- Email: [email protected]
Visualization of patent analysis for emerging technology
Available online at www.sciencedirect.com
Expert Systems with Applications Expert Systems with Applications 34 (2008) 1804–1812 www.elsevier.com/locate/eswa
Visualization of patent analysis for emerging technology Young Gil Kim, Jong Hwan Suh, Sang Chan Park
*
Department of Industrial Engineering, Korea Advanced Institute of Science and Technology, Guseong-dong, Yuseong-gu, Daejeon 305-701, South Korea
Abstract Many methods have been developed to recognize those progresses of technologies, and one of them is to analyze patent information. And visualization methods are considered to be proper for representing patent information and its analysis results. However, current visualization methods for patent analysis patent maps have some drawbacks. Therefore, we propose an alternative visualization method in this paper. With colleted keywords from patent documents of a target technology field, we cluster patent documents by the k-Means algorithm. With the clustering results, we form a semantic network of keywords without respect of filing dates. And then we build up a patent map by rearranging each keyword node of the semantic network according to its earliest filing date and frequency in patent documents. Our approach contributes to establishing a patent map which considers both structured and unstructured items of a patent document. Besides, differently from previous visualization methods for patent analysis, ours is based on forming a semantic network of keywords from patent documents. And thereby it visualizes a clear overview of patent information in a more comprehensible way. And as a result of those contributions, it enables us to understand advances of emerging technologies and forecast its trend in the future. Ó 2007 Elsevier Ltd. All rights reserved. Keywords: Visualization; Patent analysis; k-Means clustering; Semantic network; Ubiquitous computing technology
1. Introduction It has been a critical issue to understand technological trends not only to avoid unnecessary investment but also to gain the seeds for technological development. So, many methods have been developed to recognize those progresses of technologies, and one of them is to analyze patent information. However, it is hard for non-specialists to analyze patent information because patent information is enormous and rich in technical and legal terminology. Therefore, patent information needs to be transformed into something simpler and easier to understand. And visualization methods are considered to be proper for representing patent information and its analysis results. Generally, visualization methods are known as one of the best data mining ways to understand because graphical display methods often offer superior result compared to other
*
Corresponding author. Tel.: +82 42 8692920; fax: +82 42 8693110. E-mail address: [email protected] (S.C. Park).
0957-4174/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2007.01.033
conventional techniques (Westphal & Blaxton, 1998). Especially to top managers who decide directions of technology investments, visualization methods are more useful than conventional ways such as textual, tabular, and list for quick and easy knowledge discovery (Ganapathy, Ranganathan, & Sankaranarayanan, 2004). Those visualization methods for patent analysis are called a patent map at large. A patent map is the visualized expression of total patent analysis results to understand complex patent information easily and effectively. And it is produced by gathering related patent documents of a target technology field, processing, and analyzing them (WIPO, 2003). In general, a patent document contains dozens of items for analysis which are classified into structured and unstructured item groups. Structured items are uniform in semantics and in format across a patent document such as patent number, filing date, or investors. On the other hand, the unstructured items are free texts and quite different in length and content for a patent document like claims, abstracts, or descriptions of the invention. The visualized analysis results of the former items are called
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
patent graphs, and those of the latter are called patent maps, although loosely a patent map may refer to both cases (Liu, 2003) Likewise, current visualization methods for patent analysis are based on mapping patent information and its analysis results. However, current patent maps have some drawbacks. First, most of them are time based, ranking based or matrix maps which consider only one aspect between structured items or unstructured items of each patent document. More integrated and balanced visualization approach is required to provide the overall structure of patent information effectively. Second, they are complicated networks of patent documents though they use different methods. As a result, they make patent analysis results incomprehensible and unclear to analyzer. Consequently, those deficient patent maps fail to provide an intuitive insight into the concerned technology field. And this is the third drawback we’d like to make mention