Research on Frontier Discovery of Technological Innovation

Based on Knowledge Flow

Dechao Wang

, Yongjie Li

, Jian Zhu

and Xiaoli Tang

Institute of Medical Information & Library, Chinese Academy of Medical Sciences and Peking Union Medical College,

Beijing, P. R. China

Keywords: Technological Innovation Frontier, Knowledge Flow, Science-Technology Linkage, Knowledge Meme.

Abstract: Amidst intensifying technological competition, technological innovation fundamentally arises from the flow

of scientific knowledge into technical knowledge. Precisely characterizing the features and connotation of

this knowledge flow is therefore crucial for identifying the frontiers of technological innovation. Adopting a

semantic flow perspective, this study developed a simulation framework to model semantic flow between

documents, progressing from key knowledge elements to the full-text level. Leveraging deep learning models,

it then re-identified knowledge flow relationships between documents. The concept of knowledge meme was

introduced to quantify the propagation dynamics (intensity and scope) of knowledge units across scientific

and technical knowledge systems. Subsequently, a knowledge flow network connecting patents and academic

papers in the lung cancer domain was constructed. Building upon this network, the substantive content of the

knowledge flows was measured. This research achieved the identification and reconstruction of knowledge

flow relationships between scientific and technical documents. Furthermore, by analyzing the content and

communication patterns of computable knowledge units, it elucidated the frontiers of technological

innovation. This approach holds significant implications for understanding science-technology linkages and

identifying emerging technological innovation frontiers.

1 INTRODUCTION

Amidst an increasingly complex and volatile global

landscape, rapid scientific advancement and

accelerated technological iteration have elevated

technological innovation to a critical determinant of

national scientific prowess and international

competitiveness. The frontiers of technological

innovation, characterized by novelty,

interdisciplinarity, and high visibility, represent

clusters of research achievements that spearhead

progress within technological domains (Shibata et al.,

2008). These frontiers play a pivotal role in guiding

future scientific and technological development.

Crucially, science and technology serve as key

enablers for discovering innovation opportunities,

with most patent innovations drawing upon scientific

foundations. The flow of knowledge from scientific

https://orcid.org/0000-0002-1838-8168

https://orcid.org/0009-0004-6306-8288

https://orcid.org/0009-0008-8222-0280

https://orcid.org/0000-0001-6946-3482

publications into technical domains, exemplified by

patents, is instrumental in generating technical

knowledge (Roh et al., 2023). Consequently, delving

into science-technology linkages offers an effective

pathway to uncover technological opportunities and

pinpoint innovation frontiers (Robinson et al., 2013).

This study adopts the theoretical lens of

knowledge flow to articulate the content and direction

of knowledge absorption, growth, and dissemination

among distinct entities (Hai, et al., 2006). Within this

framework, we define the frontier of technological

innovation as a cluster of knowledge content flowing

from science to technology.

Leveraging knowledge flow relationships to

model science's contribution to technology, this

research devised a knowledge flow recognition

algorithm capable of identifying semantic inclusion

relations. Building upon this algorithm, we

294

Wang, D., Li, Y., Zhu, J. and Tang, X.

Research on Frontier Discovery of Technological Innovation Based on Knowledge Flow.

DOI: 10.5220/0013708800004000

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 294-301

constructed a knowledge flow network connecting

academic papers and patents. Subsequently, utilizing

this network, we investigated the propagation

pathways of knowledge meme during knowledge

dissemination. This analysis aims to elucidate the

current frontiers of technological innovation within

the target technological field, thereby providing

foundational insights and references for related

scientific and technological research endeavors.

2 RELATED WORK

2.1 Identification of the Frontiers of

Technological Innovation

Research on identifying the frontiers of technological

innovation encompasses bibliometric or cluster

analyses conducted solely within the technological

domain (Roche et al., 2010; Kim et al., 2017; Xu et

al., 2022; Li et al., 2015). However, a greater number

of studies opt to explore these frontiers by analyzing

science-technology linkages. This approach involves

examining the relationships among knowledge units

across scientific and technological systems,

measuring the direction, intensity, and structure of

their dissemination and transfer to reveal interactions

between science and technology (Ba et al., 2021). It

also facilitates research into discovering

technological opportunities through these linkages,

tracking transformation patterns of scientific and

technological knowledge, or identifying innovation

frontiers (Han et al., 2022; Du et al., 2019; Tian et al.,

2024; Xu et al., 2020; Du et al., 2019). Scholarly

papers and patents are widely recognized as

representative outputs of scientific and technological

research, respectively (Ahmadpoor et al., 2017).

