Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Identifying the latent technology opportunities based
on a perspective of coupling publications with patents
Munan LI 1, 2 Wenshu Wang 1 Keyu Zhou 1
2019/10/20
1. School of Business Administration, South China University of Technology
1. Guangdong Province Key Laboratory on Decision & Innovation Management
Outline
Related workIntroduction 1
2
Methodology 3
5
Case study
4
Conclusion & Discussions
• Within a specific topic, the publications and patents are the outputs ofscientific research and technology transformation respectively, and also theyare often utilized to evaluate the levels of R&D.
Introduction
Publications
Patents
R & D(Nations, Institutes
and Staffs)
Level
Performance
Efficiency
……
• Under the traditional framework on identifying the TO (technology opportunity), thepatents are the usually data source. And, the classic and emerging analytical methodson patents are the mainstream of current studies.
Introduction
PatentsIdentification or evaluation of TO
Efficacy matrix
Topic modeling
SAO(Subject-Action-Object) Analysis
……
Patent mapping
• Research questions
• RQ1: Within some topics or under the specific situations, the coupling relationsbetween publications and patents can be conducted to identify the latent technologyopportunities?
• RQ2: And how can we use these coupling relations between publications and patentsto identify TO?
• RQ3: The new TO based on the proposed coupling methods can be identified by theother methods or can be compared with the other methods including traditional oremerging?
Introduction
• The driving force of technology progress has been generally discussed since 1950s ,and it should be endogenous from the view of productivity development (Solow,1957). In consequence, to forecast the evolution of technology, or technologyforesight has become one of critical approaches for strategic planning and industrial-policy making (Martin, 1995; Hasan and Tucci, 2010; Miles, 2010; Li et al., 2018).
• In general, the data of publications is often utilized to evaluate the scientific outputand academic performance, efficiency or collaboration between different institutes orscientists, and so forth (Guenter et al., 2007; Kyle et al., 2015).
• Some latest studies have integrated social media data, e.g. Twitter, Facebook orWeChat into the framework of technological forecasting and TOA (Chen, 2018).
Related work
• Regarding the relevant studies on TOA (technology opportunity analysis), technologyforecasting, emerging technologies, TCE(technological capability evaluation) andtech mining etc., the patent data play the key role in those processing (Breitzman &Mogee, 2002; Tseng et al., 2007; Youtie et al, 2008; Lee et al, 2009; Porter &Newman, 2011; Zhang et al, 2016; Wang et al., 2017).
• Also, the technique on combing bibliometrics with patent analysis has beenconducted to forecast those emerging technologies or R&D evaluation (Hullmannand Meyer, 2003; Daim et al., 2006).
Related work
Related work
Bibliometrics & publications data
Patent data & Patent analysis
Bibliometrics & Patent analysis
Multiple data source analyses
Scientific evaluation on output, efficiency, performance and international collaborations and so forth.
TOA (Technology Opportunity Analysis), TCE (Technology Capability Evaluation), Technology Evolution Analysis and so on.
Forecasting emerging technologies,Evaluating the performance or efficiency of R&D
Social medias data(Twitter, Facebook, WeChatInternet data (homepages, search results and so on)
Basically, the general computing-framework based on the coupling relations betweenpublications and patents is still insufficient, especially for those multi-disciplinary ortrans-disciplinary topics, for example, artificial intelligence, smart cities and IndustrialInternet etc.
Methodology
Here, an hypothesized framework is proposed to forecast the latent technologyopportunities based on the coupling relations between scientific articles andcorresponding patents within a specific topic.
Publications
Patents
Publications Patents Approximate classification
More More High R(Research) & High D(Development)
More Less High R(Research) & Low D(Development)
Less More Low R(Research) & High D(Development)
Less Less Low R(Research) & Low D(Development)
Methodology
Corresponding concepts and equations for computing:Perplexity: In natural language processing, it is used to measure the quality of the trainedlanguage model and evaluate the generalization ability of the model (Blei et al., 2003).
Perplexity
= exp −(� � log(��(��
�
���
��
���
�
���
|��)�(��|��)))/(� ��)
�
���
Coherence: The coherence of topics can be measured by calculating the degree of correlationbetween the features with higher scores in the topics, which is helpful to classify the topicsinto understandable categories (Newman et al., 2010; Stevens, 2012).
