Introduction

Patent documents are up-to-date and reliable information sources that reflect technological change (Yoon et al. 2011). Patent analysis has a narrow functional focus related only to technological knowledge of inventions, but it nevertheless has wide applicability to many different businesses (Feldman and Sanger 2007) and has been considered a vital tool for the analysis of technological trends (Yoon and Park 2004). Recently, the number of patents granted increased rapidly, and the need to analyze large quantities of patents has made relying only on experts’ knowledge to analyze technological trends almost impossible (Kostoff 1998). In response, tools for patent analysis have been developed to assist researchers and research and development (R&D) policy makers to better concentrate on their own knowledge services, including product development planning and technology strategy formulation.

One tool for patent analysis is a patent map that visualizes overall relationships among patents in a given technology. A widely adopted method for patent mapping is the keyword-based approach that constructs patent maps using the keywords and key phrase patterns that occur in a given patent set (Chang et al. 2010; Lee et al. 2009; Yoon and Park 2004; Yoon et al. 2002). By exploiting the vector space model (Salton et al. 1975) that was originally developed for information retrieval (IR), the approach predefines a set of patterns, represents each patent according to the vector codifying occurrences of the patterns, and then maps the relationships among the vectors onto a lower-dimensional space. Although the keyword-based approach has been widely used because of its simplicity and applicability, it has some substantial limitations: (1) predefining keywords and key phrases relies heavily on the knowledge of experts who may be expensive or unavailable (Choi et al. 2011; Yoon et al. 2011; Yoon and Kim 2012), (2) frequencies and co-occurrences of the patterns may not reflect substantial characteristics of inventions (Wanner et al. 2008), and (3) the patent maps may differ according to which expert was employed because experts select the keywords and key phrases (Yoon and Kim 2012). In the strictest sense, the uniqueness of an invention can be determined not by the occurrences of keywords and key phrases but by inventive key findings or fresh expertise.

For these reasons, by introducing the subject-action-object (SAO)-based approach, this paper constructs dynamic patent maps to identify technological competition trends effectively in a given technology. SAO structures are syntactically ordered sentences extractable using the natural language processing of the patent text; they are the key concepts that explicitly include technological objectives and structures of a relevant patent (Cascini et al. 2004; Mann 2002), and the set of SAO structures is considered a detailed picture of the inventor’s expertise (Moehrle et al. 2005; Park et al. 2012). In this study, we produce dynamic patent maps by combining textual and bibliographic analysis of patents and present in-depth strategic analyses of the maps to identify useful insights into technology areas in which patents have not been granted (“patent vacuums”), areas in which many patents have actively appeared (“technological hot spots”), R&D trends of competitors, and characteristics of patent clusters. The proposed SAO-based dynamic patent maps and their strategic analyses are illustrated using patents related to the synthesis of carbon nanotubes (CNTs). We expect that the proposed SAO-based dynamic patent maps and their strategic analyses can be incorporated into R&D planning processes to assist technology strategy decision makers and R&D policy makers in identifying technological competition trends.

Section “Theoretical background” presents an overview of the theoretical background. Based on the groundwork, section “SAO-based dynamic patent maps” briefly introduces a procedure for SAO-based dynamic patent maps, and section “Illustration and strategic analyses of dynamic patent maps” presents the practical analyses of the dynamic maps. Section “Conclusion and future research” concludes the paper and discusses future research topics.

Theoretical background

Constructing SAO-based dynamic patent maps requires SAO-based patent textual analysis, measurement of similarities between SAO structures, and patent mapping. Therefore, this section reviews the theoretical literature on SAO-based patent analysis, semantic sentence similarity measurement, and multidimensional scaling (MDS).

SAO-based patent analysis

The method proposed in this paper requires extracting SAO structures from patent text; these are composed of a subject (noun phrase), action (verb phrase), and object (noun phrase). This syntactically ordered sentence explicitly describes a relationship between components that appear in the relevant patent. Given a simple sentence, “soap cleans hands,” the subject is “soap,” the action is “cleans,” and the object is “hands.” The action “cleans” clearly defines a relationship between the subject “soap” and the object “hands.” SAO structures are substantially related to the concept of function, which is defined as “the action changing a feature of any object” (Savransky 2000), so subjects and objects may refer to components of a system and actions may refer to functions performed by and on components (Cascini et al. 2004; Choi et al. 2010). Furthermore, SAO structures from patent claims are considered to represent intensive knowledge related to the inventor’s expertise and the patent’s key concepts (Moehrle et al. 2005; Cascini et al. 2004).

