bs_bs_banner

An SAO-based text-mining
approach for technology
roadmapping using patent
information
Sungchul Choi1, Hongbin Kim1, Janghyeok Yoon2,
Kwangsoo Kim1 and Jae Yeol Lee3
1
Technology Strategy Group, Samsung Advanced Institute of Technology, San 14-1, Nongseo-ri,
Giheungeup, Yongin, Gyeonggi 446-712, Republic of Korea. [email protected]
2
Department of Industrial Engineering, Konkuk University, 120 Neungdong-ro, Gwangjin-gu, Seoul
143-701, Republic of Korea. [email protected]
3
Department of Industrial Engineering, Chonnam National University, 77 Yongbong-ro, Buk-gu,
Gwangju 500-757, Republic of Korea. [email protected]

A technology roadmap (TRM) links technologies with a company’s strategic objectives and
so supports acquisition of required technologies in advance of needs. It is a powerful tool for
strategic planning and technology management. Because technology is changing rapidly and
market competition is fierce, the role of a TRM is becoming increasingly important. To
support the role of a TRM, many firms and governments that use roadmapping are becom-
ing interested in reducing costs while retaining objectivity during TRM development. One
suggestion to achieve these goals is to use the keyword-based quantitative approach to
creating a TRM, but the information provided by the approach is limited because of the
characteristics of keyword information. To solve this limitation, this research uses the
concept of ‘function’ to support quantitative analysis for developing a TRM. The concept of
function can provide information on the uses and purposes of a technology. To represent a
function, a subject–action–object structure is commonly used. The suggested approach
allows research and development (R&D) managers to extend the views of product and
technology during development of a TRM. In addition, by reducing the time required to
develop a TRM, the proposed approach supports quick and accurate decision-making by
R&D managers.

1. Introduction cant influence on changes. Since 1960, due to the

impact and importance of technologic innovation,

S chumpeter argued that the growth of a firm or

nation depends on innovation (Schumpeter and
Opie, 1934), including innovation in technology,
predicting future technology and anticipating market
needs for a predicted technology have been the main
considerations. Since the beginning of the 1990s in
marketing, and organization. Historically, however, particular, as companies’ research and development
technological innovations have had the most signifi- (R&D) activities have been combined with corporate

52 R&D Management 43, 1, 2013. © 2012 The Authors. R&D Management © 2012 Blackwell Publishing Ltd,
9600 Garsington Road, Oxford, OX4 2DQ, UK and 350 Main St, Malden, MA, 02148, USA.
 An SAO-based text-mining approach for technology roadmapping

strategy, establishing the direction of R&D invest- cate how they affect other technologies, or identify
ment according to strategic goals and market needs the purpose of technology keywords. This limitation
has become increasingly essential. Accordingly, restricts the value of the output of keyword-based
extensive research on strategic R&D planning has TRM. For example, Tseng et al. (2007) analyzed the
been conducted during the past decade (Miller and keywords CYLINDRICAL LENS ARRAYS and
Morris, 1999). associated terms such as BUS BAR and SPATIAL
Recently, to support strategic R&D planning, tech- PHASE; although their analysis revealed how often
nology roadmaps (TRMs) have been widely used by the keywords are related, it could not express exactly
R&D organizations (Phaal et al., 2003). A TRM has how they are related.
the advantage that it enables developments in tech- To solve this problem, this research uses the
nology to be mapped and linked to product evolution concept of ‘function’ to support quantitative analysis
and market opportunities (Phaal et al., 2003). TRMs when developing a TRM. ‘Function’ is defined as
are extensively applied by firms and nations for ‘the action changing a feature of any object’ (Savran-
strategic planning, and for linking technology and sky, 2000). The concept of function can provide
product developments to business goals and market information on the uses and purposes of a technol-
opportunities (Garcia and Bray, 1997). Many organi- ogy. To represent a function, a subject–action–object
zations use TRMs to develop new products and tech- (SAO) structure is commonly used.
nologies (Phaal et al., 2001; Vatananan and Gerdsri, In this paper, we propose an SAO-based method of
2010). developing TRMs, called SAO-TRM, that uses SAO
Generally, TRMs have been developed by utilizing structures and patent information to support effective
a qualitative approach that depends on experts in the decision-making during technology planning and to
target technology. Technology analysis based on eliminate the limitations of the keyword-based TRM
experts’ knowledge has numerous advantages, but it approach. To this end, we propose a new TRM format
entails some limitations such as high cost, complex- that includes a function layer to utilize function infor-
ity, and inconsistency (Kostoff, 1998). For example, mation for developing TRM. In addition, we suggest
traditional TRMs generally tend to overstate qualita- product–function–technology (PFT) maps to support
tive and expert-dependent knowledge rather than decision-making while developing TRMs. Finally,
incorporating quantitative and objective information we explain how to interpret the proposed maps and
(Lee et al., 2008). In addition, processes used to how to apply the interpretation to the process of
produce TRMs often require vast amounts of infor- constructing TRMs using the new format.
mation and also have considerable costs in terms of The remainder of this paper consists of five sec-
time and human resources (Albright and Kappel, tions. In Section 2, we describe work related to the
2003). Furthermore, firms often find that TRMs are proposed approach. In Section 3, we provide a
too expensive (Kostoff et al., 2004). detailed description of our approach. In Section 4, to
To remedy this problem, researchers have tried to illustrate the proposed approach to developing TRM,
develop TRMs by applying quantitative analysis we analyze an example case: polymer electrolyte
techniques using a keyword-based approach. The membrane (PEM) technology in proton exchange
keyword-based approach utilizes text-mining tech- membrane fuel cells (PEMFCs). Finally, in Section
niques to analyze technology information. Using 5, we present concluding remarks and directions for
text-mining techniques, the keyword-based approach further study.
extracts important keywords that represent the
important contents of documents and discovers pat-
terns that contain technological implications. The
2. Related work
keyword-based approach improves the efficiency
and effectiveness of the roadmapping process. For
2.1. A keyword-based technology
example, developing a system to support visualizing,
roadmapping
documenting, disseminating, and updating TRMs
may increase roadmapping efficiency, and providing The main objective of a TRM is to offer a framework
meaningful information to support systematic to visually integrate market, product, and technology
decision-making may increase roadmapping effec- evolution based on time series structures (Garcia and
tiveness (Lee et al., 2008). Bray, 1997; Kostoff and Schaller, 2001; Petrick and
Although the keyword-based approach to TRM Echols, 2004). This framework can be used for gath-
development can overcome the limitations of previ- ering information from a wide variety of sources to
ous qualitative approaches, it cannot represent how develop dynamic short-, mid-, and long-term plans
technologies are used in their technology area, indi- for R&D (Petrick and Echols, 2004). To support

© 2012 The Authors R&D Management 43, 1, 2013 53

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

R&D planning, most TRMs are represented as three- technological information and TRM process than did
layered charts that represent strategic plans in terms previous research.
of the evolution and development of technology, of Lee et al. (2008) suggested a keyword-based TRM
products, and of markets (ERIMA, 1997). method. To increase roadmapping effectiveness to
The purpose of a keyword-based TRM method is improve TRM applicability, they proposed keyword-
to provide objective and quantitative information to based product–technology maps from which objec-
support strategic decision-making. To this end, the tive and quantitative information can be derived.
method uses text-mining techniques to analyze tech- They suggested how to apply implications of the
nological documents and to generate various types proposed map to the development of TRMs. Lee
of information including technological implica- et al. (2009) proposed a technology-driven roadmap-
tions. Generally, the target source, which is used for ping process that starts with capability analysis for
analyzing information, is based on a technical docu- technology planning and ends with business oppor-
ment such as a patent, so the keyword-based TRM tunity analysis for market planning. Using patent
method has been utilized to develop a technology data, this research suggested three maps for develop-
driven roadmap (Lee et al., 2009). Since text-mining ing a TRM: an actor–relations map, a technology–
received attention as a useful technique for technol- industry map, and a technology–affinity map.
ogy analysis, the research of keyword-based TRM Yoon et al. (2008a) proposed a new roadmapping
has been actively studied. The details of related work approach that uses systematic approaches to analyze
are as follows. quantitative data obtained from readily available
Lee and Jeong (2008) conducted research regard- sources. This method is an attempt to overcome some
ing methods to support the process of using co- of the existing limitations of qualitative roadmapping
word analysis to construct TRMs. They used stra- processes. Text mining is utilized to pre-process the
tegic diagram (Callon et al., 1991) to develop the raw data, which is in the form of natural language
TRM of robot technology in Korea. Suh and Park and is therefore difficult to input directly into the
(2009) proposed a patent map and service-oriented morphology analysis (MA) method (Zwicky, 1948)
TRM (SoTRM) using the patent map. SoTRM that is the principal approach used for developing
helps decision makers from public and private TRMs. The proposed MA-based methods can con-
sectors to select and concentrate on the most impor- tribute to improving the process of developing road-
tant services and related technologies in R&D of maps, by utilizing large volumes of product and
the service industry. For developing a patent map, technology data that might not be accessible other-
Suh and Park (2009) considered five values of wise. Yoon et al. (2008b) identified characteristics of
keyword information: level, time, relative growth, data that can be potentially used in roadmapping and
relative share and relative order. Li (2009) con- presented an approach to extracting important infor-
ducted a case study based on both qualitative and mation from such raw data by applying various data
quantitative data, by explaining how Cisco Systems mining techniques including text mining, multidi-
has been so successful in utilizing its strategy of mensional scaling, and k-means clustering. Moreo-
mergers and acquisitions for corporate growth based ver, Yoon et al. (2008b) explained how this approach
on a business ecosystem, especially from a techno- can be applied in each step (preliminary, mapping,
logical perspective. Li (2009) used text-mining and managing) of roadmapping.
techniques to analyze Cisco’s technology ecosys-
tem. Finally, Lee et al. (2011) proposed a formal
2.2. An SAO-based technology analysis
concept analysis (FCA)-based approach to develop-
ing a dynamic patent lattice that can analyze The concept of SAO structure originated from the
complex relations among patents and monitor theory of inventive problem solving (Russian
trends in technological changes. FCA is a math- acronym: TRIZ), which describes patterns for
ematical tool for grouping objects with shared prop- solving technical problems. Genrich Altshuller uti-
erties; it is based on lattice theory. lized the concept of function and developed TRIZ
Keyword-based TRM method is being actively by generalizing and abstracting technologies after
researched by Lee et al. (2008, 2009) and Yoon et al. extensive analysis of 200,000 patents, then used the
(2008a,b). Whereas previous research focused functional approach to suggest technology evolution
mainly on using text-mining techniques to find tech- patterns (Altshuller, 1984). An SAO structure is
nological implications from patents, Lee and Yoon based on the functional approach used in TRIZ.
each considered how to use extracted technological An SAO structure allows ‘key concepts’ to be rep-
implications for developing TRM, i.e., their research resented, rather than ‘key words’ (Cascini et al.,
focused more on the tight coupling between analyzed 2004) and can express means–end relationships

