibliographic coupling to evaluate

technological proximity among firms. Song et al.

(2016) extracted action-object structures from patent

texts to assess technological needs and then applied

three patent-based indicators to evaluate candidate

partners. Jee and Sohn (2020) used latent Dirichlet

allocation (LDA) topic modeling to identify firms

working in complementary areas, followed by

bibliometric analysis to evaluate technological strength

and relatedness. Qi et al. (2022) applied LDA to both

scientific publications and patent documents to extract

topic-level similarities and then employed link

prediction methods to recommend partners based on

innovation capacity and openness. More recently,

Wang et al. (2025) constructed three interconnected

network layers and introduced three metrics,

technological similarity (patent citation metric),

technological heterogeneity (patent indicators), and

collaboration status (based on link prediction

indicators), to recommend potential R&D partners.

Although these hybrid approaches represent a

meaningful advance over single-method models,

important limitations remain. Most studies do not

fully integrate all three core dimensions, textual

analytics, bibliometric analysis, and network-based

link prediction, within a unified analytical framework.

Moreover, few adopt advanced machine learning

methods capable of modeling complex, non-linear

relationships among diverse features. This

methodological fragmentation restricts the capacity

of existing models to reflect the multifaceted and

dynamic nature of R&D collaboration.

3 DATA AND METHODOLOGY

3.1 Data Collection and Preprocessing

This study collected 197,203 patent applications

related to self-driving vehicles from the United States

Patent and Trademark Office (USPTO) database,

spanning from 2001 to October 2023. To align with

the goal of recommending organizational R&D

partners, all patents filed by individual applicants

were excluded. The final dataset consisted of 165,775

organizational patents. As shown in Figure 1, the

number of SDV-related patent applications has

steadily increased each year, suggesting a growing

market and technological momentum in this domain.

The decline in 2023 filings reflects a lag due to the

required publication and examination period for

patent applications.

Figure 2 further shows that co-patent applications,

those involving collaborations between organizations,

have also grown since 2013. This trend highlights a

rising demand for inter-organizational cooperation in

the SDV industry. Notably, 2015 marks a critical

milestone, with Tesla launching its third-generation

electric vehicle (Model X), signaling the commercial

viability of autonomous driving technologies.

Accordingly, this study focuses on the period from

2015 to 2023, splitting the data into two segments:

Period 1 (2015–2020) and Period 2 (2021–2023), to

support time-aware modeling for potential R&D

partner recommendation.

Figure 1: Yearly trend of patent applications.

Figure 2: Yearly trend of collaborative patents.

KDIR 2025 - 17th International Conference on Knowledge Discovery and Information Retrieval

3.2 Methodology

The methodological framework of this study,

illustrated in Figure 3. From the applicant information

in patent records, a collaboration network was

constructed to represent co-application relationships

between organizations. Various indicators were

computed and categorized into three feature groups:

node features, edge features, and topological features.

These served as inputs for a deep learning model

based on Park and Geum (2022), which integrates a

Graph Neural Network (GNN) and a Deep Neural

Network (DNN). In this framework, node features are

processed by the GNN to generate structural

embeddings, which are combined with edge and

topological features in the DNN for link prediction.

This two-stage approach supports accurate prediction

and practical recommendation of R&D partners in the

SDV domain.

Figure 3: Research framework.

3.2.1 Constructing the Collaboration

Network

To construct the collaboration network, each

undirected edge represented at least one jointly filed

patent between two organizations. The adjacency

matrix was defined such that if firms and co-applied

for at least one patent between 2015 and 2020, and

otherwise, where edge weights reflected the number

of co-patents. This network served as the foundation

for feature generation and potential collaboration

prediction tasks.

3.2.2 Defining Features for the DNN Model

Node features (organizational-level) are categorized

into three categories.

Technological strength (Geum et al., 2013) includes:

 Patent quantity: number of application and

granted patents;

 Patent quality: forward citations of application

and granted patents;

 R&D resources: number of inventors and

granted-to-application ratio;

 Technological presence: share of total patents

and simple patent family ratio;

 Technological impact: proportion of highly

cited (top 10%) patents.

R&D openness (Geum et al., 2013) captures:

 Organizational openness: share of co-invented

patents;

 Inventor and assignee openness: average

number of nationalities per patent.

Other variables developed in this study include:

 Organization nationality (via one-hot

encoding);

 Technological fields: A vector indicating the

number of patents held by the organization in

each CPC subclass.

Edge features (dyadic-level) describe collaborative

strength and synergy and consist of two categories.