Recently, several SAO-based patent design structure analyses have been used to visualize the design structures of patents as SAO-based networks (Cascini et al. 2004) and to identify patent similarity by measuring similarity among the SAO networks (Cascini and Zini 2008). Other SAO-based patent analyses have operated on the basis of patent-based merger and acquisition strategies, patent-based human resource decisions (Moehrle et al. 2005), patent infringement risk evaluation (Bergmann et al. 2008; Park et al. 2012), product forecasting (Gerken et al. 2010), and the identification of rapidly evolving technological trends (Yoon and Kim 2012).

Although the patent maps of this paper introduce an SAO-based patent map approach similar to that introduced in our previous works (Park et al. 2012; Yoon and Kim 2011; Yoon and Kim 2012), this paper has a differentiation in that it constructs SAO-based patent maps from a “dynamic” perspective and presents practical analyses of the dynamic maps to address several dimensions, including patent vacuums, technological hot spots, R&D overlaps of competitors, and characteristics of patent clusters.

WordNet-based semantic sentence similarity measurement

The semantic patent similarity between two patents is computed using semantic sentence similarities between pairs of SAO structures of the patents. In general, the measurement process of semantic similarity between two sentences is composed of (1) tokenizing the sentences, (2) stemming works, (3) tagging parts of speech, (4) determining the most likely meaning of each word in each sentence, and (5) computing the sentence similarity on the basis of the similarity between pairs of corresponding words (Simpson and Dao 2005; Yoon and Kim 2012). A measure of similarity between two concepts (Resnik 1999) is defined as follows:

$$ {\text{sim}}(c_{1} ,c_{2} ) = \frac{{2 \times {\text{depth}}\;({\text{lcs}}(c_{1} ,c_{2} ))}}{{{\text{depth}}\;(c_{1} ) + {\text{depth}}\;(c_{2} )}} $$
(1)

where lcs is the lowest common subsumer of two concepts, c 1 and c 2, and depth is the distance from a concept node c i to the root of a concept hierarchy. The similarity score of the two concepts is \( 0 < {\text{sim}}(c_{1} ,c_{2} ) \le 1 \), where 1 means that the concepts are identical. Building on the measure of similarity between two concepts, the matching average (Simpson and Dao 2005) to compute similarity between sentences X and Y is

$$ MatAvg(X,Y) = \frac{2 \times Match(X,Y)}{|X| + |Y|} $$
(2)

where |X| and |Y| denote the number of set tokens in sentences X and Y, respectively, and \( {\text{Match}}(X,Y) \) is the sum of similarity of the matching word tokens between the sentences. The matching average score between the two sentences is \( 0 < {\text{MatAvg}}(X,Y) \le 1 \), where 1 means that the sentences are identical. Using the WordNet semantic dictionary (Miller 1995) as a concept hierarchy, the measure for similarity measurement between two words, and the measure for similarity measurement between two sentences, a .NET-based semantic sentence similarity measurement has been implemented as an application programming interface library (Simpson and Dao 2005). Because SAO structures are complete sentences, the procedure proposed in this paper uses the C# library as a basis for the automatic measurement of semantic similarity between SAO structures.

Although an advanced semantic similarity measurement method was developed to support the exchange of information among enterprise applications (Jeong et al. 2005), it requires specific typed data, including a tree structure. However, SAO structures extracted from patents do not construct such tree structures, so we adopt the semantic sentence similarity measurement to measure similarities among patents.

Multidimensional scaling

MDS is a set of related statistical techniques often used in information visualization for exploring similarities or dissimilarities in data. Using dissimilarity between objects, or attributes of objects, MDS maps their relationships onto a lower-dimensional space. The quality of an MDS can be identified using the stress value V, ranging from 0 (no stress) to 1 (maximum stress). In general, MDS with V < 0.2 is accepted to avoid degeneration (Kruskal 1964). The strength of MDS is that it provides analyzers with a visual understanding of data. MDS has been applied to chemical engineering science maps (Peters and Van Raan 1993a, b), the evaluation of technological strategies (Schmoch 2009), and analysis of technology opportunities (Lee et al. 2009). These studies were based on co-occurrences of predefined keywords or key phrase patterns; this paper is different from previous studies in that it exploits SAO-based semantic patent similarity.