54 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

(Moehrle et al., 2005). An SAO structure can be how to use the SAO structures for technology analy-
organized in a problem–solution format if the action– sis. However, SAO structures can provide more tech-
object (AO) states the problem and the subject (S) nological information than can keywords. Therefore,
forms the solution (Moehrle et al., 2005). For we think that SAO structures can provide a range of
example, an SAO structure can simply represent the technology information and can be useful for devel-
function of a battery as ‘battery energizes bulb’. In oping TRMs.
this example, ‘battery’ is the subject, ‘energizes’ is
the action, and ‘bulb’ is the object. The technology
2.3. Patent information as a source
purpose of ‘battery’ is to energize the ‘bulb’, and the
of TRMs
function of ‘battery’ is ‘to energize the bulb’.
Many researchers have attempted to determine how In this paper, we use patent information as a resource
to use the SAO structure for technology analysis. for technology analysis. Patents are useful for esti-
Cascini et al. (2004) developed PAT Analyzer, which mating the status and trends of emerging technolo-
uses SAO models extracted from patent documents to gies (Archibugi and Planta, 1996). Patents also
search for patent contents. PAT Analyzer allows the contain valuable up-to-date technological informa-
search for ‘key concepts’instead of ‘keywords.’Using tion and provide trusted, objective information with
the function information extracted using PAT Ana- reference to R&D (Fleisher and Bensoussan, 2003;
lyzer, Cascini conducted a research to search for TRIZ Wanner et al., 2008). For these reasons, patent infor-
(Theory of inventive problem solving) contradictions mation has been widely used to identify technologi-
(Cascini and Russo, 2006) and to measure patent cal trends and to help formulate technology planning
similarity by comparing functional trees of inventions (Narin, 1993; Reitzig, 2004). Recently, patent
(Cascini and Zini, 2008). mining, which is a technique for systematic survey
Another approach is to use the SAO structure to and analysis of patent document to achieve corpora-
analyze patent similarities. Moehrle et al. (2005) tions’ strategic technology objectives, has been
suggested a method that uses inventor profiles to researched in technology management area.
support human resource decisions, and Bergmann Due to the characteristics of patent information,
et al. (2008) used patent similarities to analyze the all mentioned research regarding development of
risk of patent infringement. Yoon and Kim (2011) keyword-based TRM, and SAO-based technology
proposed a patent network based on semantic patent analysis used patent mining approach for technology
analysis using SAO structures. They suggested new analysis. Especially, because patent information is
indices to identify the technological importance of managed as a huge database and has well-defined
patents, the characteristics of patent clusters, and the bibliographic data, this information is easy to access
technological capabilities of competitors. Recently, and process for computer-based technology analysis.
some research has focused on how to use SAO struc- The contents of a patent document can be grouped
tures to improve the accuracy of comparison methods into two categories. The first group includes struc-
to evaluate patent similarities (Sternitzke and Berg- tured items called bibliographic data (such as title,
mann, 2009; Moehrle, 2010). patent number, issued date, or assignees), and the
SAO structures can also be applied to network second group is composed of unstructured items that
analysis techniques. Choi et al. (2011), suggested are descriptions of the invention. In this paper, we use
using SAO network analysis of patents to identify structured data for time series and assignee analysis,
technology trends. They presented a procedure that and unstructured data for SAO structure extraction.
formulates an SAO network and a method that
applies actor network theory to analyze technology
implications of the SAO network. 3. An SAO-based approach to
SAO structures can be used to provide information technology roadmapping
for the development of TRMs. The Samsung
Advanced Institute of Technology developed TRMs In this chapter, to suggest an SAO-TRM, we propose
using technological functional information defined as using the SAO-based TRM format, the PFT map, and
similar SAO structures. Yoon et al. (2011b) also sug- an SAO-based TRM procedure that uses the pro-
gested a framework for technology reuse by utilizing posed map.
functional models to define modulated technology;
the format of their suggested framework is similar to
3.1. SAO-based TRM format
TRM format.
Research into technology analysis using SAO The basic TRM type is a three-layered chart that
structures is in its early stages; it must still consider contains time-based information (ERIMA, 1997).

© 2012 The Authors R&D Management 43, 1, 2013 55

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

Figure 1. (a) A subject–action–object (SAO)-based TRM format and (b) TRM example. MEA, membrane electrode assembly.

Table 1. The layers of the SAO-based TRM format We give a simple example of SAO-TRM regarding
Layer type Definition membrane electrode assembly (MEA) technology in
fuel cells. A MEA consists of two components:
Product The layer including products and product catalyst and membrane (Figure 1b). Therefore, the
components for technology activities development technology of MEAs can be simply
that can be procured in the market,
including services. divided into catalyst formulation technology and
membrane composition technology. The main
Function The layer including functions for the
development purposes of technologies purpose of catalyst formulation technology is to
and products. increase conductivity to increase the efficiency of
Technology The layer including technologies required MEAs; this technology is related to catalyst products.
for the design, manufacture and test The main purpose of membrane composition tech-
of a product, including all related nology is to increase the membrane’s mechanical
processes, manufacturing processes or strength; this technology is related to the durability of
methods.
MEA products.
Definitions are referred to Guglielmi et al. (2010).

3.2. SAO-based PFT maps

This three-layer type TRM consists of market, In this section, we propose PFT maps to support
product, and technology information. Of these three construction of an SAO-TRM. The proposed PFT
layers, the product and technology layers are the consists of three maps: a PFT portfolio map, a PFT
most useful for product planning (Phaal et al., 2001; relationship map, and a PFT deployment map. The
Lee et al., 2008). In this research, we focus on these PFT portfolio map provides the information regard-
two layers and add a ‘function’ layer to the suggested ing which technological factors (products, functions,
SAO-based TRM format. and technologies) are currently important. To deter-
The SAO-TRM format suggested in this paper mine these factors, we consider the absolute number
(Figure 1a) consists of three layers: product, func- and rates of increase of occurrences of PFT terms.
tion, and technology (Table 1). By inserting a func- The PFT relationship map represents relationships
tion layer, the suggested SAO-TRM format more among factors; network analysis techniques are used
clearly represents the relationships among products to analyze these relationships. The PFT deployment
and technologies than does the existing TRM format. map clearly defines the associations between the
The items of the layer are related to the technologies purpose of a product and the direction of technology
and present their purposes. In addition, the items of development; for this map, we use the format of the
the function layer are related to the products. This quality function deployment (QFD) method (Dale,
connection shows the characteristic of the target 2003). The proposed maps are developed by referring
products. By presenting the information that is not to the keyword-based product–technology map (Lee
provided in the exiting TRM format, the SAO-TRM et al., 2008) and the technology-tree-based method
format should help decision-making in strategic of constructing TRMs (Cheong, 2006).
R&D planning. The following is a description of To build these maps, SAO structures must be
each layer. extracted from patents and tagged to represent

