Academic journal article Innovation: Organization & Management


Academic journal article Innovation: Organization & Management


Article excerpt

(ProQuest: ... denotes formulae omitted.)

1. Introduction

Nowadays, environmental issues such as improving life quality, dealing with climate change and managing water resources have been given more and more attention. The market size of the global environmental industry was approximately US$800 billion in 2010 and it is expected to grow to over US$1,000 billion in 2017. Furthermore, most fields in the environmental industry are at the stage of initial market formation and thus, the growth potential of the environmental industry is very high. Countries all over the world are establishing eco-friendly policies for economic growth; in particular, developed countries like the US, Japan and the EU regard eco-innovation as an opportunity for economic growth with sustainability. Although defining eco-innovation is not an easy task, in general, literatures emphasize that eco-innovations reduce the environmental impact caused by consumption and production activities, whether the main motivation for their development or deployment is environmental or not (Carrillo-Hermosilla, Del Río, & Könnölä, 2010).

Research issues related to eco-innovation can be summarised into four categories (Kemp & Pearson, 2007). The first is the innovation of environmental technologies (ETs). The second is the introduction of organizational methods and management systems for dealing with environmental issues in production and product. The third is new or environmentally improved products and environmentally beneficial services. The last is the innovation of green system such as biological agriculture system and renewables-based energy system. Among them, the innovation of ETs is regarded as a crucial issue, since ETs play the fundamental role in eco-innovation and influence the direction of other issues (Foray & Grübler, 1996; Petruzzelli, Dangelico, Rotolo, & Albino, 2011). The rise of the greening of markets has been pointed to as part of an overall paradigm change of ET. (Andersen, 2008). Due to the intractable complexity and volatility of ETs, it becomes more important to grasp the technological trends and advances by analyzing the overall structure of ETs and interaction among them (Kemp & Oltra, 2011). Identifying and assessing technological advances critical to the company's competitive position is now recognized as a crucial activity for achieving and maintaining competitive positions in eco-innovation fields. It is considered to be indispensable, in particular, for the innovation of ETs that seek technological possibilities through technological fusion among various fields of technologies. Consequently, there have been attempts to identify the technological structure and relationship among ETs.

The identification of the technological structure and relationship among ETs is primarily conducted through the patent analysis (Markatou, 2012). It is reported that patents contain about 80% of all the knowledge of technological innovation (Blackman, 1995). Moreover, they can be easily accessed and analyzed through various types of public or private databases. In spite of controversial discussions regarding the limitations of patent analysis (Archibugi & Pianta, 1996; Ernst, 2003), patents have a unique advantage of quantitativeness over the conceptual or qualitative approaches. Also, they contain empirical information with regard to most innovation in developed countries. Patents are, hence, perceived as useful information for measuring eco-innovation (Kemp & Oltra, 2011).

The most commonly used information of patents is citation and co-classification. Citation analysis is based on the citation relationship among different patents, assuming that the knowledge of a cited patent is transferred to a citing patent, and there exists a technological linkage between them (Lee, Kim, & Park, 2009; Lee & Kim, 2010). The identification of the structure of ETs with citation information is mainly conducted by analyzing the number of cited patents (Acosta, Coronado, & Fernández, 2009; Arundel & Kemp, 2009; Popp, 2006) and citation networks (Nemet, 2012; Wu, Hsu, Lee, & Su, 2011). However, citation analysis has some shortcomings such as that there is a timelag between the citing and cited patents and furthermore, it only considers the individual patent level (Yoon & Park, 2004).

Co-classification analysis analyzes the technological relationships based on the fact that patents are classified into some technological classifications considering their technological characteristics (OECD, 1994). The assumption of the co-classification analysis is that the frequency by which two classification codes are jointly assigned to a patent document can be interpreted as a sign of the strength of the knowledge relationships in terms of knowledge links and spillovers (Breschi, Lissoni, & Maleraba, 2003). In contrast to citation analysis, it is based on the hierarchical technological classification system and therefore, technological relationships can be analyzed not at the level of individual patents but at various technological levels according to the purpose of studies. Furthermore, errors from the time-lag problem are relatively insignificant because the time of classification information of a patent is equal to that of patent registration.