SAO-based dynamic patent maps

The procedure for SAO-based dynamic patent maps is composed of four steps (Fig. 1): (1) collecting patent data, (2) extracting SAO structures from patent text using NLP, (3) generating a patent dissimilarity matrix by computing semantic patent dissimilarities based on semantic sentence similarities between the SAO structures of patents, and (4) visualizing the patent dissimilarity matrix as dynamic patent maps by exploiting MDS and patent bibliographic information. To provide this procedure in an automated way, we developed a .NET-based prototype system that interfaces external systems including information mapping software. The prototype system provides several functions for strategic analyses of the dynamic patent maps in section “Illustration and strategic analyses of dynamic patent maps”.

Fig. 1
figure 1

Overall procedure for SAO-based dynamic patent maps

Full size image

Collection of patent data

The proposed patent map is generated using a specific patent set. In general, gathering patents requires a retrieval query composed of textual information related to a target technology and bibliographic information such as international patent codes, applicants, and application date. A final patent set for analysis is prepared by eliminating irrelevant patents.

Although patent documents have various sections, syntactic analysis of patents can use only narrative sections, including title, abstract, background summary, detailed description, and claims. Among various narrative sections, the proposed procedure uses only claims because they clearly state the innovative knowledge that requires legal protection (Fujii et al. 2007; Yoon et al. 2011). The data source is stored in an electronic format, such as Microsoft Excel, for syntactic analysis in the next step.

Syntactic analysis of patent text

In this step, SAO structures are output after the syntactic analysis of patent text. NLP parsers such as the Stanford parser (Stanford 2010) and MiniPar (Lin 2010) assist people without linguistic expertise to automatically extract the specific syntactic structures from the text. Commercial linguistic analyzers such as Knowledgist2.5™ (www.invention-machine.com) are also available. Because some of the SAO structures in patents may be duplicated or irrelevant, they are filtered out using a set of stopwords (STOPWORDS 2010); this is followed by human expert screening. Finally, SAO structures for the semantic analysis of the next step can be prepared. As a complete sentence, each SAO explicitly describes a relationship between components, that is, subjects and objects, including the tools and materials used in the relevant patent (Table 1).

Table 1 An example of SAO extraction (Patent US2008-232042)
Full size table

Measurement of SAO-based semantic patent similarities

In this step, a patent dissimilarity matrix is generated by first computing the semantic similarities between pairs of patents (Fig. 2). The semantic patent similarity between two patents is computed by measuring the semantic similarity between the SAO structures of the patents.

Fig. 2
figure 2

Concepts for computation of the patent dissimilarity matrix

Full size image

Using the C# library for the WordNet-based semantic sentence similarity measurement (Simpson and Dao 2005), semantic similarity between two SAO structures can be computed. WordNet defines the concept hierarchies of most words, including synonyms, hypernyms, and hyponyms, by word meaning: for example, the chemical symbol for the element “Mg” is the same as “magnesium,” and the noun “automobile” is a variant of the noun “vehicle”. However, WordNet does not contain abbreviations that are strongly domain-specific: for example, CNT researchers use specialized acronyms, such as “MWNT” for “multiple-walled nanotube” and “CVD” for “chemical vapor deposition”. Therefore, the C# library source code was modified to refer to a synonym set that can be grouped by investigating the SAO structures of the data source.

To determine whether two SAO structures are same, a threshold value p is used. For example, when p = 0.90 and MatAvg(SAO i , SAO j ) = 0.92, then the two SAO structures can be considered semantically the same. The determination of similarity between SAO i and SAO j is

$$ {\text{SAO}}_{ij} = \left( {\begin{array}{*{20}l} {1,\quad {\text{if}}\;{\text{MatAvg}}({\text{SAO}}_{i} ,{\text{SAO}}_{j} )\; \ge \;p} \\ {0,\quad {\text{otherwise}}} \\ \end{array} } \right. $$
(3)