56 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Figure 2. The product–function–technology portfolio map.

technological factors (product, function, or technol- technological attention in the past. Because they may
ogy). Pre-processing is described in detail in Section also have high developmental potential, many com-
3.3. panies invest heavily in them.
Established technological factors have high fre-
quencies, but low rates of increase of occurrence.
3.2.1. PFT portfolio map These factors may have already been sufficiently
The main function of the PFT portfolio map is to developed and implemented to satisfy the required
represent the importance of technological factors. performance.
For this, we assume (1) that if technological factors Declining technological factors have low rate of
related to particular fields appear frequently in increase and low frequency of occurrence. These
patents, then that factor is receiving significant inter- factors likely have not been of interest since the early
est among innovators in the relevant technology, and stage of R&D. This type is of no interest in the
(2) that related R&D activities are likely to be in technological field.
progress (Lee et al., 2008). The technological factors With the PFT portfolio map, an individual techno-
are divided into four groups (Figure 2a) based on logical factor is analyzed using the technological
two criteria computed as relative levels: the absolute factor life-cycle map (Figure 2b), which represents
number of occurrences and the increase in frequency the life cycle of an individual technological factor
of occurrence. using its number of occurrences and the rate of
Emerging technological factors have relatively low increase of occurrences per specific period of time.
frequencies, but high rates of increase in frequency of The life cycle consists of four phases: emergence,
occurrence. This kind of technological factor is very growth, maturity, and decline. As a technological
likely to have not been of interest in the past, but to be factor progresses through these phases, its total
actively developed at the time of analysis. number of occurrences increases, but its rate of
Core technological factors have both high frequen- increase in occurrences per period decreases, except
cies and high rates of increase in frequency of occur- during the emergence phrase. In the decline phase,
rence. The technological factors belong to the core the number of occurrences per period decreases. The
type are very likely to have attracted considerable product, function, and technology maps provide

© 2012 The Authors R&D Management 43, 1, 2013 57

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

the QFD method (Dale, 2003). In the same way that

QFD is used to connect user demands to measures
of manufacturing products, the proposed map is
used to support the process of using function infor-
mation to connect products and technologies. The
proposed map consists of three matrices: effect by
purpose, technology by effect, and purpose by
product; these matrices represent the function
relationships, and technology operating principles,
and purposes of products, respectively. Each matrix
is built using the results of analyzing SAO
structures.
Figure 3. The product–function–technology relationship map. To construct this map, we divide functions into
two types: purpose and effect. A purpose function
represents reasons for development of technologies
information regarding the overall change of the tech- and products; an effect function represents how tech-
nological factors; the life-cycle map represents the nologies or products affect other materials or techno-
change of the individual factor over time. logical attributes. Generally, the effect function
Based on this analysis, the proposed map helps provides the technology operation principle.
decision makers who are developing a TRM; it does Using functions to represent how to connect prod-
so by providing answers to three questions: (1) What ucts to technologies, this map supports three techno-
is the important purpose of products and technolo- logical decision-making processes: (1) Discovering
gies that are developing in certain period? (2) Which new technology opportunities, (2) selecting technol-
technological factors may not be considered for ogy indicators, and (3) establishing product develop-
future development? and (3) Which products and ment direction.
technologies should be developed in the future and First, the effect-by-purpose and technology-by-
for what purposes? effect matrices are utilized to discover new technol-
ogy opportunities. In the effect-by-purpose matrix,
an R&D manager should check which effect func-
3.2.2. PFT relationship map tions can be utilized to accomplish purpose func-
The PFT relationship map visually presents informa- tions. If the effect functions are not utilized for the
tion regarding the relationships among technological purpose functions despite the possibility that they
factors. The format of this map is similar to that of can be utilized, the R&D manager considers the
an SAO-TRM, which uses a network graph to repre- technology-by-effect matrix to check technologies
sent how products, functions, and technologies are related to the effect functions. Then, the R&D
related (Figure 3). To this end, we use SAO struc- manager adds the technologies as candidates for
tures extracted from patent document to find techno- development if the technologies can be utilized to
logical factors to be used as nodes of a network. In an support the effect functions.
SAO structure, generally, a product or technology is Second, the purpose-by-product matrix is utilized
denoted as S, and a function is denoted as an AO pair. to select technology indicators. An R&D manager
To build the network, pre-processing should be con- can identify the purposes of the products using this
ducted using natural language processing (NLP) matrix, which represents the relationship between
techniques (Section 3.3). products and purpose functions. The R&D manager
The formalized network including the PFT rela- uses the purpose functions to categorize the purposes
tionship information is analyzed using network of the products then uses the measure methods to
analysis techniques, which yield the following tech- categorize the technology indicators. For example,
nological implications: which technological factors ‘reduce thickness’ and ‘improve stability’ are the
significantly affect other factors; which functions are purpose functions of the PEM technology of
related to important products or technologies; and PEMFCs (Choi et al., 2011). To accomplish ‘reduce
why the important products were developed. thickness’, the micrometer can be used as the
measure. To accomplish ‘improve stability’, two
kinds of stability must be considered: stability of
3.2.3. PFT deployment map high-temperature conductivity and mechanical
The last map is the PFT deployment map stability. The transition temperature can be used as
(Figure 4). The proposed map is developed using the measure of the stability of high-temperature

58 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Purpose Effect
Technologies
functions functions

functions

functions
Products

Purpose

Effect
Purpose by product Effect by purpose Technology by effect
matrix matrix matrix

Figure 4. The product–function–technology deployment map.

Figure 5. Overall process of developing subject–action–object (SAO)-based technology roadmap (TRM). PFT, product–function–
technology.

conductivity, and the result of a tearing test can be cators. Similarly, the R&D manager considers the
used as the measure of the mechanical stability. technology development priorities by only checking
Finally, the R&D manager utilizes the information the effect functions related to the purpose functions
obtained from analysis of the effect-by-purpose that affect the technology indicators, if suitable can-
and technology-by-effect matrices to establish the didate technologies are not available.
product development direction. The R&D manager
can prioritize technology development options by
considering the candidate technologies that need to
3.3. The SAO-based TRM procedure
be developed, with the selected technology indica-
tors. Because the technology indicators are related to The procedure for developing SAO-based TRM is
the purpose functions and the candidate technologies composed of five steps (Figure 5): (1) collecting
are related to the effect functions, the R&D manager patent data source, (2) extracting SAO structures
can check the relationship between the purpose and using syntactic analysis of patent text, (3) identifying
the effect functions to decide which candidate tech- types of technological factors by analyzing extracted
nologies are related to the selected technology indi- SAO structures, (4) constructing PFT maps, and (5)

© 2012 The Authors R&D Management 43, 1, 2013 59

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

The SAO extraction example(from US5360745) 3.3.3. Identification of technological factors

The third step is identification of the type of techno-
Original Sentence:
The MG-Si powder is melted by the high-temperature
logical factors (product, technology, or function) by
plasma and sprayed onto the base substrate 11. analyzing extracted SAO structures. In SAO struc-
ture, subject and object each consist of a word
Extracted SAO Model: phrase. The phrases can be divided into four types:
high-temperature plasma - melt - MG-Si powder product, technology, material, and technology
attribute (Table 2). In this paper, we regard a material
Figure 6. An example of subject–action–object (SAO) extraction as a product and a technology attribute as a technol-
(patent US5360745). ogy. Therefore, a word phrase represents a product
including a material, or a technology including a
technology attribute. The AO combination also rep-
resents a function.
developing an SAO-TRM based on the information To identify the types of a technological factor in
from the PFT maps. an SAO structure, we use an SAO linked list that
represents relationships among SAO structures
(Figure 7). In the case of a product or technology
3.3.1. Collection of patent documents type, generally, a product is manufactured by a tech-
The first step is collection of patent documents. To nology and comprised a product component. For
develop the TRM, the proposed method uses a spe- example, chemical vapor deposition is a technology
cific patent set. Gathering of patents requires a patent for manufacturing amorphous silicon thin film, and
retrieval query that is composed of textual informa- amorphous silicon thin film is a product component
tion related to a target technology, and bibliographic of a thin-film solar cell.
information such as International Patent Code, appli- To accurately identify the type of a word phrase, we
cants, and application date. After collecting patents, consider the action word that links the subject to the
irrelevant patents are eliminated by applying a filter- object. Based on the type of action, we can identify the
ing rule. type of word phrase used in the subject and the object.
This process requires a reference table that defines the
types of action (Table 3). In an SAO structure, if the
3.3.2. Syntactic analysis of patent text subject and object are word phrases regarding a
The second step is extraction of SAO structures from product type, they are linked by an action word such as
the collected documents. SAO structures can be include, comprise, or compose. If the subject is a
extracted from any description in textual format product or technology type, and the action word is
including the abstract, claims, and description sec- ‘intensify’ or ‘increase’, then the word phrase that
tions of a patent document. Because an abstract is describes the object is highly likely to be a type of
precise and has been regarded as the most meaning- technology attribute. By referring to Table 3, the types
ful part (Chen et al., 2003), most existing technology of word phrases can be manually identified.
analysis research using SAO structures generally The next task is to use AOs to identify types of
extracts SAO structures from abstracts only (Cascini function. The first type is a purpose function; this
et al., 2004; Moehrle et al., 2005; Bergmann et al., type represents the purpose of the subject word. This
2008), but in this paper, we used the full text of type combines a technology attribute, which can be
patents. The abstract mainly describes important measured by quantitative estimate, and the related
components of an invention, but it cannot describe its action word (Table 4). The second is an effect type;
purpose or provide technological details. The full text this type represents how the subject word affects the
of a patent document includes information concern- object word. This type is constructed as the combi-
ing ‘problems to be solved’ (Tseng et al., 2007), and nation between related action word (see Table 4) and
this information can represent technology details. any type of key phrase.
Thus, to extract necessary SAO models, we consider Although not considered as the source for an SAO-
all parts of patent documents. TRM, a partative type also exists. The AO of this
For automatically extracting SAO structures using function type contains an action word that represents
an NLP parser, we use a commercial linguistic parser, an inclusion relationship (Cascini et al., 2001) and an
KnowledgistTM 2.5. (Invention Machine, Boston, MA, object word regarding a product or a technology. This
USA; http://www.invention-machine.com). Using type of function is only used to identify the techno-
NLP technology, we can extract SAO structures logical factor type of a word phrase that represents a
(Figure 6). product or technology.