of. Especially for an emerging technology like ubiquitous computing and bio informatics, it is essential to recognize its advance and make an estimate of the hereafter. Hence, we propose an alternative visualization method for patent analysis to overcome drawbacks of current patent maps. Our approach contributes to establishing a patent map which considers both structured and unstructured items of a patent document. We expect to keep the balance of analysis features by using filing dates as structured items and keywords and its frequency as unstructured items. Besides, differently from previous visualization methods for patent analysis, ours is based on forming a semantic network of keywords from patent documents. And thereby it visualizes a clear overview of patent information in a more comprehensible way. And as a result of those contributions, it enables us to understand advances of emerging technologies and forecast its trend in the future. The rest of the paper is structured as follows. In Section 2, we introduce related works with visualization methods for patent information and its analysis results. In Section 3, we explain an overview of our approach. In Section 4, we apply our visualization method to develop a patent map of the ubiquitous computing technology as an emerging technology field. And in Section 5, finally we conclude the paper with a discussion of the proposed patent map’s implications in the ubiquitous computing technology. 2. Literature review Related to researches using patent information, there are two mainstreams. One of them is to study visualization methods for patent information and its analysis results. And in this paper, we are interested in the former research area. This research area has attracted attention of the persons concerned. That’s because current technological development necessitates it to avoid unnecessary investment as well as gaining the seeds for technological develop-
1805
ment and the applicable fields. Also, the attention is increased to promote the efficient use of patent information by deepening related institutions’ understanding of patent information. On the basis of these awareness, the Japan Patent Office has been producing and providing more than 50 types of expressions and more than 200 maps for several technology fields since 1997 (JIII, 2000). In addition, many other countries such as Korea (Ryoo & Kim, 2005), Italy (Camus & Brancaleon, 2003; Fattori, Pedrazzi, & Turra, 2003) and the USA (Morris, DeYong, Wu, Salman, & Yemenu, 2002) also provide many kinds of patent maps. Currently, most patent analyses use patent citations to represent the meaningful relationship in patent information. But it is known that patent citation analysis has some serious drawbacks. According to Yoon and Park (2004), there are four drawbacks of patent citation analysis described. First, it is difficult to grasp the overall relationship among patent documents. Second, related to the first problem, the scope of analysis and the richness of potential information are limited. Third, citation has no capability of considering internal relationship between patent documents. Finally, citation analysis is a time-consuming task because it needs only an exhaustive search. So, Yoon and Park (2004) proposed a network-based patent analysis as an alternative method. But the network patent analysis still has some limitations. Researches on the intelligent methods for patent analysis have been made as well. The neural methods for mapping scientific and technical information (articles, patents) and for assisting a user in carrying out the complex process of analyzing large quantities of such information are concerned by Lamirel, Shehabi, Hoffmann, and Francois (2002). Based on text mining techniques, Tseng, Wang, Juang, and Lin (2005) created a real world patent map for an important technology domain: carbon nano-tube experimentally. And the other mainstream is concerned about patent classification. By Black and Ciccolo (2004), machine learning technology is applied to text classification on United States’ patent information to automatically differentiate between patents relating the biotech industry and those unrelated. Fall, To¨rcsva´ri, Fie´vet, and Karetka (2004) reported the results of applying a variety of machine learning algorithms for training expert systems in Germanlanguage patent classification tasks. And Trappey, Hsu, Trappey, and Lin (2006) developed a platform for patent document classification and search using a back-propagation network. 3. Methodology Our visualization method steps are summed up as follows. With colleted keywords from patent documents of a target technology field, we cluster patent documents by the k-Means algorithm. With the clustering results, we form a semantic network of keywords without respect of
1806
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
3.2. Clustering patent documents with keywords Extracting keywords from patent documents related to a target technology field
Clustering patent documents with keywords using k-Means algorithm
Forming a semantic network of keywords
Forming Patent Map
Fig. 1. Framework of visualization method for patent analysis.