Primary methodologies for analyzing science-

technology linkages include: citation correlation

analysis between papers and patents (Nguyen et al.,

2019; Kenan-Flagler et al., 2011; Li et al., 2024),

author-inventor analysis (Boyack et al., 2008; Breschi

et al., 2010; Ning et al., 2020), and knowledge

structure analysis (Xu et al., 2022; Du et al., 2024;

Zhang et al., 2022; Ran et al., 2024).

These methods collectively aim to construct

specific relationships between patents and papers,

enabling the identification and analysis of knowledge

clusters flowing between science and technology

domains, thereby uncovering innovation frontiers.

Nevertheless, these approaches exhibit significant

limitations. Citation correlation analysis suffers from

inherent constraints of citation relationships

themselves — including the questionable semantic

correlation between citing and cited documents (Li et

al., 2014; Meyer, 2000), the scarcity of citations

between patents and papers (Xu et al., 2022; Callaert

et al., 2006), the difficulty in constructing potential

citation links, and the potential intentional

concealment of patent citations (Wu et al., 2017)—

resulting in compromised data quality. Author-

inventor analysis establishes connections via

researcher identities but neglects the underlying

semantic association between science and

technology, failing to accurately capture deep data

linkages. Among knowledge structure analysis

methods, vocabulary- or topic-based approaches are

more effective at revealing semantics; however, they

face challenges in systematically constructing

knowledge structure networks (Ba et al., 2021) and

delineating propagation pathways.

2.2 Knowledge Flow Relationships

Knowledge flow describes the process by which

knowledge disseminates and transfers among

different entities, domains, or systems. Knowledge

flow relationships are conventionally measured using

citation data (Criscuolo et al., 2008; Lyu et al., 2022;

Zhao et al., 2022) and can also be inferred through

indirect citation chains (Feng et al., 2023). However,

the implicit, ambiguous, and complex nature of

technological interactions complicates the revelation

of intrinsic relationships within scientific and

technological knowledge (Chen et al., 2023).

Crucially, citation relationships cannot fully represent

knowledge flow (Meyer, 2000); even when

incorporating indirect citations, they inadequately

capture the knowledge contribution and academic

influence across disciplines (Roh et al., 2023).

From a knowledge flow perspective, its core

elements encompass the subject, content, and

direction. To address the shortcomings of the

aforementioned methods, researchers require

methodologies capable of constructing accurate

knowledge flow relationships between scientific and

technological entities and analyzing the substantive

content of these flows through effective semantic

techniques (Kang et al., 2022; Zhang et al., 2024).

2.3 Knowledge Meme

The term "meme" originated in Dawkins' seminal

work The Selfish Gene, conceptualized as the

functional unit of knowledge inheritance and

variation, reflecting the process of knowledge flow

and dissemination (Yang et al., 2021). Within the

Research on Frontier Discovery of Technological Innovation Based on Knowledge Flow

295

scientific domain, extracting meme facilitates the

discovery of semantic information embedded in

academic papers (Zeng et al., 2023), and their

propagation serves to quantify the diffusion patterns

of knowledge flow (Mao et al., 2024; Kamada et al.,

2021). Fundamentally, innovation constitutes a process

of meme-based search, combination, experimentation,

and adjustment. Kuhn et al. (2014) defined scientific

meme as short text units within scientific publications

whose semantics are replicated when the publications

are cited. Similarly, Sun et al. (2018) defined technical

meme as short text units within patents whose

semantics are replicated upon citation. However,

research on scientific or technological meme

predominantly relies on citation relations (Araújo et al.,

2018). Consequently, tracking meme propagation

remains largely confined to citation analysis, failing to

transcend its inherent limitations. Nevertheless, this

research provides a novel conceptual framework for

reflecting knowledge flow— specifically, considering

the direction and volume of knowledge movement at

the level of semantic content.