Studies on the examination of the structure and linkage of ETs based on patent co-classification have been conducted from the perspectives of three technological interrelationships - intensity, relatedness and cross-impact. First, those from the intensity perspective used the number of co-occurrence, based on the fact that the larger the number of patents affiliated to two classes, the more the intensity of interrelationship is between the two classes (Englesman & van Rann, 1992). For example, Meyer (2006) represented a co-classification map of ETs by applying a clustering and multidimensional scaling algorithm to patent co-classification frequency. Tanner (2012) identified the technologically related knowledge fields of fuel cell technologies with the frequency of patent co-classification. Second, those from the relatedness perspective applied the cosine index, which has the form of a correlation (Breschi et al., 2003; Jaffe, 1986). For instance, Dibiaggio and Nasiriyar (2009) measured a level of substitutability between fuel cell technologies by applying cosine index to their joint occurrences in patent classification. Fornahl, Broekel, and Boschma (2011) estimated the relatedness between biotechnologies by means of a similarity measure using patent co-classification codes. Finally, those from a cross-impact perspective employed a cross-impact index, which measures an impact of an event on other events' occurrence (Choi, Kim, & Park, 2007; Kim, Lee, Seol, & Lee, 2011). For example, Weenen, Ramezanpour, Ramezanpour, Commandeur, and Claassen (2013) measured technological interrelations among medical nutrition technologies by applying a cross-impact analysis to the patent co-classification data. Yoon and Jeong (2013) applied a patent-based cross-impact analysis to estimate the degree of impact of biotechnology to other technologies.

Although these studies could offer many implications for understanding the technological interrelationships among ETs, they are subject to some limitations. First, they only tried to investigate technology pairs with a high interrelationship and did not take into account the overall effects each technology gave to others. Second, they conducted an analysis on a specific level of patent classification system and scarcely considered the indirect relationships among technologies in the upper and lower levels. Finally, they focused only on one of the three perspectives, not on all of them.

To address these limitations, this study proposes a new approach in order to identify the core ETs using patent co-classification information with the consideration of the overall interrelationships among ETs. Association rule mining (ARM) is employed to more easily capture the technological interrelationships among ETs, and three technological interrelationship matrixes are constructed with the derived support, liftand confidence values in order to take into account the overall effects each ET gives to others. The ANP (Analytic Network Process) is conducted to produce priorities of ETs with consideration of their direct and indirect impacts. DEA (Data Envelopment Analysis) is then applied to calculate the total importance of each ET, putting three perspectives of technological interrelationship together.

The remainder of this study is organized as follows. Section 2 reviews the theoretical background of ARM, ANP and DEA. Section 3 explains the overall process of the proposed approach. Section 4 shows how to apply the proposed approach to identify core ETs. Section 5 concludes this study with directions for future research.

2. Literature review

2.1. Association rule mining (ARM)

ARM is one of the data mining techniques that searches for interesting relationships among items in a large database. An association rule is the co-occurrence of two items, indicating that if two items occur together frequently, they have a strong association relationship (Han & Kamber, 2001). ARM has primarily been applied to firm activities, particularly to marketing (Liao & Chen, 2004; Wong, Zhou, & Yang, 2005; Rong, Vu, & Law, 2012). It has also been used in various areas, such as medicine (Kumar, Ranjan, & Kumar, 2013; Sebastian & Then, 2011), resources (Jiang, Zhang, & Wang, 2011), bioinformatics (Creighton & Hahash, 2003; Wright, Chen, & Maloney, 2010) and finance (Hsieh, 2004; Keasey & McGuinness, 2008). In particular, ARM has recently been applied to analyze the patent database for various purposes, such as technological trend analysis (Shih, Liu, & Hsu, 2010), core technology identification (Kim et al., 2011), complementary technology exploration (Wang, 2012) and technology classification (He & Loh, 2010).

The three measures for evaluating the rule interestingness are support, liftand confidence. Their brief descriptions are presented in Table 1. In this study, they are applied for constructing technological interrelationship matrices. Specifically, the values of support, liftand confidence between all technology pairs calculated by conducting ARM on the patent co-classification information are used for the values of the cells in the intensity matrix, cross-impact matrix and relatedness matrix, respectively.

2.2. Analytic network process (ANP)

The ANP is a generalization of the Analytic Hierarchy Process (Saaty, 1996), one of the most widely used multiple-criteria decision-making (MCDM) methods. The AHP decomposes a problem into a hierarchy in which each decision element is considered to be independent and thus, it cannot accommodate the interrelationships among elements. The ANP extends the AHP to problems with dependence and feedback. It allows for more complex interrelationships among decision elements by replacing the hierarchy in the AHP with a network (Meade & Sarkis, 1999). Therefore, in recent years, there has been an increase in the use of the ANP to the patent data in order to analyze technological interrelationships in a variety of problems, such as technology selection (Shen, Lin, & Tzeng, 2011), core technology identification (Kim et al., 2011; Lee, Kim, Cho, & Park, 2009), research and development (R&D) project evaluation (Jung & Seo, 2010), technological convergence (Geum, Lee, Kim, & Kim, 2012) and R&D partner selection (Geum, Lee, Yoon, & Park, 2013).