Therefore, a measure for semantic similarity between patents X and Y can be defined on the basis of how many SAO structures the two patents share:

$$ {\text{SIM}}(X,Y) = \frac{{2 \times N_{\text{SAO}} (X,Y)}}{{N_{\text{SAO}} (Y) + N_{\text{SAO}} (Y)}} $$
(4)

where N SAO (X) is the number of SAO structures in patent X, and N SAO (X,Y) is the number of the semantically identical SAO structures shared by patents X and Y. The dissimilarity between patents A and B is

$$ {\text{DSIM}}(A,B) = 1 - {\text{SIM}}(A,B) $$
(5)

The semantic dissimilarity score of two patents is \( 0 \le {\text{DSIM}}(A,B) \le 1 \), where 0 means that they are identical. Using the dissimilarities between pairs of patents, a patent dissimilarity matrix can be obtained that codifies the semantic distances between patents. Given that the number of patents is M, a lower-triangular M × M patent dissimilarity matrix can be prepared.

Visualization of SAO-based dynamic patent maps

This step visualizes the relative positions of patents on the lower-dimensional spaces by considering the semantic distances between patents. For this purpose, this paper uses MDS. Various MDS algorithms such as PREFSCAL, PROXCAL, and ALSCAL are available, and SPSS 17.0, a widely used commercial statistics software package, currently provides these algorithms. Using the M × M patent dissimilarity matrix, the software outputs the relative positions of patents on a two-dimensional space. After the positions of patents are combined with the patent bibliographic information, the patent map can provide the dynamic features of a given technology with respect to application dates, applicants, and patent clusters.

Illustration and strategic analyses of dynamic patent maps

To illustrate the proposed method, this section generates dynamic patent maps using patents related to the synthesis of CNTs, which are allotropes of carbon with a cylindrical nanostructure. These cylindrical molecules exhibit extraordinary strength and unique electrical properties and are efficient thermal conductors, so they are potentially applicable to nanotechnology, electronics, optics, and material science. Over the past decade, many CNTs-related patents have been applied for. This paper considers 136 patents collected from patent databases for Europe (EU), Japan (JP), Korea (KR), and the United States (US), the top four countries in the Patent Cooperation Treaty application (WIPO 2010). This paper used a patent set identical to that in our previous study (Yoon and Kim 2011); readers can refer to the retrieval query for the collected patents from the study. These patents were related only to methods of synthesizing CNTs, so most patents were classified into nano-materials and manufacture related international patent classifications (IPC) including C01B (non-metallic elements; compounds thereof), B82B (nano-structures; manufacture or treatment thereof) and C23C (coating metallic material; coating by vacuum evaporation etc.) (Table 2).

Table 2 4 digit IPC groups of patents
Full size table

Because the SAO-based analysis in this paper considers only English text, patent text in the KR and JP databases were translated into English text using language translation services such as K2E-PAT (KIPO 2010); because the grammatical structure of Korean is identical to that of Japanese, the text of JP patents can be perfectly translated into Korean text using the Google translating program. The representation format of application numbers differs among patent databases, so the gathered patents were labeled from P1 (oldest patent JP1992-172242) to P136 (newest patent JP2009-223352) in application date order.

Section “Dynamic patent maps related to the synthesis of carbon nanotubes” provides the generated SAO-based semantic patent map and section “Strategic analyses of patent maps for R&D planning” presents ways to identify technological insights for decision supports on R&D strategies.

Dynamic patent maps related to the synthesis of carbon nanotubes

The claims of each patent were stored in a Microsoft Excel file, and Knowledgist2.5™ was then used to extract their SAO structures. After eliminating duplicate and irrelevant SAO structures, 1174 SAO structures (an average of 8.63 per patent) were extracted. Then, by examining the extracted SAO structures, we defined a synonym set that includes abbreviations and phrases with the same meaning.

The semantic patent similarities were computed on the basis of the WordNet-based semantic sentence similarities between their SAO structures. To determine whether two SAO structures are same, the threshold value p was set to 0.80 after discussion with a nano-technology domain expert. Then, dissimilarities between pairs of patents were calculated and a 136 × 136 patent dissimilarity matrix was obtained (Fig. 3), and PROXCAL of SPSS 17.0 was used to convert the matrix into a two-dimensional patent map with an MDS quality stress value of V = 0.15 (Fig. 4). In the map, patents that are near each other have a strong possibility of being technologically similar or complementary, whereas they are unrelated if they are far from each other. Although the patent map gives an intuitive understanding about a given patent set, this understanding becomes more difficult as the size of the patent set increases. Therefore, further analysis to effectively identify technological insights from the patent map is introduced in the next sub-section.