Building upon the concepts of scientific and

technological meme, this study defines knowledge

meme as "short text units within scientific or

technological publications whose semantics are

replicated when knowledge flow occurs among

publications."

3 DATA AND METHODS

3.1 Data

This study selected the field of lung cancer as the

empirical research domain. The primary dataset

comprised lung cancer patents (Dataset A) retrieved

from the Dimensions database, covering patent grants

issued between January 2019 and December 2023.

Following the consolidation of patent families, Dataset

A contained 6,671 unique patents. Building upon this

foundation, the referenced patents (source: incoPat)

and referenced scientific publications (source:

PubMed) cited by these lung cancer patents were

collected by matching patent numbers with PMID

identifiers. These referenced documents were

amalgamated into Dataset B, which contained 19,453

referenced patents and 12,394 referenced publications.

3.2 Research Design

This research is predicated on three fundamental

assumptions:

1) Knowledge flow occurs within citation

relationships between patents and

publications.

2) This knowledge flow is effectively captured

by the propagation of knowledge meme

semantics.

3) Knowledge flow extends beyond the existing

citation network, encompassing relationships

between patents and publications where

semantic transfer of knowledge meme occurs

independently of direct citation links.

Aligned with the characteristics of knowledge

meme dissemination, this study conceptualizes

citation-based knowledge flow as the semantic

containment of key knowledge elements (represented

by meme) from the knowledge-outflow entity

(publication) within the text of the knowledge-inflow

entity (patent). Consequently, a knowledge flow

identification model was designed. This model learns

the characteristics of knowledge flow within the

known citation network (Dataset B) to predict

knowledge flow relationships existing outside this

network.

Figure 1: Frontier discovery of technological innovation

based on knowledge flow.

This study designed and trained a deep learning

algorithm, based on PubmedBERT+Bi-LSTM, to

identify whether the semantics of knowledge meme

derived from a publication are contained within the

text semantics of a patent. This identification process

enables the construction of knowledge flow

relationships between publications and patents. The

model was trained using the lung cancer patent data

(Dataset A) and their corresponding references

(Dataset B). Subsequently, the trained model was

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

296

applied to identify potential knowledge flow

relationships between all patents and publications

within the lung cancer domain. This facilitated the

construction of a comprehensive knowledge flow

network for lung cancer. Finally, leveraging this

network, the propagation intensity and propagation

scope of knowledge meme were calculated. This

analysis aimed to assess the impact of science on

technological development and identify the frontiers

of technological innovation within the lung cancer

field. The overall research workflow is depicted in

Figure 1.

3.3 Knowledge Meme Analysis

Traditionally, meme extraction employs rule-based

methods, often utilizing high-frequency words from

the corpus after removing stop words, functional

terms, and meaningless tokens to form a candidate

meme set. However, such rule-based approaches are

semantically constrained and struggle to identify and

merge candidate meme sharing identical semantics.

Within natural language processing (NLP), the

keyword extraction task automatically selects phrases

from a document to summarize its content. Keyword

extraction algorithms leveraging pre-trained language

models capture semantic information within

documents, enabling effective identification and

output of key information as keywords. This

capability aligns well with the defining characteristics

of meme and offers the potential for enhanced meme

recognition. Therefore, this study employed the

PromptRank model — a current state-of-the-art

(SOTA) algorithm for keyword extraction (Kong et

al., 2023) —to extract keywords from all patent and

publication documents. The resulting keywords were

treated as the knowledge meme for subsequent

knowledge flow analysis and computation.

Driven by the features of meme propagation

mechanisms, this study adopted the method proposed

by Kuhn et al. to calculate a comprehensive score for

candidate meme. To address variations in term

expression, including synonyms and lexical variants,

we incorporated an additional word sense

disambiguation step. Specifically, a word embedding

generation model was utilized to generate

embeddings for candidate meme words. Knowledge

meme exhibiting a cosine similarity score exceeding

0.9 were subsequently merged.

The keywords extracted from all patents within

the lung cancer dataset (Dataset A) served as the

initial candidate knowledge meme. Following this,

the comprehensive score for each candidate

knowledge meme was calculated sequentially based

on the constructed knowledge flow network, utilizing

Formulas (1) and (2).