To cluster patent documents with merged keywords, several steps are required (see Fig. 3). First, we check existence of each keyword within texts of a patent document. So we form a keyword existence matrix with a column index of keywords (1, . . ., j, . . ., n) and a row index of patent documents (1, . . ., i, . . ., m). If j-keyword exists within texts of i-patent document, then an element of (i, j) is filled with the number of 1. But if it does not, then the element of (i, j) is filled with the number of 0. As a result, we make a keyword existence matrix of which elements are filled with 0 or 1. Next step is to cluster patent documents by k-Means algorithm using the keyword existence matrix. Here each keyword’s value between 0 or 1 plays as a feature’s value for a patent document. So the keywords’ values are used to classify patent documents into k groups (k P 2). 3.3. Forming semantic networks of keywords
filing dates. And then we build up a patent map by rearranging each keyword node of the semantic network according to its earliest filing date and frequency in patent documents. An overview of our approach is described in Fig. 1. 3.1. Collecting keywords Our approach begins by targeting a domain technology which analyzers are interested in. And then initial keywords needs to be collected from experts to search related patent documents. After searching patent documents, we collect keywords from patent documents. And then, they are merged with the initial keywords. As a result, the list of merged keywords are completed to be used for next steps (see Fig. 2).
Now with clustered patent documents, we investigate what keyword each group has. For example, let’s assume that patent documents ‘A’ and ‘B’ belong to the group 1. According to the matrix in Fig. 4, patent document ‘A’ has keywords of ‘a’ and ‘c’. And patent document ‘B’ has keywords of ‘b’ and ‘c’. Then, the group 1 consists of three keywords of ‘a’, ‘b’, and ‘c’. Like this way, we investigate keywords for each group (see Fig. 4). And using the list of keywords for each group, we make a semantic network. How to form a semantic network is described in Fig. 5. According to it, group 1 has keywords of ‘a’, ‘b’, and ‘c’. On the other hand, group 2 has keywords of ‘c’ and ‘d’. Then two groups share ‘b’, and therefore relationship between two groups can be represented by three nodes: (a, b), (c), and (d). Here the shared node is higher than the others, so arrows are drawn from (c) to (a, b)
Target a concerned field of technology
Inquire recommendable keywords from experts
Search patent documents using the recommended keywords
Extract keywords predefined by inventors from the patent documents
Merge the recommended and Patent document databases
predefined keywords
Fig. 2. Extracting keywords from patent documents.
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
Check existence of keywords occurring in a patent document
Form a keyword existence matrix
1807
k-Means algorithm
Fig. 3. Clustering patent documents using k-Means algorithm.
a
b
c
d
A
1
0
1
0
B
0
1
1
0
C
0
0
1
1
D
0
0
0
1 a, b, c
A keyword existence matrix
c, d
Group 1
Group 2
Investigation of keywords for each group
A, B
C, D
Group 1
Group 2
Clustered patent documents: (A, B), (C, D)
Fig. 4. Investigating keywords belonging to clustered groups.
a, b, c a, b, c
c, d
Group 1
Group 2
c, d
a, b
c
d
Investigation of keywords for each group
keyword Group 1
a, b, c
Group 2
c, d
Level 0
c
List of keywords for each group
a, b
d
Level 1
Formation of a semantic network using keywords
Fig. 5. Forming a semantic network of keywords.
and (d). Like this way, we make a semantic network which consists of nodes with a keyword or more than two keywords.