The applications of the ANP can be divided into three types according to their primary purpose (Kim et al., 2011). As a MCDM method, first, ANP has widely been used for modeling a decision problem and selecting the best alternative with conflicting and interrelated criteria, which is the original aim of the ANP. Second, the ANP has also been applied to the quantification of an existing framework by prioritizing the interrelated elements in the framework (Asan & Soyer, 2009; Yüksel & Dagdeviren, 2007). The third type is focused on capturing the indirect influences among elements in a network and producing their relative importance (Lee et al., 2009; Lee, Seol, Sung, Hong, & Park, 2010). The use of the ANP in this study can be included in this type. The process of ANP is composed of four steps (Chung, Lee, & Pearn, 2005; Saaty, 1996): (1) network model construction, (2) pairwise comparison and local priority vectors, (3) supermatrix formation and transformation, and (4) final priorities.

2.3. Data envelopment analysis (DEA)

DEA is a non-parametric approach for evaluating the relative efficiencies of decisionmaking units (DMUs) with multiple inputs and outputs. It does not require any assumptions regarding the functional form of a production function and a priori information on the importance of inputs and outputs. The relative efficiency of a DMU is measured by estimating the ratio of the weighted outputs to the weighted inputs and comparing it with those of others. A DMU's efficiency is restricted to be less than or equal to one. Under this restriction, efficiency is measured based on the weight of each element maximizing the efficiency of target DMUs. DEA is a useful tool for benchmarking by providing the information for improving efficiency through a reference set for inefficient DMUs. It also allows for the calculation of the required amount of improvement in the inefficient DMU's inputs and outputs to make it efficient. For this reason, DEA has been used for performance evaluation and benchmarking in various areas, such as education, law, hospital and banking (Metters, King-Metters, & Pullman, 2002). Application of DEA to the patent information includes new business area identification (Seol, Lee, & Kim, 2011), technological trend analysis (Tseng, Hsieh, Peng, & Chu, 2011), technological forecasting (Anderson, Daim, & Kim, 2008) and technological strategy efficiency evaluation (Lee, 2010). While DEA was originally developed for measuring the efficiency of multiple units performing a transformation process of several inputs and several outputs, it is now playing a broader role as a tool for MCDM problems (Bouyssou, 1999).

The first DEA model proposed by Charnes, Cooper, and Rhodes (1978) is the CCR (Charnes, Cooper, and Rhodes) model, which assumes that production exhibits constant returns to scale. Banker, Charnes, and Cooper (1984) extended it to the BCC (Banker, Charnes, and Cooper) model for the case of variable returns to scale. DEA models are also distinguished by the objective of the model: maximize outputs (output-oriented) or minimize inputs (input-oriented). In some MCDM problems, there is no negative (or positive) evaluation item. More specifically, all criteria are preferred to be high (or low); thus, only outputs (or inputs) will exist when using DEA. To accommodate this kind of situation, Lovell and Pastor (1999) suggest the pure output (or input) model without inputs (or outputs). They proved that an output-oriented CCR model with a single constant input and an input-oriented CCR model with a single constant output coincide with the corresponding BCC models; however, a CCR model without inputs (or outputs) is meaningless. The output-oriented BCC model employed in this study is formulated as


where X is the matrix of input vectors, Y is the matrix of output vectors, X0;Y0 is the DMU being measured, k is the reverse of the efficiency score, and k is the vector of intensity variables.

3. Proposed approach

3.1. Overall process

The procedure of identifying the core ETs is as follows. First, patent data of ET is collected. Second, three technological interrelationship matrices - intensity matrix, relatedness matrix and cross-impact matrix - are constructed with support, liftand confidence values calculated by conducting ARM on the co-classification information of the gathered patent data. Third, technological importances are derived by applying ANP to the technological interrelationship matrices. Finally, core ETs are identified through employing DEA to the derived technological importance values. Figure 1 depicts the overall process of the proposed approach. More detailed explanations are provided below.

3.2. Data collection

For collecting patent data related to ET, it is necessary to identify what classifications are for ETs on the patent classification system. The patent classification system stands for the hierarchical system that classifies and manages patents considering their technological characteristics (Kim et al., 2011). In general, the technological characteristics of patents can be identified with claim. Claim is a disclosure of new features of patents, and patents are classified based on the technological characteristics of each claim. Patents have more than two claims for diversifying subjects and thus, they are affiliated to more than two classifications based on the patent classification system (USPTO, 2012). Classification, therefore, indicates which technological areas the patents (individual technologies) are affiliated in the patent classification systems.

This study applies IPC (International Patent Classification) as a technology classification system for analysis. It is composed of eight sections and each section is divided into the hierarchy of class, subclass, maingroup and subgroup. For example, at 'A62D 1/02', 'A', '62', 'D', '1' and '02' correspond to section, class, subclass, maingroup, and subgroup, respectively. As presented in Table 2, there are eight subclasses and 39 maingroups in IPCs for ET (Acosta et al., 2009). Therefore, in this study, a subclass and a maingroup serve as the units of core ET identification, and collected patents of ET assigned to ET-related classifications are affiliated to their relevant categories.