Coherence=������(��,
���
�� , �)
Methodology
For a specific topic, the coupling relationships can be described as such two aspects: For a specific term, the amount of publications and the amount of patents can present
some useful information. Based on the topic modeling, some sub-topics can be extracted from the retrieved
publications or patents, and the similarities between these sub-topics can be computed andconducted to evaluate the strength of coupling relations.
Sub topic #1
Sub topic #2
Sub topic #n
Sub topic #1
Sub topic #2
Sub topic #n
Publications Patents
…… ……
Top modeling
Top modeling
Methodology
SDS (Similarity between Different Sub-topics): the Euclidean Distance between two sub-topics, which also needs some techniques for computation.
Two new concepts are proposed:TI (Transformation Intensity): we just utilized two factors: (1) the probability of a term
belonging to a specific sub-topic in publications; (2) this term has been transformed intohow many sub-topics of the corresponding patents.
TE(Transforming Efficiency): here, the concept of HLP(half-life period) is conducted tocalculate the TE.
� �, � = �� − ��� + �� − ��
� +⋯+ �� − ���
����= ����
�
���
⋅ ��
����
= ���� ⋅ ��
�
���
+����
�
���
⋅ ��⋅1
2
�
+����
�
���
⋅ �� ⋅1
2
�
March, 2016, the milestone event about Go game, AlphaGo vs Sedol Lee
(4:1)
Case study: Artificial intelligence
A mazing hand in the center from
Sedol Lee
s1
幻灯片 13
s1 scutlimn, 2019/7/23
In 1956 , the Dartmouth conference is thought as the mark event about the origins of artificial intelligence.
Case study: Artificial intelligence
In August 1956, at Dartmouth college, JohnMcCarthy , Marvin Minsky, Claude Shannon,Alan Newell , Herbert Simon and otherscientists are together, talking about acomplete out-of-touch theme: Using machinesto mimic human learning and other aspectsof intelligence.
The meeting lasted for two months, andalthough there was no general consensus, aname was given to the discussion: artificialintelligence. Thus, 1956 became the first yearof artificial intelligence.
In the 1960s, artificial intelligence evolved different theories and branches such as computational complexity, cognitivecomputing and computational linguistics. With the promotion and application of relevant models and algorithms of artificialintelligence, research related to human labor union intelligence has spread to different fields and disciplines.
Case study: Artificial intelligence
Considering that there are many researches on artificial intelligence and the scope of knowledge diffusion is wide, that is, itis more or less involved in most research fields, the author considers topic modeling to conduct more in-depth contentmining and presentation. LDA(Latent Dirichlet Allocation)LDA is a popular and mature topic mining algorithm at present.In essence, LDA is a Bayesian model including subject, document and topic, which is completely based on bayesianreasoning mechanism and has good knowledge interpretation ability (Blei 2003; 2012).
Case study: Artificial intelligence
Case study: Artificial intelligence
0
500
1000
1500
2000
2500
3000
3500
4000
Amount
CHINESE ACAD SCI
ISLAMIC AZAD UNIV
INDIAN INST TECHNOL
ZHEJIANG UNIV
UNIV TEHRAN INTELLIGENCE
SHANGHAI JIAO TONG UNIV
NANYANG TECHNOL UNIV
NATL UNIV SINGAPORE
Case study: Artificial intelligence
3770
7351
0
1000
2000
3000
4000
5000
6000
7000
8000
199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017
Pat
ents
Patentgrants
Patent applications
Rank Regions Amout of Patents
1 CHINA 140222 USA 68893 JAPAN 30854 KOREAN 1775
5 GERMANY 724
Patent applications surged 7,351 in 2017, of which 5,655 were from China, accounting for 76.93 percent of the total growth in 2017.