60 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Table 2. The types of a keyword phrase

The types of keyword Definition Example keyword phrases
phrase

Product Products or product components are all items needed for Amorphous silicon thin film,
technology activities that can be procured in the market, PEMFC
including services (Guglielmi et al., 2010)
Technology A technology is defined as the technical know-how that is CVD, MOCVD
required for the design, manufacture and test of a space
product, including also all related processes,
manufacturing process, or method (Guglielmi et al.,
2010)
Material Materials that are not only belong to specific technologies Water, gas, air, powder,
or products, used in various area (Hirtz et al., 2002). For oxygen, hydrogen
example, gas, liquid, solid, plasma, mixture
Technology attribute The performance measurement of technology or product Conductivity, thickness of film

type of a function is identified and manually modified

as needed.

3.3.4. Construction of PFT map

3.3.4.1. The PFT portfolio map. To build the PFT
portfolio map, the occurrence of technological
factors per time of period is calculated. The portfolio
map consists of two axes: rate of increase and abso-
lute size. Rate of increase is defined as follows:

Rate of increase ( F , T )

=
∑ Occurrence ( F, T ) − ∑ Occurrence ( F, T − 1) ,
∑ Occurrence ( F, T − 1)
(1)
Figure 7. Example of a subject–action–object (SAO) linked list
for identifying the type of a word phrase. where F is a technological factor, T is a period of time,
SOccurrence(F, T) is the total number of occurrences
of F during T, and T–1 means the preceding time
Table 3. The examples of action reference table for period. The rate of increase represents how the occur-
identifying the types of a word phrase
rence of a technological factor changed between a
Subject Example action words Object specific time period and the preceding time period.
Rate of increase (F,T) and SOccurrence(F, T) are used
Technology Manufacture, produce, Product
form, supply as the axes when building the technological factor
life-cycle map (Figure 2a).
Product Include, have, composed Product
of, comprise, supply
Technology Use Materials 3.3.4.2. The PFT relationship map. To build the
PFT relationship map, we use co-occurrence of tech-
Product Convey, supply –
nological factors in an SAO structure. To represent
Technology Intensify, Technology
the relationship as a network, a product and function
Product Increase, measure, Attribute
stabilize
are linked if they exist as an S and AO in one SAO
structure. Similarly, a technology and function are
linked if they exist in one SAO structure. A product
and technology are indirectly connected if they are
connected to the same function.
The process of identifying the types of a function For instance, from Choi et al. (2011) (Figure 8), we
depends on the type of combinations between the get four SAO structures. In the structures, the subject
action word and the type of word phrase used to ‘cross-linking’ is a technology for manufacturing
describe the object. Based on the combinations, the ‘MEA’ and ‘PEM’, which are product components of

© 2012 The Authors R&D Management 43, 1, 2013 61

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

Table 4. Types of AO
The type of AO Definition Example action words KP type of object

Purpose function AO representing what the purpose Stabilizes, Vibrates, intensify, Technology attribute
of subject word is increase, measure, stabilize.
Effect function AO representing how the subject Absorb, Accumulate, Cleans, Any types
word affects the object word Condense, Cools, Destroys,
Detects, Erodes, Evaporate,
Extracts, Freezes, Boil, Heats,
Holds, Mixes, Moves, Polishes,
Produces, Removes, Separates.
Partative type AO representing inclusion Include, have, composed of, Product technology
relationship between products or comprise, supply, assembles
technologies.
AO, action–object; KP, key phrase.

Figure 8. An example of constructing product–function–technology (PFT) relationship map using subject–action–object (SAO) structures.
Components are described in the text.

PEMFCs. ‘Cross-linking’ can increase conductivity same SAO structure; therefore, we use the indirect
and mechanical strength, and these two functions are relationship between a product and technology. In
also the purpose functions of ‘MEA’ and ‘PEM’. an SAO structure, a product is related to a purpose
Then, ‘MEA’ and ‘PEM’, which are products, indi- function, and a technology is related to an effect
rectly connect to the technology ‘cross-linking’ function. Therefore, if a product and technology are
through the functions ‘increase conductivity’ and related, we consider that the purpose function and
‘mechanical strength’. effect function that exist in one SAO structure with
the product or technology are also related. In this
3.3.4.3. PFT deployment map. Finally, to build the paper, we assume two cases, when the purpose func-
PFT deployment map, we also use co-occurrence of tion is indirectly related to the effect function, con-
technological factors in an SAO structure. Similar sidered as they are indirectly related. The first is to
to a PFT relationship map, a product-by-purpose indirectly connect a product and technology, which
matrix and a technology-by-effect matrix are con- are related to same functions. This case is same as
structed using the relationship between a product the indirect relationship between a product and
and purpose function or a technology and an effect technology, mentioned in the PFT relationship
function, respectively. To check the relationship, we map. Second case is to use a partitative relationship
verify that they exist in an SAO structure as an S between a product and a technology. If a product
and AO. In a purpose-by-effect matrix, the purpose and technology have a partitative relationship, we
and effect functions are represented as AO in an consider that they have an indirect relationship. The
SAO structure so that they cannot both exist in the purpose functions related to the product can also

62 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Figure 9. Subject–action–object (SAO)-based technology roadmapping process. PFT, product–function–technology; R&D, research and
development.

be connected to the effect functions related to the new product concepts identified during product plan-
technology. By considering these two cases, we con- ning. To identify a target technology, the first task is
struct a purpose-by-effect matrix. to identify technology indicators. The purpose-by-
product matrix of the PFT deployment map is uti-
lized to select technology indicators (Section 3.2.3).
3.3.5. Development of SAO-based TRM After finishing this task, the decision makers use the
Finally, we explain in detail how to utilize the pro- PFT deployment map to choose the technology
posed maps to generate a TRM (Figure 9). Generally, that affect the indicators, then set the technology
the process of developing a TRM proceeds from development priorities. When prioritizing the target
product planning to R&D planning. In product plan- technology, the emerging and core technologies,
ning, decision makers set the direction of product provided by the PFT technology layer of the PFT
development. To this end, the three types of PFT map portfolio map, are also considered to help identify
can be utilized. First, the product layer of the PFT target technologies.
map portfolio shows the product trends in emerging In this step, the PFT relationship map is utilized to
and core product areas. The products in these areas help make ‘make-or-buy’ decisions for each technol-
are likely to be important or trendy in the market. The ogy by comparing between the own technologies
second approach is to utilize function information. and the competitors’ technologies. Using bibliomet-
The function layer of the PFT portfolio map identi- ric information, the PFT relationship map can
fies emerging and core functions. Because the rela- provide information regarding the major technolo-
tionships between products and functions can be gies of each company. After making the decisions, if
observed by checking the PFT relationship map, the technologies must be developed in-house, the
product development directions can be identified PFT relationship map helps the decision makers set
by considering the function information. Lastly, the development strategies such as whether technologies
PFT deployment map can be also utilized to identify should be developed concurrently. In addition, the
new product concepts. This development map is used decision makers also consider whether technology
to identify the product’s development directions. can be reused if the company owns a technology that
Because the directions from the PFT development provides similar functions (Yoon et al., 2011b).
map define the new product concept based on the
possibility that new technology will appear, this can
be utilized to file patents in advance of specific tech-
4. SAO-TRM: a case study of PEMFC
nological development. Among the three approaches,
the decision makers choose the approach that is most
4.1. Technology overview
suitable for their objectives in developing a TRM,
and determine the product concept by considering To illustrate the proposed approach, we used it to
market trends. develop an SAO-TRM regarding PEMFC technol-
After completing product planning, we proceed to ogy. PEMFCs constitute a next-generation energy
technology planning, to find technology related to the technology; they are being developed for applica-