𝑃





𝑑

→

𝑑

→



𝛿

𝑑

→

𝛿

𝑑

→



𝛿

(1)

𝑀





𝑓



𝑃



(2)

Where, P

is the propagation score of knowledge

meme; f

is the document frequency of knowledge

meme; M

is the composite knowledge meme score.

m→m

indicates the number of knowledge-inflow

documents containing m and at least one of the

associated knowledge-outflow documents containing

m; d

→m

indicates the number of knowledge-inflow

documents with at least one associated knowledge-

outflow document containing knowledge meme m;

m→m

indicates that the knowledge meme m are

contained, but the associated knowledge outflow

documents do not contain the number of knowledge

inflow documents of the knowledge meme; d

→m

indicates that all associated knowledge-outflow

documents do not contain the number of knowledge-

inflow documents of the knowledge meme. To

prevent the denominator from being zero, set δ as

the smoothing factor.

3.4 Knowledge Flow Identification

Algorithm

3.4.1 Algorithm Design

As illustrated in Figure 2, the knowledge flow

identification algorithm developed in this study

comprises three primary stages:

Embedding Generation: Producing sentence-level

and word-level embeddings.

Sequence Data Processing: Handling the

sequential nature of input features.

Classification Prediction: Determining the

presence of a knowledge flow relationship.

Figure 2: The structure of knowledge flow identification

algorithm.

Research on Frontier Discovery of Technological Innovation Based on Knowledge Flow

297

Stage 1: Embedding Generation

To enhance the model's capacity for capturing

domain-specific semantics and terminology within

the biomedical field, the PubMedBERT pre-trained

language model was utilized.

The text of the knowledge-inflow document

(typically a patent) was directly processed by

PubMedBERT to generate sentence-level

embeddings.

The keywords (representing knowledge meme)

extracted from the knowledge-outflow document

(typically a publication) were initially embedded

using PubMedBERT to generate initial word

embeddings. Subsequently, these initial embeddings

were further optimized by training a dedicated word

embedding vector model. This optimization

incorporated the full-text contextual information

from both patents and publications to refine the

semantic embedding representation of the keywords.

Stage 2: Sequence Data Processing

Given the sequential dependence and contextual

relevance inherent in the input features for knowledge

flow identification, the generated word embeddings

(representing outflow meme) and sentence

embeddings (representing inflow document context)

were concatenated and fed into a Bidirectional Long

Short-Term Memory (Bi-LSTM) layer. The Bi-

LSTM network effectively processes this sequential

input to capture complex dependencies.

Stage 3: Classification Prediction

The output vectors from the Bi-LSTM layer were

sequentially passed through a linear layer and a

Softmax activation layer. This process yielded a

probability score indicating the likelihood of a

semantic containment relationship (i.e., a positive

knowledge flow relationship) existing between the

input pair. All predicted results were ranked by their

probability scores. The top n results, based on the

highest probability scores, were selected as positive

predictions.

3.4.2 Construction of Algorithm Dataset

The algorithm training and evaluation datasets were

constructed from the lung cancer patent dataset

(Dataset A) and the reference dataset (Dataset B,

containing cited publications and patents).

Positive Samples: Pairs were formed based on

explicit citation relationships (e.g., a lung cancer

patent citing a paper or another patent).

Negative Samples: For a given document (Feature

A), documents within the reference dataset (Dataset

B) that had no citation relationship with it were

randomly selected to form negative pairs (Feature B).

Patents within the lung cancer dataset (Dataset A)

that possessed citations were partitioned into training,

validation, and test sets using a 6:2:2 ratio. Reflecting

the typical imbalance in such tasks, the number of

randomly generated negative samples in both the

training and validation sets was set to four times the

number of positive samples. Within the test set,

negative samples were generated at ratios of either

five or ten times the number of positive samples.

The neural network model was trained using input

pairs consisting of the two sets of embedding features

along with their corresponding classification labels

(indicating positive or negative knowledge flow

relationship). The model's objective was to identify

the semantic containment relationship between the

embeddings of Feature A and Feature B.

Post-processing for Chronological Consistency:

To ensure temporal validity of the predicted

knowledge flow relationships, a post-processing step

was applied. Any positive prediction where the

publication date of the knowledge-outflow document

(Feature B) occurred later than the publication date of

the knowledge-inflow document (Feature A) was

reclassified as a negative result, as knowledge cannot

logically flow from a future document to a past one.