R&D Linkage (Geum et al., 2013) includes:

 Joint R&D capability: Share of patents co-

invented with partners out of the organization’s

total co-invented patents.

 Joint application intensity: Share of patents

jointly filed with partners out of all co-invented

patents across organizations.

 Joint invention intensity: Share of patents

jointly invented with partners out of all co-

invented patents across organizations.

Collaborative Synergy (Geum et al., 2013) includes:

 Knowledge inflow: Ratio of the average

backward citations in partner co-invented

patents to the organization’s overall average.

 Knowledge criticality: Ratio of the average

forward citations in partner co-invented patents

to the organization’s overall average.

 Knowledge similarity: Sum of shared citations

between the organization and its partner,

normalized by the number of application

patents held by both.

Topological features (network-level) were based on

several link prediction indicators (Mutlu et al., 2020)

and network centrality measures (Hansen et al.,

2011):

 Link similarity: common neighbors, shortest

paths, Jaccard index, Salton index, Sorensen

index, preferential attachment, resource

allocation index, Adamic-Adar index, Katz

index.

 Node centrality: degree, closeness, and

betweenness centrality.

1. Construct an R&D

collaboration network

2. Define features for

R&D partner

recommendation

3. Construct a two-

stage DNN model

4. Partner

recommendation

A Two-Layer Deep Learning Approach for R&D Partner Recommendation in the Self-Driving Vehicle Industry

3.2.3 Building a Two-Stage DNN Model

The two-stage deep learning model (Figure 4)

consisted of:

 Stage 1: GNN embedding: GraphSAGE

(Hamilton et al., 2017) aggregated each node's

features and neighborhood context into vector

embeddings that preserved local topology and

semantic content.

 Stage 2: DNN classifier: The embeddings from

the GNN were concatenated with edge and

topological features and input to a DNN. The

DNN used two hidden layers with ReLU

activation, dropout, and focal loss to address

class imbalance. Training was conducted on the

2015–2020 data, and predictions were

validated against 2021–2023 collaborations.

Figure 4: Two-stage DNN model.

Negative samples were created by randomly

sampling unlinked firm pairs to balance the dataset

for binary classification.

3.2.4 Partner Recommendation

The final output was an R&D partner

recommendation list for each focal organization.

Candidate firms were ranked based on their predicted

link probabilities. The top-N ranked entities

represented high-potential partners with strong

structural and technical fit. These results offer

practical guidance for alliance planning and partner

identification in the self-driving vehicle innovation

ecosystem.

4 RESULTS

4.1 Two-Stage Deep Learning

Performance

The model was trained on 28,203 organization pairs

constructed from 238 firms that collaborated in both

periods. Data were split 70:30 into training and test

sets. Using PyTorch, the two-stage deep learning

framework was implemented: GNN embeddings

were generated from node features, and then

combined with edge and topological features to train

the DNN. Hyperparameters were optimized using

Optuna. The best configuration included a learning

rate of 0.0089, three hidden layers in both GNN and

DNN, Leaky ReLU for GNN activation, ReLU for

DNN, and a MultiStepLR scheduler.

Model performance on the test set showed high

accuracy (96.65%), precision (70.83%), recall

(66.92%), and F1 score (68.82%). The area under the

precision-recall curve (AUPRC) was 78.93,

indicating strong performance across varying

thresholds.

To assess the contribution of each feature set,

separate models were trained using node, edge, or

topological features alone. Results showed that while

all single-feature models achieved reasonable

accuracy, the integrated model outperformed them

across all metrics. Notably, the GNN model yielded

an F1 score of 58.87% and AUPRC of 72.70%, while

edge and topological features were slightly lower.

Combining all three feature sets enhanced precision

and robustness.

4.2 Collaboration Network and Cluster

Analysis

Community detection on the collaboration network

revealed three dominant clusters:

 Cluster 1: Centered on Toyota, included 160

organizations, where Toyota, Honda, and

Denso emerged as key nodes across centrality

metrics.

 Cluster 2: Led by Hyundai, comprising 101

entities including Kia, LG, and Samsung, with

strong interconnectivity among Korean firms.

 Cluster 3: Represented by Volkswagen,

BMW, and Valeo, focusing on European and

U.S. collaborations.

Centrality measures showed Toyota and Hyundai

held the highest degree, betweenness, and

eigenvector centralities within their clusters,

confirming their influence.