© 2012 The Authors R&D Management 43, 1, 2013 63

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

Figure 10. Structures of a fuel cell stack and membrane electrode assembly (MEA). PEM, polymer electrolyte membrane.

tions in transport, and to stationary and portable cells. Table 5. Patent retrieval query
The core component of a PEMFC is a cell stack Patent retrieval query Collected
consisting of repeated structures of an MEA and a patents
bipolar plate (Figure 10).
A MEA contains a PEM sandwiched between an PEM (membrane* or (electrolyte* and 688
membrane*) or film).ti. and
anode plate and a cathode plate. The plates each ((fuel near cell) or (PEMFC
includes a catalyst layer surrounded by gas diffusion or DMFC)) and (polymer* or
layers. In PEMFCs, when hydrogen reacts electro- proton* or molecule or
chemically with oxygen, chemical energy is gener- hydrogen* or macromolecule
ated; PEMFCs transform this chemical energy to or ionomer) and @AD>=
19910101<=20110531
electrical energy. A stream of hydrogen is delivered
to the anode side of the MEA, where a catalyst splits *Zero or more repetitions of the expressions.
the atoms into protons and electrons.
In PEMFCs, the MEA is a key component, and
the key technological challenge of PEMFCs is to 4.2. Data collection
develop a PEM for ion exchange. Therefore, in this
research, we develop an SAO-TRM that focuses on As a first step, US patents relating to PEMFC tech-
PEM technology and concomitantly analyzes other nology were collected from the WIPS database
technologies such as a catalyst electrode, gas diffu- (http://www.wips.co.kr). The collected patents were
sion layers, bipolar plate, and MEA. published from January 1, 1991 to May 31, 2011.
Qualitative patent analysis of PEM technology has After removing irrelevant patents, 688 patents were
already been performed by the Korean Intellectual used to formulate and analyze an SAO-TRM
Property Office (KIPO). We choose the patent set (Table 5).
from the report, compare the result obtained by
applying the proposed approach to those obtained by
4.3. SAO processing
Korean Intellectual Property Office, and use a tech-
nology tree of the relevant patents to test whether After selecting the patent set, we used SAO structures
new technological findings can be added. to utilize the sources of an SAO-TRM. First, we used

64 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Table 6. Numbers of patents and SAOs, by time period

Period 1987–2001 2002–2004 2005–2006 2007–2010

Number of patents 143 203 169 173

Number of SAOs 2,605 4,543 4,495 4,588
SAO, subject–action–object.

Figure 11. Product portfolio map of PEFMC technology.

Knowledgist™ 2.5 to extract SAO structures after their frequency of occurrence changes. The rate of
syntactic analysis of the full text of patent documents. increase of technological factor occurrence from
The syntactic analysis yielded 94,762 SAO struc- period 3 to period 4 (y-axis) is plotted against the
tures; 16,231 of these remained after duplicated, and absolute number of occurrences in period 4 (x-axis).
irrelevant SAO structures were deleted. In these Patents in each map are divided into four groups
remaining SAO structures, 1,403 word phrases according to the average value of rate of increasing
remained after the categorization step. Among the and absolute numbers.
remaining word phrases, 983 were product types First, we examine the product portfolio map
(including materials), and 420 were technology types regarding membrane technology (Figure 11). The
(including technological indicators). Of the AO com- map clearly shows that ‘membrane’, ‘fuel’, ‘MEA’,
binations, 4,312 remained after categorizing; 1,418 ‘catalyst’, and ‘PEMFC’ are in the core product area.
were partative type, 2,308 were effect type, and 586 These product words are generally used in the PEM
were attribute type. technology. In detail, ‘platinum alloy’ is located in
To analyze changes in patent information over the Core area and can be considered as an important
time, patents were further classified into four time a product or product material. In fact, platinum is
periods: period 1 from 1987 to 2001, period 2 from used in membrane technology as a catalyst to
2002 to 2004, period 3 from 2005 to 2006, and period increase rates of a chemical reaction. The interesting
4 from 2007 to 2010 (Table 6). point of this figure is that it can compare the types of
polymer used for source material in membrane tech-
nology. Two kinds of polymer type are used:
4.4. PFT maps perfluoro-based polymer and carbon-based polymer.
4.4.1. PFT portfolio map In the map, ‘fluorinated compound’ and ‘sulfonic
The axes in the portfolio map are the absolute acid group’, which are in the position of core area
number of keyword occurrences and the rate at which with extremely high frequency, are types of

© 2012 The Authors R&D Management 43, 1, 2013 65

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

Figure 12. Function portfolio map of PEFMC technology.

perfluoro-based polymer. In fact, these polymers are actively executed at present. In addition, the func-
considered to be important materials in membrane tions related to physical durability, such as ‘have
technology. In addition, some word phrases regard- thickness (AO32)’ and ‘have mechanical strength
ing perfluoro-based polymer are located on the (AO19)’, are also located in the core area. Physical
core area; these include ‘Nafion(P31)’, ‘hydroxyl durability is likely to be considered as an important
group(P92)’, ‘phosphonic acid group(P168)’, and purpose of a technology development.
‘phosphoric acid(P169)’. In contrast, carbon-based Next, we observe the emerging area. This area
polymer materials are located in the emerging area contains functions related to heat and chemical sta-
or the core area near the emerging area. The materials bility. The following functions are belong to this
of this type are as follows: ‘polytetrafluoro ethylene type: ‘increase temperature (AO41)’, ‘limit tempera-
(P186)’, ‘polybenzoxazole polymer (P177)’, ‘per- ture (AO42)’, ‘increase stability (AO40)’, ‘have
fluorocarbon polymer (P161)’, ‘polybenzimidazole heat resistance (AO17)’, ‘have temperature (AO29)’,
compound (P176)’, ‘polyazole compound (P175)’, ‘have resistance (AO26)’, and ‘increase heat resist-
‘polymerizable monomer (P180)’, ‘perfluorocarbon ance (AO37)’. In particular, because the functions
compound (P160)’, ‘hydrocarbon structure (P85)’, related to heat stability have high rates of increase
and ‘polyarylethersulfone polymer (P174)’. To (> 0.5), they may become a core technology devel-
summarize, in PEM technology, perfluoro-based opment purpose in the future.
polymers demand careful consideration, and carbon- Finally, in the ‘decreasing’ area, we detect func-
based polymers are likely to be core products in the tions related to methanol crossover (MC) problems,
future. such as ‘have crossover (AO10)’ and ‘decrease MC
Second, we examine the function portfolio map (AO3)’. MC is a phenomenon in which methanol
regarding membrane technology (Figure 12). In the diffuses from the anode to the cathode through the
core area, we found functions related to membrane membrane without reacting. MC degrades the effi-
performance: ‘have conductivity (AO9)’, ‘have elec- ciency of PEMFCs, so preventing the phenomenon
trochemical conversion (AO15)’, ‘have performance is very important. The location of patents related to
(AO22)’, ‘have permeability (AO23)’, and ‘provide MC in the ‘decreasing’ area of the map suggests that
conductivity (AO44)’. In case of ‘conductivity’, to the MC problem has been solved.
improve the power efficiency of PEMFCs, improve- Finally, we examine the technology portfolio map
ment of PEM conductivity is regarded as a very regarding membrane technology (Figure 13). Tech-
important technology development theme. Viewed nologies related to manufacturing PEM occur in
from this perspective, we think that performance- the core area; these include ‘polymer chemistry
related functions are research objectives being techniques (T49)’, ‘cracking (T12)’, ‘crosslinking

66 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Figure 13. Technology portfolio map of PEFMC technology.