4 RESULTS

4.1 Algorithm Evaluation

The model was evaluated on two distinct test sets,

characterized by class imbalance ratios (positive to

negative samples) of 1:5 and 1:10, respectively. This

design reflects the inherent scarcity of knowledge

flow relationships compared to non-flow pairs within

the data. Given that the model outputs a probability

score for the existence of a knowledge flow

relationship between any document pair, the top n

ranked predictions by probability were selected as

positive identifications.

As the knowledge flow identification algorithm

functions as a ranking task, standard information

retrieval metrics were employed for evaluation: Mean

Reciprocal Rank (MRR), Mean Average Precision

(MAP), Recall, Precision, and F1-score. The

performance metrics of the algorithm on both test sets

are presented in Table 1.

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

298

Table 1: The algorithm test results.

Positive/

negative

sample

ratio

MRR MAP Recall Precision F1

1:5 0.8309 0.6601 0.8038 0.7692 0.7861

1:10 0.7579 0.5713 0.8029 0.6967 0.7460

Despite potential data quality limitations inherent

in the training samples (notably, the non-correlative

nature of some citation relationships), the knowledge

flow identification algorithm demonstrated robust

performance on the citation-based dataset. This

validates its applicability for constructing interactive

networks of scientific and technological knowledge.

4.2 Memetic Comprehensive Score

Calculation

An empirical dataset was constructed comprising

patents granted between 2022 and 2023 and

publications issued between 2019 and 2023 within the

lung cancer domain. The trained model was then

applied to identify potential knowledge flow

relationships between these publications and patents.

For each patent, the top 5 publications (published

prior to the patent application date) with the highest

predicted probability scores were identified as

potential knowledge-outflow sources.

Applying a smoothing factor of 0.5, the

comprehensive scores for all candidate knowledge

meme were computed and ranked. Table 2 lists the

top 10 ranked candidate knowledge meme alongside

their comprehensive scores.

Table 2: Ranking of Candidate Knowledge meme by

Comprehensive Score (Top 10).

Keywords Comprehensive score

anti-pd-l1 antibody 2.013221

peptides 1.423944

anti-tumor immune responses 1.312993

bronchial asthma 1.214736

anti-pd-1 antibody 1.182252

compounds 0.990032

tumor immunotherapy 0.912560

nucleic acid aptamer 0.899915

air pollution 0.895928

cell peptide epitopes 0.883943

Subsequent to calculating the comprehensive

meme scores, and leveraging the knowledge flow

relationships identified among patents within the lung

cancer dataset by the trained model, a "patent-patent

knowledge flow network" was constructed. This

network enables the exploration of meme propagation

dynamics within the technological knowledge system

after their initial flow from scientific publications into

patents.

5 DISCUSSION

This study addresses critical challenges in identifying

the frontiers of technological innovation—namely, the

inadequacy of capturing science-technology linkages

and the limited semantic recognition capabilities

prevalent in current methodologies. The ability of

traditional approaches to accurately reveal deep-level

semantic flow relationships is fundamentally

constrained by inherent data limitations and

methodological shortcomings.

To overcome these limitations, this research

proposed a semantic-driven approach centered on

knowledge flow. A pivotal contribution is the

introduction of knowledge meme as computable units

of knowledge transfer. By enabling the extraction and

quantitative measurement of these knowledge meme,

this study provides novel semantic-level perspectives

and methods for analyzing the frontiers of

technological innovation.

The knowledge flow identification model was

trained based on citation relationships. However, the

representativeness of citations as proxies for genuine

knowledge flow relationships is inherently limited

(Meyer, 2000; Chen et al., 2023), potentially

introducing bias into the model's predictions.

Future research will focus on several key

directions:

Enhancing Meme Interpretation: Delving deeper

into the semantic information embedded within

knowledge meme, employing techniques such as

classification, combinatorial analysis, and logical

deduction to enhance the interpretability of the

results.

Multidimensional Validation: Employing

bibliometric methods to analyze the constructed

knowledge flow networks and cross-validating the

findings derived from knowledge meme analysis with

these network-based insights.

Research on Frontier Discovery of Technological Innovation Based on Knowledge Flow

299

REFERENCES

Ahmadpoor, M.A., & Jones, B.F. (2017). The dual frontier:

Patented inventions and prior scientific advance.