4.3 Predicted R&D Collaboration

Network and Partner

Recommendation

Based on the trained two-stage model, this study

predicted potential R&D collaborations for the 2024–

2026 period using patent data from 2021–2023. A

total of 590 likely collaboration pairs were identified,

exceeding the 572 observed links in the prior

Node features

Edge features

Topological

features

GNN model

DNN model

KDIR 2025 - 17th International Conference on Knowledge Discovery and Information Retrieval

network. Community detection and centrality

analysis were applied to the predicted network,

revealing new clusters and high-potential

organizations.

 Cluster 1: Toyota remained central with the

highest degree, betweenness, and eigenvector

centralities, followed by Honda, Denso, Jtekt,

and Stanford University.

 Cluster 2: Hyundai led the second-largest

cluster, with Kia, Samsung, and Korea

Advanced Institute of Science and Technology

also prominently involved.

 Cluster 3: Emerging leaders included

Increment P Corp and Pioneer Corp, along with

research institutions such as Fraunhofer and

Northeastern University.

Visualization of the predicted collaboration network

(Figure 5) illustrated the structural role of core firms

and highlighted new connections. Node size reflected

the number of predicted links, while colors indicated

cluster memberships.

Figure 5: Predicted R&D collaboration network.

For example, the Toyota case revealed 35 predicted

partners—most of which were previous collaborators

such as Denso and Subaru, suggesting ongoing

cooperation. However, new recommendations like

Tokyo Cosmos Electric and Pioneer Corp were also

identified based on CPC complementarity. Tokyo

Cosmos Electric, with strengths in B60R and H05B,

shares only one CPC class with Toyota but aligns

with prior Toyota partners like Kojima Ind Corp,

suggesting technical complementarity.

Recommended partners were categorized into:

 Existing Partners: reinforcing current

alliances through shared CPC domains (e.g.,

Denso, Subaru).

 New Partners: identified based on

complementary CPC profiles and inferred

similarity to Toyota's prior partners (e.g.,

Tokyo Cosmos Electric, Univ Milano

Politecnico).

These results demonstrate the model’s capacity to

uncover both strategic extensions of existing

partnerships and novel opportunities based on

technology fit.

5 CONCLUSIONS

This study developed a DNN framework to predict

potential R&D partnerships in the self-driving vehicle

industry based on patent data. Bibliometric

indicators, link prediction measures, and social

network analysis were used to generate features. A

two-stage deep learning model, comprising a GNN

and a DNN, was trained to predict collaboration links.

The model achieved acceptable predictive

performance by leveraging node, edge, and

topological features extracted from a co-patent

network. The obtained results revealed influential

clusters centered on firms like Toyota and Hyundai,

and the partner recommendation system successfully

identified both established and emerging

collaborators, offering practical insights for strategic

alliance planning.

The limitations of this study are discussed as

follows. First, this study analyzed only SDV related

patent data in the USPTO database. Therefore, it is

not cover patents related to SDV in other countries.

For example, BYD Auto, one of the world's largest

electric vehicle manufacturers in China, has files

most of their patents in China. There are only 15

patents filed in US according to the collected patent

data in this study. Second, this study is restricted to

uncover R&D collaborations using patent data. It may

not be able to find R&D collaboration in other forms,

such as strategic alliance, new venture, and licensing

(Schilling, 2012). Future research can include broader

data sources that can yield a more comprehensive

view of R&D collaboration in industries.

ACKNOWLEDGEMENTS

This research is supported by Grant No. NSTC 113-

2221-E-005-076 -MY2 from the National Science

and Technology Council, R.O.C.

A Two-Layer Deep Learning Approach for R&D Partner Recommendation in the Self-Driving Vehicle Industry

REFERENCES

Geum, Y., Lee, S., Yoon, B., & Park, Y. (2013). Identifying

and evaluating strategic partners for collaborative

R&D: Index-based approach using patents and

publications. Technovation, 33(6-7), 211-224.

Gwak, J., Jung, J., Oh, R., Park, M., Rakhimov, M. A. K.,

& Ahn, J. (2019). A Review of Intelligent Self-Driving

Vehicle Software Research [Review]. Ksii

Transactions on Internet and Information Systems,

13(11), 5299-5320. https://doi.org/10.3837/tiis.2019.

11.002

Hamilton, W. L., Ying, R., & Leskovec, J. (2017).

Inductive Representation Learning on Large Graphs.

arXiv:1706.02216. Retrieved June 01, 2017, from

https://ui.adsabs.harvard.edu/abs/2017arXiv17060221

Hansen, D. L., Shneiderman, B., & Smith, M. A. (2011).

Chapter 5 - Calculating and Visualizing Network

Metrics. In D. L. Hansen, B. Shneiderman, & M. A.