3.3. Technological interrelationship matrices construction

As aforementioned, there are three technological interrelationships - intensity, relatedness and cross-impact. Therefore, the three types of technological interrelationship matrix are constructed by using the values of technological interrelationships between all the pairs of ETs, calculated with the co-classification information of the collected patent data. Let us indicate with M all the patents that are determined. Each of the M patents is assigned to more than one classification code. Let FAm ¼ 1 if a patent m contains the classification code A and FAm ¼ 0 otherwise. The number of patents with classification code A is therefore given by ....Thus, the number of patents that are classified into both classifications A and B is indicated as .... If the number of co-classifications that M patent can have is N, the intensity index of technologies A and B is defined as Int ... (Englesman & van Rann, 1992). It is the normalized number of patents that are classified by both technologies A and B. The relatedness index of technologies A and B is defined as Rel ... (Han & Kamber, 2001; Jaffe, 1989). It is the cosine index, which measures the correlation between them. This index measures the angular separation between the vectors representing the co-occurrences of technologies A and B, respectively, with all other technologies. The cross-impact index of technologies A and B is defined as .... It has a form of the conditional probability. With Int ..., technological interrelationship matrices having a square form, whose row and column stand for the technological classification, can be constructed. Namely, it can be constructed by inserting the values calculated by applying three equations to all the technological combinations to their relevant positions. However, it requires a huge amount of calculation with the patent data to calculate all the technological interrelationship indexes among the technologies. This study thus applies ARM to calculate such indexes because the intensity index, relatedness index and cross-impact index have the same formula with the support, liftand confidence of the association rule AB in ARM, as shown in Table 1. Specifically, the values of support, liftand confidence between all technology pairs calculated by conducting ARM on the patent co-classification information are used for the values of the cells in the intensity matrix, relatedness matrix and cross-impact matrix, respectively. Table 3 represents a form of the intensity matrix. Ti means the technologies in the ith technological area and Sup(TiTj) indicates the support value of the association rule TiTj. The diagonal values of the matrix are set to 1 because there is a 100% interrelationship between the same technologies. Since there are three technological interrelationships and two IPC levels, six technological interrelationship matrices are constructed in this way.

3.4. Technological importance derivation

Technological importances of each ET from the perspectives of intensity, relatedness and cross-impact are derived. To calculate the importances of ETs based on the direct and indirect relationships among technologies, the ANP is applied to three technological interrelationship matrices as follows.

First, the ET network model is constructed. Basically, a network model in the ANP is constructed based on expert judgments in order to model an abstract decision problem. However, the network in the proposed approach is made on the basis of technological interrelationships represented in the constructed matrices. A cluster in the ANP network corresponds to an upper level classification (subclass) and elements in a cluster are equivalent to lower lever classifications (maingroups) in an upper level classification. In the ANP context, then, the resulting network model only includes alternative clusters, contrary to the general network model in the ANP comprised of a goal cluster, criteria clusters and alternative clusters. Thus, the importance of alternatives is only evaluated with respect to the impacts or influences on other alternatives, not with respect to some criteria or goal.

Second, the priority vector is derived. The basic form of measurement in the ANP is a pairwise comparison with a scale of 1-9 because subject judgments have to be made on qualitative aspects. However, pairwise comparisons do not have to be conducted in the proposed approach. It implicitly assumes that the intensity, relatedness and crossimpact index between a pair of node are the proxies for the degree of influences from each perspective. Thus, pairwise comparisons are not required and priority vectors can be directly obtained from technological interrelationship matrices. For example, let .... It can be interpreted that technology B is 2 (=0.2/0.1) times more important than technology C in terms of relatedness with technology A. Then, the number 2 is inserted to position (B, C) and the reciprocal value of 0.5 is assigned to position (C, B). In this way, the pairwise comparison matrix with respect to the technology among all technologies can be obtained. Then, the priority vector for technology A is derived from the eigenvector method. This priority vector is naturally the same as the column relevant to technology A in a normalized technological interrelationship matrix, whose column sums to one (Lee et al., 2009). Therefore, the priority vectors can be directly obtained from the technological interrelationship matrices without pairwise comparisons.

Third, the supermatrix is constructed and transformed. The local priority vectors are entered into the appropriate columns of a supermatrix, which is a partitioned matrix where each segment represents a relationship between two clusters. To transform the supermatrix into a weighted supermatrix, each matrix segment of the supermatrix is multiplied by the corresponding cluster weights. However, in general, the resulting matrix is not column stochastic because there can be several matrix segments which have columns all of whose entries are zero. When this is the case, the weighted column of the supermatrix must be renormalized (Saaty, 1996). The renormalized matrix, which is now column stochastic, is what is called the weighted supermatrix. This column stochastic feature of the weighted supermatrix in which each of the columns sums to one allows convergence to occur in the limit supermatrix. Then, each weighted supermatrix is transformed into the limit supermatrix by raising itself to powers. The columns of the limit supermatrix converge identically, which is called limit centralities, capturing all of the direct and indirect influences among ETs.