Science, 357, 583 - 587.

Araújo, T., & Fontainha, E. (2018). Are scientific meme

inherited differently from gendered authorship?

Scientometrics, 117, 953–972.

Ba, Z., & Liang, Z. (2021). A novel approach to measuring

science-technology linkage: From the perspective of

knowledge network coupling. Journal of Informetrics,

15(3), 101167.

Boyack, K. W., & Klavans, R. (2008). Measuring science-

technology interaction using rare inventor-author

names. Journal of Informetrics, 2, 173–182.

Breschi, S., & Catalini, C. (2010). Tracing the links

between science and technology: An exploratory

analysis of scientists’ and inventors’ networks.

Research Policy, 39(1), 14–26.

Callaert, J., Van Looy, B., Verbeek, A., et al. (2006). Traces

of Prior Art: An analysis of non-patent references found

in patent documents. Scientometrics, 69, 3–20.

Chen, X., Ye, P., Huang, L., et al. (2023). Exploring

science-technology linkages: A deep learning-

empowered solution. Information Processing &

Management, 60(2), 102.

Criscuolo, P., & Verspagen, B. (2008). Does it matter

where patent citations come from? Inventor vs.

examiner citations in European patents. Research

Policy, 37, 1892–1908.

Du, C., Yao, K., Zhu, H., et al. (2024). Mining technology

trends in scientific publications: A graph propagated

neural topic modeling approach. Knowledge and

Information Systems. Advance online publication.

Du, J., Li, P., Guo, Q., et al. (2019). Measuring the

knowledge translation and convergence in

pharmaceutical innovation by funding-science-

technology-innovation linkages analysis. Journal of

Informetrics, 13, 132–148.

Du, J., Sun, Y., Li, Y., et al. (2019). Identifying innovation

frontier at the interface of science and technology: A

bibliometric framework and empirical study [In

Chinese]. Information Studies: Theory & Application,

42(1), 94–99.

Feng, S., Li, H., & Qi, Y. (2023). How to detect the sleeping

beauty papers and princes in technology considering

indirect citations? Journal of Informetrics, 17, 101431.

Hai, Z. (2006). Discovery of knowledge flow in science.

Communications of the ACM, 49(5), 101–107.

Han, X., Zhu, D., & Wang, X. (2022). Research on the

method of technology opportunity discovery promoted

by science [In Chinese]. Library and Information

Service, 66(10), 19–32.

Kamada, M., Asatani, K., Isonuma, M., et al. (2021).

Discovering interdisciplinarily spread knowledge in the

academic literature. IEEE Access, 9, 124142–124151.

Kang, X., Jia, X., Deng, L., et al. (2022). Research on the

characteristics of high-impact patent knowledge

diffusion based on all generation citation network [In

Chinese]. Library and Information Service, 66(22), 83–

94.

Kenney, M. R. (2011). Lens or prism? A comparative

assessment of patent citations as a measure of

knowledge flows from public research. Management

Science, 59(2), 504–525.

Kim, G., & Bae, J. (2017). A novel approach to forecast

promising technology through patent analysis.

Technological Forecasting and Social Change, 117,

228–237.

Kong, A., Zhao, S., Chen, H., et al. (2023). PromptRank:

Unsupervised keyphrase extraction using prompt.

Proceedings of the 61st Annual Meeting of the

Association for Computational Linguistics (Vol. 1, pp.

9788–9801).

Kuhn, T., Perc, M., & Helbing, D. (2014). Inheritance

patterns in citation networks reveal scientific meme.

Physical Review X, 4(4), 041002.

Li, B., & Chen, X. (2015). Identification of emerging

technologies in nanotechnology based on citing

coupling clustering of patents [In Chinese]. Journal of

Intelligence, 34(5), 35–40.

Li, B., Ding, K., Sun, X., et al. (2024). Research on the

diffusion speed and diffusion effects of scientific papers

into the technological domain [In Chinese]. Information

Studies: Theory & Application, 47(7), 35–47.

Li, R., Chambers, T., Ding, Y., et al. (2014). Patent citation

analysis: Calculating science linkage based on citing

motivation. Journal of the Association for Information

Science and Technology, 65.