compound (T13)’, and ‘crosslinking reaction (T14)’. include ‘porous polymer (T50)’, ‘filling pores
Because these technologies are related to methods method (T31)’, and ‘diameter (T20)’.
of assembling membrane and electrode, they are
likely to be core technologies in membrane manu-
facturing. Technological indicators related to physi- 4.4.2. PFT relationship map
cal durability of membranes are located in the In the PFT relationship map (Figure 14), we use
established area; these include ‘thickness (T64)’, network analysis techniques to present how products
‘weight (T69)’, and ‘mechanical strength (T40)’. and technologies are related to functions. With this
This location on the map suggests that physical map, we can identify the objectives of products and
durability is a positively necessary technologic indi- technologies. The network analysis techniques used
cator in PEM technology. Technology related to heat in the map are degree analysis and centrality analysis.
stability is in the emerging area or the core area near In network theory, degree indicates the number of
it. For example, ‘heat resistance (T33)’ and ‘operat- links incident to a node (Diestel, 2010). The PFT
ing temperature (T45)’ are located in the emerging relationship map also expresses the number of rela-
area with a high rate of increase, and ‘heat treatment tionships between product or technology and func-
(T34)’ and ‘temperature (T60)’ are located in the tion. Centrality is an index that reveals the location of
core area. This figure suggests that heat stability has a given node in relation to the center of the whole
received much attention recently, and that the network. Various methods have been used to measure
current PEM technology satisfies the demand for centrality, and these methods may yield differing
physical durability to a certain degree. Technologies values. In this research, we use closeness-centrality
for spreading catalyst on membranes are also analysis. To compute closeness centrality of nodes,
located in the emerging area; these include ‘dispers- we compute short geodesic distances between nodes
ing solution (T24)’, ‘catalyst dispersion (T5)’, (Sabidussi, 1966). We accepted that a function with
‘coating (T10)’, ‘dispersing solution (T22)’, and high closeness centrality can be strongly related to a
‘CCM (T6)’. CCM is a recently developed spread- key or target use of a product or technology (Yoon
ing method; it shows high rate of increase. In addi- et al., 2011a).
tion, technologies related to porosity, which can We utilize a degree analysis to determine the
decrease weight of PEM and increase mechanical overall technological trends and to understand which
strength, are also located in the emerging area; these technologies or products are related to the important

© 2012 The Authors R&D Management 43, 1, 2013 67

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

Figure 14. Product–function–technology relationship map regarding PEFMC technology (1991–2011). Thickness of connecting lines
indicates the strength of the relationship.

functions. In addition, a centrality analysis shows Second, we use centrality analysis to examine the
how the core functions changed over time. Net- change over time of key or target functions of mem-
Miner® 3 (Cyram, Seoul, Republic of Korea; http:// brane technology (Figure 15). This analysis detects
www.netminer.com/) was used to analyze the four technological trends related with functions.
network information. First, functions related to conductivity have been
The PFT relationship network of membrane tech- considered core functions in all periods. The func-
nology from 1991 to 2011 is illustrated in Figure 14. tions, such as ‘have conductivity’ and ‘provide con-
Overall, the technological function in the portfolio ductivity’, show high centrality in period 1, and the
map has a high association with the product ‘mem- centrality of the function ‘increase conductivity’,
brane’. This fact means that many functions are which is a function directly related with the perform-
achieved by improving the membrane. ance of PEMFC, rose sharply after period 1. Second,
As another approach, we can find specific products functions related to physical durability have received
or technologies utilized to achieve specific functions. increasing attention. Specially, they show meaning-
For example, functions related to physical durability, ful growth in periods 2 and 3, and have higher cen-
such as ‘increase mechanical strength’ and ‘provide trality value in period 4 than do functions related to
mechanical strength’, are connected with ‘porous conductivity. Third, the centrality values of functions
structure’ and ‘cracking’. ‘Porous structure’ is a type related to MC are decreasing. In period 1, these were
of a membrane form. Because the meaning of the considered to be core functions but show a downward
function ‘have cavity’ is related to ‘porous structure’, trend after period 2. Finally, functions related to heat
a membrane with a porous structure increases con- stability have newly appeared. These functions firstly
ductivity by transporting ions throughout cavities, show up in period 2 and keep their positions in
but it maintains mechanical strength by keeping the periods 3 and 4.
thickness of the membrane. ‘Cracking’ of the catalyst
layer speeds deterioration of catalysts, decreases the
reaction of catalyst, and reduces the durability of 4.4.3. PFT deployment map
the membrane. The function ‘have heat resistance’ is In this case, we used a PFT deployment map to
related to ‘fluorinated compound’; this means that present effect functions that are high related to core
research into the heat stability of perfluoro-based purpose functions, and technologies that are related
polymer was conducted. ‘Platinum alloy’, a core to the effect functions. We assume that effect func-
material of catalyst, shows a strong relationship with tions that are related to technology affect the purpose
the function ‘limit temperature’. This fact indicates functions that are related to the product if a relation-
that ‘platinum alloy’ has a strong influence on the ship occurs between a technology and a product. In
operating temperature of PEMFCs. addition, the strength of the relationship between

68 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Figure 15. Centrality analysis of functions in PEM technology.

Figure 16. Part of purpose-by-effect matrix.

them increases with their frequency of co-occurrence divided into three technology categories: (1) mem-
in patent documents. brane manufacturing technology, (2) porous structure
PEM technology includes many purpose and effect technology, and (3) catalyst spreading technology.
functions; we chose important purpose functions The PFT deployment map reveals that the function
mentioned in the function portfolio map and the PFT ‘increase mechanical strength’ is highly related to
relationship map, and effect functions related to the manufacturing technology and porous structure
purpose functions (Figure 16). There are three technology. The function ‘increase conductivity’ is
purpose functions: ‘increase mechanical strength’, related to all technology considered. The matrix also
which is related to physical durability; ‘increase con- shows that the function ‘have heat resistance (or use
ductivity’, which is related to performance; and ‘have heat treatment)’ is influenced by membrane manufac-
heat resistance (or use heat treatment)’, which is turing technology and porous structure technology.
related to heat stability. These purpose functions are
associated with five effect functions: ‘comprise
4.4. TRM
crosslinking compound’, ‘use cracking’, ‘have
porous structure’, ‘fill porous structure’, and ‘com- In this section, we propose a TRM of PEM technol-
prise coating method’. These five functions can be ogy based on the proposed PFT maps. Because we

© 2012 The Authors R&D Management 43, 1, 2013 69

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

cannot clearly define how to use a TRM, we illustrate to physical durability, such as ‘improve mechanical
the TRM by considering the overall market situation strength’, are considered to core research fields, and
except R&D planning. Nevertheless, we expect the functions related to heat stability, such as ‘have heat
case to help to understand how to apply SAO-TRM resistance’, are recognized as emerging research
in R&D planning. All qualitative analysis of PEM objectives.
technology are from KIPO’s report (KIPO, 2006), In fact, improving heat resistance and mechanical
and we try to compare the results obtained using the strength are considered to be important technology
proposed method with the results of that report. objectives in PEM technology. Especially, because
carbon-polymer-based membranes have lower con-
ductivity than do perfluoro-polymer-based mem-
4.4.1. Market situation branes, relatively few studies of heat resistance and
Currently, public companies in the PEMFC industry mechanical strength in carbon-polymer-based mem-
are not profitable. However, the PEMFC industry is brane have been conducted. However, as operating
growing, and fuel cell industry revenues exceeded temperature increases, the chemical reactions
$750 million worldwide in 2010 (Pike Research, become more active, and these reactions improve the
2011). The Japanese car makers, Honda and Toyota, performance of carbon-polymer-based membranes.
plan to sell PEMFC vehicles to the public in 2015. Simultaneously, because the high temperature also
Many experts expect the PEMFC market to be active speeds catalyst reactions, this improvement helps to
in 2015, and sales are forecast to reach more than decrease the amount of catalyst required and to
500,000 vehicles by 2020. Based on the market situ- improve the possibility of using non-platinum cata-
ation, the proposed TRM concentrated on making lysts that could be used before. Such a change would
commercial products in 2020. decrease the cost of catalysts: for example, nickel, a
kind of non-platinum catalyst, is very cheap. In addi-
tion, high operating temperature eliminates the CO
4.4.2. Product planning poisoning problem, and this improves the durability
From the product portfolio map, we concluded that of PEMFC. Perfluoro-polymer-based membranes
the perfluoro-polymer-based membrane is a core have the fatal disadvantage of weak mechanical prop-
product, whereas the carbon-polymer-based mem- erties and the limitation that conductivity decreases
brane is an emerging product. In fact, research on drastically at high operating temperature. Nafion®
perfluoro-polymer-based membranes has been (Du Pont, Wilmington, DE, USA), the typical com-
actively conducted to improve their performance and mercial perfluoro-polymer-based membrane, is hard
durability. Some companies such as Avista (http:// to use at > 80°C because of performance degradation.
www.avistautilities.com) and Ballard Power (http:// Considering these characteristics, we conclude
www.ballard.com) have released perfluoro-based that research into carbon-polymer-based membranes
PEMFC products after reducing their manufacturing can overcome their low conductivity, compared
cost. However, perfluoro-polymer-based membranes with perfluoro-polymer-based membranes, and can
are very expensive, their performance decreases decrease the cost of catalysts. Therefore, we set the
sharply at temperatures > 80°C, and their frequency target product as a carbon-polymer-based membrane
suffers from the MC problem. In contrast, carbon- and the target functions as ‘improve heat resistance’
polymer-based membranes have low manufacturing and ‘improve durability’. In fact, the US Government
cost, high possibility of improving performance, and Department of Energy has concentrated on high-
a relatively simple manufacturing process. Because temperature PEMFC technology since 2006 (Schmit-
of these characteristics of carbon-polymer-based tinger and Vahidi, 2008).
membranes, the PEM market has the possibility of
changing from perfluoro-polymer-based membranes
to carbon-polymer-based membranes in the future. 4.4.4. Technology planning
Therefore, we assume that carbon-polymer-based From the proposed analysis, we learned that the
membranes will be dominant in the future. manufacturing technology for assembling mem-
branes and electrodes are considered to be a core
technology. In addition, the technological indicators
4.4.3. Function planning suggest that technology related to physical durability
From the proposed PFT maps, we summarized func- is an established technology. Like the result of func-
tion trends as follows. The current important func- tion analysis, technology related to ‘heat resistance’
tions are still those related to improving conductivity. has received attention recently and is therefore an
From the perspective of durability, functions related emerging technology.