Finally, final priorities are derived. The importance of technologies of each perspective can be identified based on each limit centrality of ETs.

3.5. Core ETs identification

Core ETs are identified through employing DEA to the derived technological importance values. When considering this to be a problem of MCDM in selecting ETs with high importance, ETs correspond to the alternatives and three technological interrelationships match to the criteria. The total importance of ETs implies the priority scores of each ET. Also, if applying DEA to be a method for MCDM, the ETs, three technological interrelationships and the total importance of ETs correspond to DMUs, outputs and efficiency scores derived by applying DEA, respectively. Therefore, the core ETs considering intensity, relatedness and cross-impact together are identified through DEA. Because the efficiency score in the DEA ranges from 0% to 100%, the total importance also has values from 0 to 100. Figure 2 shows the correspondence among core ET identification, MCDM and DEA. Because only three outputs exist and there is no input, the outputoriented BCC model without inputs is adopted for applying DEA. This study only tried to judge the priority scores of ETs and therefore, it was not appropriate to reflect on R&D investment or labor for inputs.

4. Results

4.1. Data collection

Documents of 33,198 patents assigned to ET-related classifications, registered until 2012, were collected from the USPTO (United States Patent and Trademark Office) database and stored in our database. Because the number of patents was so huge such that we could not collect all of them manually, the own-developed JAVA-based web document parsing and mining program was used for automatically downloading the patent documents.

4.2. Technological interrelationship matrices construction

To construct technological interrelationship matrices, ARM was applied to the co-classification information of gathered patent data at the level of subclass and maingroup, respectively. SAS E-miner release 9.3, one of the data-mining packages, was used, and Apriori algorithm was selected to search the rules. With the derived support, liftand confidence values between all technology pairs, six technological matrices were constructed. For example, Table 4 shows the intensity matrix at the subclass level and Table 5 shows the intensity matrix at the maingroup level.

4.3. Technological importance derivation

The ANP was applied to the constructed technological matrices for calculating the technological importance of each perspective. First, the ET network model was constructed. Figure 3 portrays the ET network, including eight clusters and 39 elements. Every subclass had influences on each other, and included a feedback loop that represented the technological interrelationships among the maingroups in the subclass itself.

Second, priority vectors were derived. At first, the cluster weights are determined through normalizing the technological interrelationship matrices at the subclass level priority vector. For example, Appendix A shows the priority vector for each cluster of the intensity perspective, which was derived by normalizing Table 4. Next, local priority vectors at the element level were also obtained by normalization without pairwise comparisons. For example, the importance of maingroups of subclass B01D on each maingroup of subclass C02F from the intensity perspective was obtained through the transformation of the intensity matrix at the maingroup level (Table 5), as shown in Appendix B. What is important here is that the normalization of columns has to be done for each cluster. Cluster weights and local priority vectors of the relatedness and crossimpact perspectives were also obtained.

Third, the supermatrix for the ET network, a 39 x 39 matrix composed of 64 (=8 x 8) blocks, was constructed with obtained local priority vectors. A block corresponds to a set of priority vectors, a priority matrix. For example, Appendix C shows a part of the supermatrix of the intensity perspective. To transform the supermatrix into a weighted supermatrix, each matrix segment of the supermatrix was multiplied by the corresponding cluster weights shown in Appendix A and the weighted column of the supermatrix was renormalized. A part of the weighted supermatrix of the intensity perspective is shown in Appendix D. Finally, the limit supermatrix was derived by raising the weighted supermatrix to powers. Appendix E shows a part of the limit supermatrix of the intensity perspective. The limit supermatrix of the relatedness and cross-impact perspectives were also constructed in this manner.

Finally, the final priorities were derived. The columns in the limit supermatrix (Appendix E) represent the final priorities, that is, limit centralities. The technological importances of the eight subclasses and 39 maingroups of three perspectives are shown in Table 6 and Table 7, respectively. The limit centrality of a subclass is the sum of the maingroups belonging to the subclass.

4.4. Core ETs identification

Core ETs were identified through the output-oriented BCC DEA model. This model uses the derived importance values of ETs of each perspective as output data and the constant value (e.g., 10) as input data. At the subclass level, A62D (Chemical means for extinguishing fires; Processes for making harmful chemical substances harmless, or less harmful, by effecting a chemical change; Composition of materials for coverings or clothing for protecting against harmful chemical agents; Composition of materials for transparent parts of gas-masks, respirators, breathing bags or helmets; Composition of chemical materials for use in breathing apparatus), B01D (Separation) and B09C (Reclamation of contaminated soil) scored 100, as indicated in Table 6. It is obvious that these ETs have significant impacts on other ETs and therefore, they are considered as the core ETs in terms of three perspectives together. In addition, B09B (Disposal of solid waste) and F23J (Removal or treatment of combustion products or combustion residues; Flues) followed them. On the contrary, C02F (Treatment of water, waste water, sewage or sludge) scored 73.81 in efficiency, and it was the least important.