Smith (Eds.), Analyzing Social Media Networks with

NodeXL (pp. 69-78). Morgan Kaufmann.

https://doi.org/https://doi.org/10.1016/B978-0-12-3822

29-1.00005-9

Jee, S. J., & Sohn, S. Y. (2020). Patent-based framework

for assisting entrepreneurial firms' R&D partner

selection: Leveraging their limited resources and

managing the tension between learning and protection.

Journal Of Engineering And Technology Management,

57, 101575, Article 101575. https://doi.org/10.1016/j.

jengtecman.2020.101575

Kim, H. J., Kim, T. S., & Sohn, S. Y. (2020).

Recommendation of startups as technology cooperation

candidates from the perspectives of similarity and

potential: A deep learning approach. Decision Support

Systems, 130, 113229.

Makri, M., Hitt, M. A., & Lane, P. J. (2010).

Complementary technologies, knowledge relatedness,

and invention outcomes in high technology mergers and

acquisitions. Strategic Management Journal, 31(6),

602-628. https://doi.org/https://doi.org/10.1002/smj.

829

Mutlu, E. C., Oghaz, T., Rajabi, A., & Garibay, I. (2020).

Review on Learning and Extracting Graph Features for

Link Prediction [Review]. Machine Learning and

Knowledge Extraction, 2(4), 672-704.

https://doi.org/10.3390/make2040036

Park, I., Jeong, Y., Yoon, B., & Mortara, L. (2015).

Exploring potential R&D collaboration partners

through patent analysis based on bibliographic coupling

and latent semantic analysis. Technology Analysis &

Strategic Management, 27(7), 759-781. https://doi.org/

10.1080/09537325.2014.971004

Park, M., & Geum, Y. (2022). Two-stage technology

opportunity discovery for firm-level decision making:

GCN-based link-prediction approach. Technological

Forecasting and Social Change, 183, 121934.

https://doi.org/https://doi.org/10.1016/j.techfore.2022.

121934

Qi, Y., Zhang, X., Hu, Z. Y., Xiang, B., Zhang, R., & Fang,

S. (2022). Choosing the right collaboration partner for

innovation: a framework based on topic analysis and

link prediction. Scientometrics. https://doi.org/

10.1007/s11192-022-04306-9

Ryan, M. (2020). The Future of Transportation: Ethical,

Legal, Social and Economic Impacts of Self-driving

Vehicles in the Year 2025. Science and Engineering

Ethics, 26(3), 1185-1208. https://doi.org/10.

1007/s11948-019-00130-2

Schilling, M. A. (2012). Strategic Management of

Technological Innovation (4th edition ed.). McGraw-

Hill.

Shah, R. H., & Swaminathan, V. (2008). Factors

influencing partner selection in strategic alliances: The

moderating role of alliance context. Strategic

Management Journal, 29(5), 471-494.

https://doi.org/10.1002/smj.656

Song, B., Seol, H., & Park, Y. (2016). A patent portfolio-

based approach for assessing potential R&D partners:

An application of the Shapley value. Technological

Forecasting and Social Change, 103, 156-165.

https://doi.org/10.1016/j.techfore.2015.10.010

Vosooghi, R., Puchinger, J., Jankovic, M., & Vouillon, A.

(2019). Shared autonomous vehicle simulation and

service design. Transportation Research Part C-

Emerging Technologies, 107, 15-33.

https://doi.org/10.1016/j.trc.2019.08.006

Wang, J. F., Wang, N. T., Zhao, W. Y., & Feng, L. J.

(2025). Identifying and evaluating R&D partners via

patent-based multilayer networks from the perspective

of knowledge complementarity: A case study of

unmanned ship technology. Computers & Industrial

Engineering, 204, Article 111085.

https://doi.org/10.1016/j.cie.2025.111085

Wang, M.-Y. (2012). Exploring potential R&D

collaborators with complementary technologies: the

case of biosensors. Technological Forecasting and

Social Change, 79(5), 862-874.

Wang, X., Wang, Z., Huang, Y., Liu, Y., Zhang, J., Heng,

X., & Zhu, D. (2017). Identifying R&D partners

through Subject-Action-Object semantic analysis in a

problem & solution pattern. Technology Analysis &

Strategic Management, 29(10), 1167-1180.

Yan, E., & Guns, R. (2014). Predicting and recommending

collaborations: An author-, institution-, and country-

level analysis. Journal of Informetrics, 8(2), 295-309.

Yoon, B., & Song, B. (2014). A systematic approach of

partner selection for open innovation. Industrial

Management & Data Systems, 114(7), 1068-1093.

KDIR 2025 - 17th International Conference on Knowledge Discovery and Information Retrieval