70 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Figure 17. SAO-based TRM of PEM technology.

Considering the results of the PFT maps, we define PEM is a precondition for market entry, and, after
the technologies for a carbon-polymer-based mem- doing so, the research theme will change to improv-
brane to improve heat resistance. To achieve a suit- ing conductivity or decreasing cost. To this end,
able membrane, heat treatment and manufacturing developing new technologies for heat treatment and
technology must be developed. In addition, to developing new polymer materials are urgent chal-
increase mechanical strength, the improvement of lenges for preparing the future PEMFC market.
porous structure technology is also considered. Next, after 2020, the proposed TRM suggests that
Finally, after achieving the required durability, a new the commercialized membrane will have high con-
catalyst spreading method must be developed to ductivity. After solving the problems of durability,
decrease cost. the membrane will satisfy the market’s demand for
performance. Even though the high-temperature
PEM has the advantage of improving conductivity,
4.4.5. TRM of PEM technology high temperature alone cannot guarantee high con-
Based on the results of analysis, we suggest the SAO- ductivity. To satisfy the demand, companies should
TRM of PEM technology (Figure 17). The SAO- continue to invest in technologies such as membrane
TRM is divided into two periods. Because many manufacturing and porous structures to improve con-
experts expect that the PEMFC market will begin to ductivity at high temperatures.
expand in 2020, the proposed SAO-TRM concen- Finally, the proposed TRM pointed out the impor-
trates on development of PEM durability before 2020 tance of developing a cost-effective technology. After
and commercialization of membrane after 2020. the PEMFC market begins in earnest, membrane
Before 2020, the proposed SAO-TRM suggests a development objectives will change from perform-
carbon-polymer-based membrane with new materials ance and durability to cost. This kind of change is
to improve durability including mechanical strength, very general phenomenon and appears in most tech-
heat resistance, and chemical stability. In fact, an nology competition. We expect that the competition
important requisite for developing a high- in PEM technology will be similar to the completion
temperature PEMFC is to increase its durability at in the computer memory industry. The performance
high temperatures; if this is not accomplished, and durability of membranes show almost no differ-
PEMFC will have the fatal problem of gas leaking. ences between companies, so cost competitiveness
Polybenzimidazole membrane, which is a promising drives the competition in PEM technology. Because a
carbon-polymer-based membrane material for use at carbon-polymer-based membrane is less expensive
high operating temperature, has the disadvantage that than a perfluoro-polymer-based membrane, decreas-
it elutes phosphoric acid during extended operation. ing the cost can be considered a long-term techno-
Therefore, solving the problem of durability of the logical challenge. To decrease cost, technologies

© 2012 The Authors R&D Management 43, 1, 2013 71

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

regarding the decrease in use of platinum and catalyst systematically, an analysis system should be con-
spreading methods will be consistently researched. structed. Currently, we use Knowledgist™ to extract
SAO structures and to develop a system of using PFT
maps to construct an SAO-TRM. However, the
5. Conclusions current system cannot fully support the overall
process of constructing an SAO-TRM. If an opti-
This paper proposed an SAO-TRM, a new quantita- mized system that includes a technological database
tive approach to technology roadmapping. To can be developed, it would improve the practicality
construct an SAO-TRM, we used text-mining and of the proposed approach.
NLP techniques to extract SAO structures from patent
documents, and we suggested a procedure for trans- Acknowledgement
forming the extracted SAO structures into base
sources for TRM. Using processed data, we proposed This work was supported by the National Research
an SAO-TRM that considers a function layer and used Foundation of Korea (NRF) grant funded by the
PFT maps to guide the decision-making process when Korea government (MEST) (No. 2009-0088379).
constructing an SAO-TRM. Finally, to demonstrate
the feasibility of the proposed approach, we devel-
oped an SAO-TRM for a case study of developing References
PEMFC technology and compared our results with
the KIPO’s (2006) report written by technology Albright, R.E. and Kappel, T. (2003) Roadmapping in the
experts. corporation. Engineering Management Review, IEEE,
As a new approach to develop TRM, the proposed 31, 31–40.
approach provides a quantitative method that uses Altshuller, G.S. (1984) Creativity as an Exact Science: The
patent information and represents the new type of Theory of the Solution of Inventive Problems. New York:
TRM diagram as a function layer. This research uses Gordon and Breach Science.
SAO structures. Because SAO structures contain Archibugi, D. and Planta, M. (1996) Measuring techno-
logical change through patents and innovation surveys.
information that is not provided by keywords, such as
Technovation, 16, 451–468.
means–end relationships and purposes of technology, Bergmann, I., Butzke, D., Walter, L., Fuerste, J.P.,
the structures can offer a variety of views regarding Moehrle, M.G., and Erdmann, V.A. (2008) Evaluating
technology in the decision-making process. Although the risk of patent infringement by means of semantic
it is hard to develop a TRM only depending on pro- patent analysis: the case of DNA chips. R&D Manage-
posed approach because of its reliability and techno- ment, 38, 550–562.
logical limitation, the proposed approach allow Callon, M., Courtial, J.-P., and Laville, F. (1991) Co-word
R&D managers to extend their views of product and analysis as a tool for describing the network of interac-
technology while developing the TRM. In addition, tions between basic and technological research: the case
by reducing the time and cost of developing a TRM, of polymer chemistry. Scientometrics, 22, 155–205.
the proposed approach supports quick and accurate Cascini, G., Fantechi, A., and Spinicci, E. (2004) Natural
language processing of patents and technical documen-
decision-making by R&D managers.
tation. In: Marinai, S. and Dengel, A. (eds), Document
To make the proposed approach more practical, Analysis Systems VI. Berlin/Heidelberg: Springer. pp.
three challenges must be addressed. First, a more 89–92.
efficient method extracting SAO structures should be Cascini, G., Lueehesi, D., and Rissone, P. (2001) Auto-
developed. Knowledgist™ is almost the only tool matic patents functional analysis through semantic
available for extracting SAO structures. Although it processing. In 12th ADM International Conference,
is easy to use, it extracts many irrelevant SAO struc- Rimini, Italy.
tures. This problem means that time must be spent Cascini, G. and Russo, D. (2006) Computer-aided analysis
removing irrelevant SAO structures and reduces of patents and search for TRIZ contradictions. Interna-
the reliability of the TRM constructed using the tional Journal of Product Development, 4, 52–67.
proposed approach. A better SAO extractor can guar- Cascini, G. and Zini, M. (2008) Measuring patent similar-
ity by comparing inventions functional trees. Interna-
antee an SAO-TRM of better quality. Second, well-
tional Federation for Information Processing 20th World
defined technology thesauri or ontologies should be Computer Congress. Milano, Italy, 31–42.
constructed. In the proposed approach, we manually Chen, L., Tokuda, N., and Adachi, H. (2003) A patent
analyzed technological word phrases to identify their document retrieval system addressing both semantic and
types. Well-defined thesauri would be more adequate syntactic properties. Proceedings of the ACL-2003 work-
for resolving identification of technology words. shop on Patent corpus processing – Volume 20. Sapporo,
Finally, to support the development of an SAO-TRM Japan: Association for Computational Linguistics, 1–6.