When it comes to the maingroup level, as shown in Table 7, B01D 53 (Separation of gases or vapours; Recovering vapours of volatile solvents from gases; Chemical or biological purification of waste gases, e.g. engine exhaust gases, smoke, fumes, flue gases or aerosols) and B09C 1 (Reclamation of contaminated soil) had the highest efficiency scores, followed by A62D 3 (Processes for making harmful chemical substances harmless, or less harmful, by effecting a chemical change in the substances), F01 N 3 (Exhaust or silencing apparatus having means for purifying, rendering innocuous or otherwise treating exhaust) and B09B 3 (Destroying solid waste or transforming solid waste into something useful or harmless). The least important ETs were those at F23 J 9 (Preventing premature solidification of molten combustion residues).

5. Conclusions

This study contributes to the eco-innovation fields by proposing a new approach to identify core ETs based on a quantitative analysis of patent information. ARM is employed more easily for capturing the technological interrelationships among ETs as well as to construct technological interrelationship matrices. ANP is used to derive the limit centrality that measures the importance of ETs in terms of impacts on other ETs in the ET network, taking indirect impacts or relationships into account. This is because ANP captures the relative importance that mirrors all of the direct and indirect interactions. DEA is applied to calculate the total importance of each ET, putting three perspectives of the technological interrelationship together. The proposed approach can be utilized for technology monitoring for both firms' planning of ETs and eco-innovation policy-making of governments.

Despite the contribution, this study has some limitations. First, although patent information has been widely accepted as a proxy for technological innovation, there is no guarantee that eco-innovation can be fully explained by the patent network analysis. Future research should address this limitation by exploring another important database, such as a publication (academic paper) database and input-output tables. Second, other than patent co-classification, other information of patents can be effectively utilized, such as citation, patent family and assignee information. Finally, the use of the IPC to define ETs may cause a problem of endogeneity in the analysis. If the patent classification changes over time, the robustness of the model may be damaged. Because IPC has been changed over time and this study analyzed all the ET-related patents registered from past to the present, fundamental problems may exist. A taxonomical classification of ETs could rectify the problem to some extent. It would be possible to use text mining with the contents of ET-related patents in order to identify the relevant keywords of ETs, construct a keyword frequency vector and define ETs by applying a cluster analysis to the keyword frequency vector. Thus, future research should address these issues.


This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2011-0007373) and also Hankuk University of Foreign Studies research fund.



Acosta, M., Coronado, D., & Fernández, A. (2009). Exploring the quality of environmental technology in Europe: Evidence from patent citations. Scientometrics, 80, 131-152.

Andersen, M. M. (2008). Eco-innovation-towards a taxonomy and a theory. In 25th Celebration DRUID Conference.

Anderson, T. R., Daim, T. U., & Kim, J. (2008). Technology forecasting for wireless communication. Technovation, 28, 602-614.

Archibugi, D., & Pianta, M. (1996). Measuring technological change through patents and innovation surveys. Technovation, 16, 451-519.

Arundel, A., & Kemp, R. (2009). Measuring eco-innovation. United Nations University Working Paper Series (2009/017), 1-40.

Asan, U., & Soyer, A. (2009). Identifying strategic management concepts: An analytic network process approach. Computers and Industrial Engineering, 56, 600-615.

Banker, R. D., Charnes, A., & Cooper, W. W. (1984). Some models for estimating technical and scale inefficiencies in data envelopment analysis. Management Science, 30, 1078-1092.

Blackman, M. (1995). Provision of patent information: A national patent office perspective. World Patent Information, 17, 115-123.

Bouyssou, D. (1999). Using DEA as a tool for MCDM: Some remarks. Journal of the Operational Research Society, 50, 974-978.

Breschi, S., Lissoni, F., & Maleraba, F. (2003). Knowledge-relatedness in firm technological diversification. Research Policy, 32, 69-87.

Carrillo-Hermosilla, J., Del Río, P., & Könnölä, T. (2010). Diversity of eco-innovations: Reflections from selected case studies. Journal of Cleaner Production, 18, 1073-1083.

Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring efficiency of decision making units. European Journal of Operational Research, 2, 429-444.

Choi, C., Kim, S., & Park, Y. (2007). A patent-based cross impact analysis for quantitative estimation of technology impact: The case of information and communication technology. Technological Forecasting & Social Change, 74, 1296-1314.

Chung, S., Lee, A., & Pearn, W. (2005). Analytic network process (ANP) approach for product mix planning in semiconductor fabricator. International Journal of Production Economics, 96, 15-36.