Case study: Artificial intelligence
1990-1999 2000-2009 2010-2017
Topic
codePublications
Topic tag Topic
code
Publicati
ons
Topic tag Topic
codePublications
Topic tag
L1-6 20957
Classificationand classifier L2-13 57945
Artificial neuralnetwork predictionmodel
L3-7
79750 Fuzzy model and fuzzysystem
L1-7 19345
Associativememory andpatternrecognition
L2-12 51028
Dynamic neuralnetwork stability
L3-18
67900 Neural network model
L1-10 13698
Nonlineardynamicsimulation
L2-8 48661
Optimization designof genetic algorithm L3-5
64369 Parametric predictionmodel
L1-14 13200Clinical medicine
L2-2 48181Protein structuresequence prediction
L3-10
63070 Support vector machineclassification
L1-9 12593Adaptive fuzzycontrol
L2-3 45952Spectral analysis ofcancer
L3-15
60883 Optimization design ofgenetic algorithm
Case study: Artificial intelligence
1990-1999 2000-2009 2010-2017
Topic
codePatents
Topic tag Topic
codePatents
Topic tag Topic
codePatents
Topic tag
P1-4 2001 Neural network P2-5 5494 Drawing methodsbased on neuralnetwork
P3-8 12276 Data transmission ofneural network
P1-1 1731 Image processing P2-3 4363 Expert system P3-10 11979 Drawing algorithm
P1-6 1686 Signal sensor P2-6 3787 Pattern recognition P3-5 11503 Vector drawing method
P1-5 1369 Computer system P2-4 3695 Image description P3-7 10980 Modular control system
P1-2 1318 System andparameters ofneural network
P2-1 3589 Traffic sensoring P3-9 10724 Signaling system
Case study: Artificial intelligence
The thickness of the line
between two topics is expressed as
follows: the smaller the distance
(the smaller the difference) is, the
thicker the line, indicating the
closeness of the relationship and
the degree of similarity.
The sub-topics evolution of
publications on artificial
intelligence
Case study: Artificial intelligence
The sub-topics evolution
of patents on artificial
intelligence
Case study: Artificial intelligence
The TI (Transformation Intensity) on those sub-topics on artificial intelligence
0.000
0.500
1.000
1.500
2.000
2.500
L1-
9L
1-7
L1
-11
L1-
1
L1-
4L
1-8
L1-
2L
1-1
2L
1-6
L1
-14
L1-
3L
1-1
0L
1-5
L1
-13
L2
-13
L2
-12
L2-
9L
2-7
L2-
4L
2-6
L2-
8L
2-3
L2
-10
L2-
1L
2-5
L2-
2L
2-1
1L
3-1
8L
3-1
L3-
2L
3-3
L3-
7L
3-1
0L
3-5
L3
-12
L3-
9L
3-1
3L
3-8
L3
-16
L3
-11
L3
-15
L3-
4L
3-6
L3
-14
L3
-17
TI
(Tra
nsfo
rmat
ion
Inte
nsit
y)
Scientific sub-topics
Case study: Artificial intelligence
The TE (Transformation Efficiency) on those sub-topics on artificial intelligence
0.000
0.200
0.400
0.600
0.800
1.000
1.200
1.400
1.600L
1-9
L1-
7
L1-
4
L1-
1
L1-
11
L1-
8
L1-
12
L1-
3
L1-
6
L1-
2
L1-
10
L1-
14
L1-
13
L1-
5
L2-
13
L2-
12
L2-
9
L2-
7
L2-
4
L2-
6
L2-
10
L2-
3
L2-
8
L2-
1
L2-
5
L2-
2
L2-
11
L3-
18
L3-
1
L3-
2
L3-
3
L3-
7
L3-
5
L3-
9
L3-
10
L3-
12
L3-
13
L3-
16
L3-
8
L3-
15
L3-
4
L3-
11
L3-
6
L3-
14
L3-
17
TE
(T
rans
form
atio
n E
ffic
ienc
y)
Scientific sub-topics
1990-2000 2001-2010 2011-2017
A framework was proposed to forecast the latent TO (technology opportunity) basedon the coupling relations between publications and patents, and the case study onartificial intelligence was conducted.
Conclusion & Discussions
However, the generalization capability of this proposed framework stillneeds more cases and verifications in the following studies.
In case study, this paper partially addresses some interesting phenomenon on artificialintelligence.
Thank you!