72 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd
 An SAO-based text-mining approach for technology roadmapping

Cheong, S.H. (2006) MOT by using scientific methodol- Miller, W. and Morris, L. (1999) Fourth Generation R&D:
ogy in Samsung R&D. Technology Management for the Managing Knowledge, Technology, and Innovation.
Global Future, 2006. PICMET 2006, xlv–lxxxi. New York: John Wiley & Sons.
Choi, S., Yoon, J., Kim, K., Lee, J., and Kim, C.-H. (2011) Moehrle, M. (2010) Measures for textual patent similari-
SAO network analysis of patents for technology trends ties: a guided way to select appropriate approaches. Sci-
identification: a case study of polymer electrolyte mem- entometrics, 85, 95–109.
brane technology in proton exchange membrane fuel Moehrle, M., Walter, L., Geritz, A., and Müller, S. (2005)
cells. Scientometrics, 88, 3, 863–883. Patent-based inventor profiles as a basis for human
Dale, B.G. (2003) Managing Quality. Malden, MA; resource decisions in research and development. R&D
Oxford: Blackwell. Management, 35, 513–524.
Diestel, R. (2010) Graph Theory. Heidelberg: Springer- Narin, F. (1993) Technology indicators and corporate strat-
Verlag. egy. Review of Business, 14, 3, 19–23.
ERIMA. (1997) Technology roadmapping delivering busi- Petrick, I.J. and Echols, A.E. (2004) Technology roadmap-
ness vision. Working Group Reports. Paris. ping in review: a tool for making sustainable new
Fleisher, C. and Bensoussan, B. (2003) Strategic and Com- product development decisions. Technological Forecast-
petitive Analysis: Methods and Techniques for Analyzing ing and Social Change, 71, 81–100.
Business Competition. Upper Saddle River, NJ: Prentice Phaal, R., Farrukh, C.J.P., Mills, J.F., and Probert, D.R.
Hall. (2003) Customizing the technology roadmapping
Garcia, M.L. and Bray, O.H. (1997) Fundamentals of approach. Management of Engineering and Technology,
Technology Roadmapping. Albuquerque, NM: Sandia 2003. PICMET ’03. Technology Management for
National Laboratories. Reshaping the World. Portland International Conference
Guglielmi, M., Williams, E., Groepper, P., and Lascar, on, 361–369.
S. (2010) The technology management process at the Phaal, R., Farrukh, C.J.P., and Probert, D.R. (2001) Char-
European space agency. Acta Astronautica, 66, 883– acterisation of technology roadmaps: purpose and
889. format. Management of Engineering and Technology,
Hirtz, J., Stone, R., McAdams, D., Szykman, S., and Wood, 2001. PICMET ’01. Portland International Conference
K. (2002) A functional basis for engineering design: on, 2, 367–374.
reconciling and evolving previous efforts. Research in Pike Research. (2011) Fuel Cell Industry is Poised for
Engineering Design, 13, 65–82. Major Change and Development in 2011 (See http://
KIPO. (2006) Patent Map Report: Proton Exchange Mem- www.pikeresearch.com/newsroom/fuel-cell-industry-is-
brane Fuel Cell. Seoul, Korea: Korea Intellectual Prop- poised-for-major-change-and-development-in-2011).
erty Office. Reitzig, M. (2004) Improving patent valuations for
Kostoff, R. (1998) The use and misuse of citation analysis management purposes – validating new indicators by
in research evaluation. Scientometrics, 43, 27–43. analyzing application rationales. Research Policy, 33,
Kostoff, R.N., Boylan, R., and Simons, G.R. (2004) Dis- 939–957.
ruptive technology roadmaps. Technological Forecast- Sabidussi, G. (1966) The centrality index of a graph. Psy-
ing and Social Change, 71, 141–159. chometrika, 31, 581–603.
Kostoff, R.N. and Schaller, R.R. (2001) Science and tech- Savransky, S.D. (2000) Engineering of Creativity: Intro-
nology roadmaps. IEEE Transactions on Engineering duction to TRIZ Methodology of Inventive Problem
Management, 48, 132–143. Solving. Boca Raton, FL: CRC Press.
Lee, B. and Jeong, Y.-I. (2008) Mapping Korea’s national Schmittinger, W. and Vahidi, A. (2008) A review of the
R&D domain of robot technology by using the co-word main parameters influencing long-term performance and
analysis. Scientometrics, 77, 3–19. durability of PEM fuel cells. Journal of Power Sources,
Lee, C., Jeon, J., and Park, Y. (2011) Monitoring trends of 180, 1–14.
technological changes based on the dynamic patent Schumpeter, J.A. and Opie, R. (1934) The Theory of Eco-
lattice: a modified formal concept analysis approach. nomic Development; an Inquiry into Profits, Capital,
Technological Forecasting and Social Change, 78, 690– Credit, Interest, and the Business Cycle. Cambridge,
702. MA: Harvard University Press.
Lee, S., Seol, H., and Park, Y. (2008) Using patent infor- Sternitzke, C. and Bergmann, I. (2009) Similarity measures
mation for designing new product and technology: for document mapping: a comparative study on the
keyword based technology roadmapping. R&D Manage- level of an individual scientist. Scientometrics, 78, 113–
ment, 38, 169–188. 130.
Lee, S., Yoon, B., Lee, C., and Park, J. (2009) Business Suh, J.H. and Park, S.C. (2009) Service-oriented technol-
planning based on technological capabilities: patent ogy roadmap (SoTRM) using patent map for R&D
analysis for technology-driven roadmapping? Techno- strategy of service industry. Expert Systems with Appli-
logical Forecasting and Social Change, 76, 769– cations, 36, 6754–6772.
786. Tseng, Y.-H., Lin, C.-J., and Lin, Y.-I. (2007) Text mining
Li, Y.-R. (2009) The technological roadmap of Cisco’s techniques for patent analysis. Information Processing
business ecosystem. Technovation, 29, 379–386. and Management, 43, 1216–1247.

© 2012 The Authors R&D Management 43, 1, 2013 73

R&D Management © 2012 Blackwell Publishing Ltd
Sungchul Choi, Hongbin Kim, Janghyeok Yoon, Kwangsoo Kim and Jae Yeol Lee

Vatananan, R.S. and Gerdsri, N. (2010) The current state of Hongbin Kim is in PhD program of Industrial and
technology roadmapping (TRM) research and practice. Management Engineering at Postech, South Korea.
Technology Management for Global Economic Growth His interests are in the area of Patent Analysis, Tech-
(PICMET), 2010 Proceedings of PICMET ’10, 1–10. nology Planning, and Technology Evaluation. His
Wanner, L., Baeza-Yates, R., Brügmann, S., Codina, J.,
current research focuses on the function network
Diallo, B., Escorsa, E., Giereth, M., Kompatsiaris, Y.,
where heterogeneous technologies are connected.
Papadopoulos, S., Pianta, E., Piella, G., Puhlmann, I.,
Rao, G., Rotard, M., Schoester, P., Serafini, L., and
Zervaki, V. (2008) Towards content-oriented patent Janghyeok Yoon, an associate professor in the
document processing. World Patent Information, 30, Konkuk University’s Department of Industrial Engi-
21–33. neering, has received his PhD degree from Pohang
Yoon, B., Phaal, R., and Probert, D. (2008a) Morphology
University of Science and Technology in Korea. He
analysis for technology roadmapping: application of text
has several years industrial experiences in companies
mining. R&D Management, 38, 51–68.
Yoon, B., Phaal, R., and Probert, D. (2008b) Structuring and research institutes such as LG and Korea Insti-
technological information for technology roadmapping: tute of Intellectual Property. His research interests
data mining approach. Proceedings of the 7th WSEAS include technology intelligence-related topics
International Conference on Artificial Intelligence, including technology forecasting, technology oppor-
Knowledge Engineering and Data Bases. Cambridge, tunity identification, technology roadmapping, and
UK: World Scientific and Engineering Academy and technology convergence.
Society (WSEAS), pp. 417–422.
Yoon, J., Choi, S., and Kim, K. (2011a) Invention property-
function network analysis of patents: a case of silicon- Kwangsoo Kim is a full professor in the Department
based thin film solar cells. Scientometrics, 86, 687– of Industrial and Management Engineering at
703. Pohang University of Science and Technology,
Yoon, J. and Kim, K. (2011) Identifying rapidly evolving Korea. He holds BS and MS degrees from Seoul
technological trends for R&D planning using SAO- National University, Korea, and PhD degree from
based semantic patent networks. Scientometrics, 88, University of Central Florida, US. His teaching and
213–228. research interest is currently focused on discovery of
Yoon, J., Lim, J., Choi, S., Kim, K., and Kim, C.-H. the future by mining patents and mimicking nature.
(2011b) Ontological functional modeling of technology
for reusability. Expert Systems with Applications, 38,
10484–10492. Jae Yeol Lee is an associate professor in the Depart-
Zwicky, F. (1948) Morphological astronomy. The ment of Industrial Engineering, Chonnam National
Observatory, 68, 121–143. University, Korea. He received his BS, MS and PhD
degrees in Industrial Engineering from Pohang Uni-
Sungchul Choi is an assistant manager in technology versity of Science and Technology (POSTECH),
strategy group at Samsung Advanced Institute of Korea, in 1992, 1994, and 1998, respectively. From
Technology. He has received his PhD degree from 1998 to 2003, he worked as a senior researcher at
Pohang University of Science and Technology in Electronics and Telecommunications Research
Korea. He has researched on technology manage- Institute, Korea. Since 2003, he has been a faculty
ment issues such as technology intelligence and tech- member of Chonnam National University. His
nology roadmapping. Currently, his research interest current research interests include semantic patent
is focused on R&D knowledge management and analysis, collaborative design and visualization, and
intelligent technology sensing system. human computer interaction.

74 R&D Management 43, 1, 2013 © 2012 The Authors

R&D Management © 2012 Blackwell Publishing Ltd