Creighton, C., & Hahash, S. (2003). Mining gene expression database for association rules. Bioinformatics, 19, 79-86.

Dibiaggio, L., & Nasiriyar, M. (2009). Rate and dimensions of the technological knowledge base underlying fuel cell innovations. Evidence from patent data. In S. Poguts, A. Russo, & P. Migliavacca (Eds.), Innovation, markets and sustainable energy: The challenge of hydrogen and fuel cells (pp. 87-101).

Englesman, E. E., & van Rann, A. F. J. (1992). A patent-based cartography of technology. Research Policy, 23(1), 1-26.

Ernst, H. (2003). Patent information for strategic technology management. World Patent Information, 25, 233-242.

Foray, D., & Grübler, A. (1996). Technology and the environment: An overview. Technological Change and Social Forecasting, 53, 3-13.

Fornahl, D., Broekel, T., & Boschma, R. (2011). What drives patent performance of German biotech firms? The impact of R&D subsidies, knowledge networks and their location. Papers in Regional Science, 90, 395-418.

Geum, Y., Lee, S., Kim, C., & Kim, M.-S. (2012). Technological convergence of IT and BT: Evidence from patent analysis. ETRI Journal, 34, 439-449.

Geum, Y., Lee, S., Yoon, B., & Park, Y. (2013). Identifying and evaluating strategic partners for collaborative R&D: Index-based approach using patents and publications. Technovation, 33, 211-224.

Han, J., & Kamber, M. (2001). Data mining: Concepts and techniques. San Diego, CA: Academic Press.

He, C., & Loh, H. T. (2010). Pattern-oriented associative rule-based patent classification. Expert Systems with Applications, 37, 2359-2404.

Hsieh, N. (2004). An integrated data mining and behavioral scoring model for analyzing bank customers. Expert Systems with Applications, 27, 623-633.

Jaffe, A. B. (1986). Technological opportunity and spillovers of R&D: Evidence from firms' patents, profits, and market value. American Economic Review, 76, 984-1001.

Jaffe, A. B. (1989). Characterising the technological position of firms, with application to quantifying technological opportunity and research spillovers. Research Policy, 18, 87-97.

Jiang, J. J., Zhang, L., & Wang, Y. Q. (2011). Association rules analysis of human factor events based on statistics method in digital nuclear power plant. Safety Science, 49, 946-950.

Jung, U., & Seo, D. W. (2010). An ANP approach for R&D project evaluation based on interdependencies between research objectives and evaluation criteria. Decision Support Systems, 49, 335-342.

Keasey, K., & McGuinness, P. B. (2008). Firm value and its relation to equity retention levels, forecast earnings disclosures and underpricing in initial public offerings in Hong Kong. International Business Review, 17, 642-662.

Kemp, R., & Oltra, V. (2011). Research insights and challenges on eco-innovation dynamics. Industry and Innovation, 18, 249-253.

Kemp, R., & Pearson, P. (2007). Measuring eco-innovation. Final report MEI project about measuring eco-innovation, OECD.

Kim, C., Lee, H., Seol, H., & Lee, C. (2011). Identifying core technologies based on technological cross-impacts: An association rule mining (ARM) and analytic network process (ANP) approach. Expert Systems with Applications, 38, 12559-12564.

Kumar, M. A., Ranjan, S. M., & Kumar, L. S. (2013). An improved data mining technique for classification and detection of breast cancer from mammograms Neural computing & applications, 22, 303-310.

Lee, L. T. S. (2010). On efficiency of integrative strategies among companies in US high-tech industry. International Journal of Management and Enterprise, 9, 311-323.

Lee, S., & Kim, M.-S. (2010). Inter-technology networks to support innovation strategy: An analysis of Korea's new growth engines. Innovation: Management, Policy & Practice, 12, 88-104.

Lee, H., Kim, C., Cho, H., & Park, Y. (2009). An ANP-based technology network for identification of core technologies: A case of telecommunication technologies. Expert Systems with Applications, 36, 894-908.

Lee, S., Kim, M.-S., & Park, Y. (2009). ICT Co-evolution and Korean ICT strategy: An analysis based on patent data. Telecommunications Policy, 33, 253-271.

Lee, H., Seol, H., Sung, N., Hong, Y. S., & Park, Y. (2010). An analytic network process approach to measuring design change impacts in modular products. Journal of Engineering Design, 21, 75-91.

Liao, S., & Chen, Y. (2004). Mining customer knowledge for electronic catalog marketing. Expert Systems with Applications, 27, 521-532.

Lovell, C. A. K., & Pastor, J. T. (1999). Radial DEA models without inputs or without output. European Journal of Operation Research, 118, 46-51.

Markatou, M. (2012). Measuring 'sustainable' innovation in Greece: A patent based analysis. Journal of Innovation & Business Best Practices, 1-10.