Fig. 3
figure 3

A portion of the patent dissimilarity matrix

Full size image
Fig. 4
figure 4

Visualization of the SAO-based patent map

Full size image

Strategic analyses of patent maps for R&D planning

By incorporating patent bibliographic information, the dynamic features of the SAO-based semantic patent map assist experts in effectively identifying technological insights. Specifically, this section identifies patent vacuums, technological hot spots, R&D trends of competitors, and characteristics of patent clusters. To analyze the patent map, a .NET-based application was developed; it shows the map in a dynamic sense by incorporating patent bibliographic information including application dates, applicants, and patent clusters (Fig. 4).

Patent vacuums and technological hot spots

Temporal analysis of the SAO-based patent map identifies patent vacuums and reveals how they change over time (the blue oval in Fig. 5); patent vacuums may represent opportunities for the creation of new technology. However, not all patent vacuums are promising, so several processes, including significance analysis, trend analysis, and feature analysis, are required to determine whether they actually represent technology opportunities (Lee et al. 2009). In this paper, the SAO structures of patents located around or in the vacuum areas were used to identify the characteristics of patent vacuums. In this respect, the SAO-based approach is substantially different from the keyword-based approach, which shows the occurrence frequency of keywords but does not show the technological structures of inventions or the relationships among components used in the inventions. However, the SAO-based approach can explicitly describe them using SAO structures (Appendix 1). This advantage can help experts infer the characteristics of patent vacuums. One such vacuum (Fig. 5) can be characterized by three technologies, including single-walled CNTs (P5[JP1996-003636], P7[JP1998-082409]), growth of CNTs at a low temperature (P57[JP2002-253317], P59[US2002-237729]), and microwave-based synthesis (P20[JP2000-000299]). Even though the characteristics of these technologies seem to be different from one another, combinations may provide technological opportunities; for example, methods of growing single-walled CNTs (P5, P7) can be made more cost-efficient by using low-temperature CNT synthesis (P57, P59) or by avoiding heating of the substrate (P20). Actually, this vacuum was later occupied by patent P105[KR2004-0112852], which was for an apparatus and method for growing CNTs at a low temperature that uses plasma etch gas and a catalyst; it thus demonstrated that this vacuum did represent a technically feasible combination of methods. Although this finding used the information of P105 to predict the appearance of P105, its intention is to show that SAO structures can provide significant insight with regard to the technological objectives and structures of the patents that may appear in the relevant patent vacuums. As another example, from the characteristics of patents around another patent vacuum (the green oval in Fig. 5), methods of multi-walled CNTs can be made to provide telescopic motion ability for inner tubes by combining the precursor materials of CNTs and multi-walled structures. A striking telescoping property of this multiple concentric CNTs can be used to create useful machines, including the world’s smallest rotational motor and a gigahertz mechanical oscillator.

Fig. 5
figure 5

Temporal analysis of patent maps (Vacuums: blue oval, green oval; Hot spots: red oval) (color figure online)

Full size image

Temporal analysis of the patent map can also be used to detect technological hot spots where new patents are being actively applied for. Therefore, the hot spots directly show areas in which technological development is most competitive during a specific period of time (Fig. 5). The characteristics of the hot spots can be directly identified by examining the SAO structures of the patents that each hot spot contains (Appendix 1).

In the patent database examined, the characteristics of hot spots changed over time. During 1999 and 2000, they consisted mainly of CNT synthesis methods that use conductive materials and arc-discharging (P14[JP1999-159180], P25[KR2000-0014246]); during 2003 and 2004 they consisted mainly of heat treatment synthesis methods using metal catalysts and CVD (P58[US2002-237695], P62[JP2002-314127], P64[JP2002-331816]); during 2005 and 2006 they consisted mainly of arc-discharging methods using purification control (P67[JP2002-378840], P68[KR2002-0086799], P81[JP2003-205629], P82[JP2003-371351]); and during 2007 and 2010 they consisted mainly of methods related to plasma-enhanced CVD (P104[JP2004-280179], P108[JP2005-132893], P122[JP2007-072270]) and methods related to low-temperature thermal CVD (P112[US2005-246063], P117[JP2006-224896]).