Lyu, H., Bu, Y., Zhao, Z., et al. (2022). Citation bias in

measuring knowledge flow: Evidence from the web of

science at the discipline level. Journal of Informetrics,

16(4), 101338.

Mao, J., Liang, Z., Cao, Y., et al. (2024). Quantifying cross-

disciplinary knowledge flow from the perspective of

content: Introducing an interdisciplinary distance

indicator. Journal of Informetrics, 17(2), 101092.

Meyer, M. S. (2000). Does science push technology?

Patents citing scientific literature. Research Policy, 29,

409–434.

Nguyen, A. L., Liu, W., Khor, K. A., et al. (2019). The

golden eras of graphene science and technology:

Bibliographic evidences from journal and patent

publications. Journal of Informetrics, 14, 101067.

Ning, Z., & Wei, L. (2020). Research on the relationship

between patent documents and academic papers based

on patent subjects: A case study of data mining [In

Chinese]. Library and Information Service, 64(12),

106–117.

Robinson, D., Huang, L., Guo, Y., et al. (2013). Forecasting

innovation pathways (FIP) for new and emerging

science and technologies. Technological Forecasting

and Social Change, 80(2), 267–285.

Roche, I., Besagni, D., François, C., et al. (2010).

Identification and characterisation of technological

topics in the field of molecular biology. Scientometrics,

82, 663–676.

Roh, T., & Yoon, B. (2023). Discovering technology and

science innovation opportunity based on sentence

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

300

generation algorithm. Journal of Informetrics, 17,

101403.

Ran, C., Tian, W., & Jia, Z. (2024). Modeling of scientific

paper-patent technology association relationship based

on mixed methods: Taking the biomedical field as an

example [In Chinese]. Information Science. Advance

online publication.

Shibata, N., Kajikawa, Y., Takeda, Y., et al. (2008).

Detecting emerging research fronts based on

topological measures in citation networks of scientific

publications. Technovation, 28, 758–775.

Sun, X., & Ding, K. (2018). Identifying and tracking

scientific and technological knowledge meme from

citation networks of publications and patents.

Scientometrics, 116, 1735–1748.

Tian, C., Dong, K., Guo, R., et al. (2024). Research on

measurement method of transformation efficiency of

scientific and technical achievements based on

knowledge association analysis [In Chinese].

Information Studies: Theory & Application, 47(5),

123–130.

Wu, H., & Ji, F. (2017). Empirical research on the

evaluation effectiveness of patent citation based on

patent application and patent censorship [In Chinese].

Library and Information Service, 61(19), 89–95.

Xu, H., Winnink, J. J., Yue, Z., et al. (2020). Topic-linked

innovation paths in science and technology. Journal of

Informetrics, 14, 101014.

Xu, H., Yue, Z., Pang, H., et al. (2022). Integrative model

for discovering linked topics in science and technology.

Journal of Informetrics, 16, 101265.

Xu, X., Wu, F., & Wang, B. (2022). Research on

identification of key core technology based on

international patent classification [In Chinese]. Journal

of Intelligence, 41(10), 74–81.

Yang, F., Qiao, Y., Wang, S., et al. (2021). Blockchain and

multi-agent system for meme discovery and prediction

in social network. Knowledge-Based Systems, 229,

107368.

Zeng, J., Cao, S., Chen, Y., et al. (2023). Measuring the

interdisciplinary characteristics of Chinese research in

library and information science based on knowledge

elements. Aslib Journal of Information Management,

75, 589–617.

Zhang, B., Wu, H., Gao, D., et al. (2022). Research on

identification of innovation fronts based on potentially

high cited papers and high value patents [In Chinese].

Library and Information Service, 66(18), 72–83.

Zhang, J., Kang, L., & Sun, J. (2024). The influence of

recency and time-span in the scientific and

technological knowledge convergence [In Chinese].

Journal of Information Resources Management, 14(4),

86–102.

Zhao, H., & Wang, X. (2022). Characteristics and

evolutionary trends of knowledge flow in

interdisciplinary research under the background of open

science: Taking the study of "Five-Metrics" as an

example [In Chinese]. Information Science, 40(4), 107–

117.

Research on Frontier Discovery of Technological Innovation Based on Knowledge Flow

301