Meade, L., & Sarkis, J. (1999). Analyzing organizational project alternatives for agile manufacturing processes: An analytic network approach. International Journal of Production Research, 37, 241-261.

Metters, R. D., King-Metters, K. H., & Pullman, M. (2002). Successful service operations management. OH: South-Western College Publishers.

Meyer, M. (2007). What do we know about innovation in nanotechnology? Some propositions about an emerging field between hype and path-dependency. Scientometrics, 70, 779-810.

Nemet, G. F. (2012). Inter-technology knowledge spillovers for energy technologies. Energy Economics, 34, 1259-1270.

OECD. (1994). Using patent data as science and technology indicators - patent manual. Paris: Author.

Petruzzelli, A. M., Dangelico, R. M., Rotolo, D., & Albino, V. (2011). Organizational factors and technological features in the development of green innovations: Evidence from patent analysis. Innovation: Management, Policy & Practice, 13, 291-310.

Popp, D. (2006). International innovation and diffusion of air pollution control technologies: The effects of NOX and SO2 regulation in the U.S., Japan and Germany. Journal of Environmental Economics and Management, 51, 46-71.

Rong, J., Vu, H. Q., & Law, R. (2012). A behavioral analysis of web sharers and browsers in Hong Kong using targeted association rule mining. Tourism Management, 33, 731-740.

Saaty, T. (1996). Decision making with dependence and feedback: The analytic network process. Pittsburgh, PA: RWS Publications.

Sebastian, Y., & Then, P. H. H. (2011). Domain-driven KDD for mining functionally novel rules and linking disjoint medical hypotheses. Knowledge-based Systems, 24, 609-620.

Seol, H., Lee, S., & Kim, C. (2011). Identifying new business areas using patent information: A DEA and text mining approach. Expert Systems with Applications, 38, 2933-2941.

Shen, Y. C., Lin, G. T. R., & Tzeng, G. H. (2011). Combined DEMATEL techniques with novel MCDM for the organic light emitting diode technology selection. Expert Systems with Applications, 38, 1468-1481.

Shih, M. J., Liu, D. R., & Hsu, M. L. (2010). Discovering competitive intelligence by mining changes in patent trends. Expert Systems with Applications, 37, 2882-2890.

Tanner, A. N. (2012). The geography of emerging industry - Regional knowledge dynamics in the emerging fuel cell industry (PhD Thesis).

Tseng, F. M., Hsieh, C. H., Peng, Y. N., & Chu, Y. W. (2011). Using patent data to analyze trends and the technological strategies of the amorphous silicon thin-film solar cell industry. Technological Forecasting & Social Change, 78, 332-345.

United States Patent and Trademark Office (USPTO). (2012). Overview of the U.S. patent classification system (USPC). Retrieved November 11, 2013, from sources/classification/overview.pdf

Wang, M. Y. (2012). Exploring potential R&D collaborators with complementary technologies: The case of biosensors. Technological Forecasting & Social Change, 79, 862-874.

Weenen, T. C., Ramezanpour, B., Pronker, E. S., Commandeur, H., & Claassen, E. (2013). Foodpharma convergence in medical nutrition - Best of both worlds? PloS ONE, 8, e82609.

Wong, K. W., Zhou, S., & Yang, Q. (2005). Mining customer value: From association rules to direct marketing. Data Mining and Knowledge Discovery, 11, 57-79.

Wright, A., Chen, E. S., & Maloney, F. L. (2010). An automated technique for identifying associations between medications, laboratory results and problems. Journal of Biomedical Informatics, 43, 891-901.

Wu, F. S., Hsu, C. C., Lee, P. C., & Su, H. N. (2011). A systematic approach for integrated trend analysis - The case of etching. Technological Forecasting and Social Change, 78, 386-407.

Yoon, B., & Jeong, S. (2013). Impact analysis of biological technology: Application of network analysis and decision making trial and evaluation laboratory. Advanced Science Letters, 19, 3610-3614.

Yoon, B., & Park, Y. (2004). A text-mining based patent network: Analytical tool for hightechnology trend. Journal of High Technology Management Research, 15, 37-50.

Yüksel, I., & Dagdeviren, M. (2007). Using the analytic network process (ANP) in a SWOT analysis - A case study for a textile firm. Information Sciences, 177, 3364-3382.

[Author Affiliation]

Chulhyun Kima and Moon-Soo Kimb*

aDepartment of Technology & Systems Management, Induk University, Seoul, Republic of Korea;

bDepartment of Industrial & Management Engineering, Hankuk Uniersity of Foreign Studies (HUFS), San 89, Wangsan-ri, Mohyeon-myun, Yongin-si, Kyungki-do 447-791, Republic of Korea

(Received 2 January 2014; accepted 16 August 2014)

*Corresponding author. Email:

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.