The identification of patent vacuums and hotspots provides valuable information. An early understanding of patent vacuums may provide opportunities for the creation of new technologies that have rarely been explored by competitors. In addition, identifying technological hot spots provides an understanding of up-to-date technological competition trends, so it assists technology strategy decision makers in responding promptly to technological changes. We expect that this temporal analysis can help such decision makers perform technology creativity activities and that it can be exploited to aid decision support in product development and R&D directions.

R&D trends of competitors

Identifying trends followed by patent applicants provides a better understanding of the technological competence of competitors (Moehrle and Geritz 2004). For example, if a company’s patents are located in a small area of a patent map, it means that the company is focusing on a specific technology. Conversely, if patents of a company are widely distributed on a patent map, this indicates that the company is exploring or developing new technologies. An R&D trend map of competitors including four companies and two individuals provides an intuitive understanding of the R&D activities by competitors (Fig. 6). The map shows that patents of ULVAC are far from other five competitors and that the company is developing its own technological area (Fig. 6). ULVAC is a leading company in vacuum technology, including vacuum coating and sputtering. In the last decade, this company has applied for patents (P27[JP2000-108319], P108[JP2005-132893], P119[JP2006-255299], P122[JP2007-072270]) related to plasma-enhanced chemical vapor deposition (PECVD), building on its vacuum technology (Appendix 2). SONY’s patent pattern overlaps with those of other applicants, including NKK, CHO, and JANG. Some SONY patents (P42[JP2001-191424], P80[JP2003-197339]) are for arc-discharging, synthesis-using heat traps, and some patents of CHO (P127[KR2007-0121569], P130[KR2008-0064683]) are for arc-discharging and synthesis-using rotating cathodes (Fig. 6). Although these patents have a similar technological basis, they use slightly different methods to attain high yields of CNTs (Appendix 2). The SAO-based approach provides plentiful information—that is, the gathered SAO structures—to assist experts in understanding competitors’ R&D trends. The occurrence frequency of keywords in the keyword-based approach cannot show the relationships among tools and materials used in the inventions, but SAO structures in the approach of this paper explicitly describe the technological structures and findings of inventions.

Fig. 6
figure 6

R&D trend map of six competitors. a ULVAC patents; b Overlap between SONY patents and CHO patents

Full size image

R&D trend maps of competitors visualize R&D trends of and technological overlaps among competitors. Therefore, they provide technology strategy decision makers with an intuitive understanding of the competitive outlook in a given technology area. We expect that the analysis of R&D trend maps can be used as a valuable source for technological decision making, including patent purchasing, M&A, and business partnerships.

Competing on different technologies

Using the relative position of each patent on the map, patent clusters can be identified. Managing patent clusters allows similar patents to be grouped as a technology package, which enables experts to identify the idiosyncratic characteristics of patent clusters (Yoon et al. 2002). Although several clustering algorithms can be used to identify patent clusters, this paper has adopted the k-means clustering algorithm. k-means clustering is a method of cluster analysis that aims to partition n observations into k clusters in which each observation belongs to the cluster with the nearest mean (MacQueen 1967). In k-means clustering, the best number k of clusters tends to show the smallest average intra-cluster distance and the largest average inter-cluster distance (Yoon and Park 2004). The average intra-cluster distance is the average distance between individual elements and the cluster center, and the average inter-cluster distance is the average distance between cluster centers. However, a particular cluster may not be completely separable from the others, which means that the cluster is different from the others in an overall sense. Applying the analysis of Yoon et al. (2011), the technological implications of patent clusters can be formulated. Clusters with many elements, low average intra-cluster distance and high average inter-cluster distance have a strong possibility of being technologically verified, cost-efficient, or dominant technologies. In contrast, patent clusters with a small number of elements, low average intra-cluster distance, and high average inter-cluster distance may not be a dominant or technologically verified technology but may deliver weak signals for the development of new technology.

By trial and error, we found that the patent cluster map of the CNTs synthesis showed the best clustering result when k = 12 (Fig. 7; Table 3). Using the SAO structures (Appendix 3), decision makers can easily understand that patent cluster 2 is composed of many new patents that are mainly related to PECVD technology. In addition, patent cluster 2 is small, with a low average intra-cluster distance and a high inter-cluster distance, which indicates that the cluster has a possibility of consisting of patents related to a novel method. Many patents in this cluster (P27[JP2000-108319], P108[JP2005-132893], P119[JP2006-255299], P122[JP2007-072270]) were applied for by one company and all are related to PECVD technology, which is currently part of the most promising next-generation CNT synthesis methods. Interestingly, patent cluster 11 shows a result similar to our previous study related to semantic patent network analysis; the previous study identified an emerging patent cluster composed of seven patents, including P131, P130, P129, P127, P48, P42, and P41. Patent cluster 11 of the patent cluster map includes all patents except P42. The previous work uses a cut-off value to construct patent networks and identify patent clusters using the bi-component method (Bollobas 1983), while the proposed analysis of this paper uses k-means clustering to identify technology packages that are similar in an overall sense. The previous work and this research may show slightly different analysis results because they adopt different techniques. This means that the two approaches are complementary in some sense; each approach can analyze the dimensions that the other approach does not identify.

Fig. 7
figure 7

Patent cluster map by k-mean clustering (k = 12). Colors represent clusters (Table 3) (color figure online)

Full size image
Table 3 Patent clusters and their properties
Full size table

Analyzing the characteristics of patent clusters is useful for identifying patent packages that are composed of similar patents and for perceiving technological insights provided by the patent packages. Therefore, we expect that the analysis of patent cluster maps can be incorporated into the patent portfolio management process.

Conclusions and future research

Patents are considered a proxy reflecting technological advancement, and thus, patent analysis has been a vital activity to identify technological trends and formulate strategic decisions in the R&D planning process. One tool for effective patent analysis is a patent map; the dynamic features of patent maps can provide significant technological implications that cannot be easily obtained by human experts’ qualitative analysis. In fact, many studies have provided techniques to identify technological trends using patent maps. To this end, the keyword-based approach has been widely used because of its simplicity and practicality in constructing patent maps. However, the keyword-based approach has some limitations in that it cannot effectively represent key technological concepts, including the structures and objectives of patents. For this reason, by introducing an SAO-based approach over the keyword-based approach in specific dimensions, this paper presents practical functions and capabilities of patent maps based on SAO structures of patents from a “dynamic” perspective. Proposed strategic analyses of the dynamic patent maps include several functions that analyze technological vacuums, technological hot spots, R&D overlaps of competitors, and competition on different technologies on patent maps. The strategic analyses of SAO-based dynamic patent maps are illustrated using patents related to the CNTs synthesis method. We expect that the functions of SAO-based dynamic patent maps will aid experts in identifying technological competition trends in the process of formulating R&D strategies.

Despite this contribution, this study has its limitations. First, this paper used the MDS to visualize the patent maps for strategic analyses and adopted the k-means clustering algorithm, a trial-and-error approach to group patents, for patent clustering. However, the MDS can cause information loss on distances between patents, and k-means clustering is considered a simple technique that is often not effective. Although the MDS and k-means clustering allowed us to identify reasonable technological implications, future research will adopt advanced techniques, such as k-medoids clustering (Park and Jun 2009) without dimension reduction and hierarchical clustering (Hair et al. 2010), to obtain more valid and effective analysis results. Second, although this research applied the strategic analyses to the synthesis of CNTs, further research will extend the analyses to strategic decision making for other fiercely competitive technology fields, including solar cells and flexible display technology. Finally, this paper only applied SAO-based semantic patent similarity to the development and analysis of patent maps. However, in our future research, we will apply the method to other applications, such as patentability analysis, strategic partner selection, and patent licensing opportunities.

Appendices

Because a full list of SAO structures of all patents requires too much space, only selected SAO structures of patents are listed in the following tables.

Appendix 1: A portion of SAO structures for analysis of vacuum areas and hot spots

 

Patent[real application number]

S (subject)

A (action)

O (object)

P5[JP1996-003636], P7[JP1998-082409]

Carbon

Use

Electrode metal additive

Collecting soot

Characterize

Single layer

Soot

Contain

Single walled carbon nanotubes

Plasma

Supply

Powdered metal catalysts

P57[JP2002-253317], P59[US2002-237729]

Arc discharge

Synthesize

Carbon nanotubes

Low-temperature thermal CVD

Comprise

Deposition precipitation

Carbon source gas

Comprise

Hydrocarbon or carbon monoxide

P20[JP2000-000299]

ECR microwave generation system

Generate

Plasma deposition chamber

Formation

Characterize

Thin film forming method

Gas supply system

Supply

Hydrogen gas

P105[KR2004-0112852]

Deposition gas

Form

Carbon-containing gas and hydrocarbon gas plasma etching gas mixture

Plasma

Etch

Gas particulate catalyst

Method of carbon nanotubes

Generate

Multiple electrodes of multi-electrode

P14[JP1999-159180]], P25[KR2000-0014246]

Cathode

Comprise

Conductive material

Cathode surface

Have

Sediment generated surface cathode

Mixture of carbon and reactive gas or inert gas

Characterize

Atmosphere

Growth of carbon nanotubes

Use

Carbon nanotube growth substrate

P58[US2002-237695], P62[JP2002-314127], P64[JP2002-331816]

Coating

Form

Bonding metal layer

Carbon source gas

Comprise

Hydrocarbon or carbon monoxide

Electroplating

Form

Metal catalyst layer

Metal catalyst layer

Comprise

Fe, Co, Ni and alloy

Vacuum sputtering

Form

Metal catalyst layer and metal support layer

Heat treatment temperature

Characterize

Carbon nanotubes

Spraying inert gas from interior of tubular cathode method

Produce

Carbon

P67[JP2002-378840], P68[KR2002-0086799], P81[JP2003-205629], P82[JP2003-371351]

Arc discharge of carbon electrode

Generate

Carbon soot

Purity of carbon

Characterize

Carbon nanotubes

Evaporation and heat

Relax

Mixed carbon electrode catalyst metal

Cooling thermal relaxation

Produce

Acid-treated carbonaceous product

Pulsed arc discharge

Trap

Carbon nanotubes

P104[JP2004-280179], P108[JP2005-132893], P122[JP2007-072270]

Substrate holder disposed in reaction chamber

Apply

Bias voltage

Plasma heating means

Characterize

Mesh

Remote plasma shielding member

Characterize

Substrate, Mo, Ti, W and WC

Sponge growth

Characterize

Plasma fabrication of carbon nanotube

P112[US2005-246063], P117[JP2006-224896]

Low temperature

Comprise

Thermal chemical vapor deposition process

Circulating reaction tube

Cool

Refrigerant

Infrared radiation condenser tube

Heat

Lamp

Reaction tube method

Produce

Carbon nanotubes

Appendix 2: A portion of SAO structures for R&D trend analysis of competitors

 

Patent[real application number]

S (subject)

A (action)

O (object)

P42[JP2001-191424], P80[JP2003-197339]

Cold cathode

Characterize

Electron emitter field emission device

Cold cathode field emission device

Have

Tube shape

Heating temperature

Form

Electron emission

Inner surface of electrode

Have

Hollow part

Guide roller

Constitute

Nanotube manufacturing method

P127[KR2007-0121569], P130[KR2008-0064683]

Cathode rod movement guidance

Engage

Anode rod

Consumable carbon rods gripper

Store

Cases

Interior of non-consumable carbon plate

Move

Negative currents

Mobile guide rod thread

Engage

Connecting member

Scraper

Rotate

Magnetic field

P108[JP2005-132893], P119[JP2006-255299]

Plasma surface treatment method

Generate

Film

Film

Include

Diamond particles

Temperature measuring unit

Measure

Temperature of chemical deposition system

Plasma

Not touch

Substrate

Appendix 3: A portion of SAO structures for characteristic analysis of patent clusters

 

Patent[real application number]

S (subject)

A (action)

O (object)

P27[JP2000-108319], P108[JP2005-132893], P119[JP2006-255299], P122[JP2007-072270]

Slot antenna

Have

Microwave ECR plasma chemical vapor deposition

Thin film

Use

Slot antenna

Plasma heating means

Characterize

Mesh

Contacting narrowing of anode

Initiate

Arc discharge

Inserting anode

Provide

Cooling

Main body of anode

Consist of

Cylinder

Passing current through electrodes

Vaporize

Carbon

Remote plasma shielding member

Characterize

Substrate, Mo, Ti, W and WC

Sponge growth

Characterize

Plasma fabrication of carbon nanotube