The semantic network described in this section is based on the previous steps such as ‘clustering patent documents with the k-Means algorithm’ and ‘investigating key words
1808
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
for clustered patent documents’. Therefore, the semantic network is dependent on the number of groups which is set temporarily by the k-Means algorithm, and there can be so many semantic networks. There are many executable programs which can perform the k-Means algorithm. Using any of them, easily we can repeat the clustering with increasing the number of groups. And for each time based on the clustering result, we repeat both steps of ‘investigation of keywords for each group’ and ‘formation of a semantic network’. As a result of n-repetitions, we get n-semantic networks. Of course, the number of groups must be more than 1. Now we have to choose one of the n-semantic networks. Usually we select one which explains the most of the relations of keywords. And this is a manual operation. But usually as the number of groups in the chosen semantic network increases, it gets better to explain the relations of keywords by the semantic network. However, too big number makes it worse to form a semantic network therefore we have to find a point of comprise. Actually, there can be a number of ways to find the proper number of groups for the semantic network. However, in this paper, we do not include it as our concern and leave it as a further work. For the semantic network obtained, we have to investigate each node’s frequency in the clustered groups of patent documents. For each node, it is defined by counting the number of keywords’ existences in the clustered groups of patent documents. For example, in Fig. 6, ‘c’ of node 1 appears in group 1 and 2. So the frequency of node 1 is 2. The frequency of node 2 is 1 because both ‘a’ and ‘b’ appear in group 1. And the node 3’s is 1 because ‘d’ belongs
a
b
c
d
A
1
0
1
0
B
0
1
1
0
C
0
0
1
1
D
0
0
0
1
node 1 (c)
node 2 (a, b)
node 3 (d)
Fig. 6. The completed semantic network of keywords.
only to group 1. Like this way, we add a frequency of each node in the semantic network of keywords. 3.4. Forming a patent map Since now, we have explained how to form a semantic network of keywords and their frequencies as unstructured items from patent documents related to the target technology. From now on, we explain how to make use of structured items in patent documents to complete a patent map on the basis of semantic networks. Let’s assume that finally we reached the semantic network as shown in Fig. 7. We have to investigate a filing date of each node in the semantic network. The filing date of each node is the earliest filing date among patent documents which have keywords of the node. For example, in Fig. 8, node 2 consists of keywords of ‘a’ and ‘b’. And if ‘a’ belongs to patent documents of ‘A’, and ‘b’ belongs to patent documents ‘B’, then the filing date which node 1 has is the earliest filing date among document ‘A’ and ‘B’. Therefore, the filing date of node 2 is ‘1997-11-27’. Similarly, the filing
c
a, b
d
Filing date A
1997-11-27
B
2000-05-14
C
2001-03-07
D
2000-08-18
node 1 (c) 1997-11-27 node 2 (a, b) 1997-11-27
node 3 (d) 2000-08-18
A semantic network with filing date information Fig. 7. Forming a semantic network with filing date information.
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
1809
c keyword Group 1
a, b, c
Group 2
c, d
a, b
d
keyword Node 1
c
node 1
Node 2
a, b
2
Node 3
d
node 2
node 3
1
1
A semantic network with frequency information
Fig. 8. Forming a semantic network with frequency information.
date of node 3 is ‘2000-08-18’. Like this way, we add a filing date of each node in the semantic network of keywords. By these two steps, the semantic network has both aspects of structured items – filing dates, and unstructured items – keywords and their frequencies in patent documents. And now we move on to the next stage for building up a patent map using the accomplished semantic network. The patent map is completed by arranging each node of the semantic network according to its filing date in an x-direction and their frequencies in a y-direction respectively. Fig. 9 shows an example of the proposed patent map.
4. Application to ubiquitous computing technology as an emerging technology The term, ubiquitous computing, was coined more than 10 years ago by Mark Weiser who, at that time, was the chief scientist at the Xerox Palo Alto Research Center. Weiser defined ubiquitous computing as the method of enhancing computer use by making many computers available throughout the physical environments, but making them effectively invisible to the user (Weiser, 1991).
frequency
node 1 (c) 2
2
1997-11-27 node 2 (a, b) 1 1997-11-27
1997-11-27
node 3 (d) 1 2000-08-18 1
A semantic network with filing date information
node 1 (c)
node 2 (a, b)
node 3 (d)
1997-11-27
2000-08-18
1997
2000
The proposed patent map
Fig. 9. Forming a patent map using a semantic network.
filing date
1810
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
In the long term, ubiquitous computing is expected to take on great economic significance (Fleisch, 2005). Numerous applications in the business environment will become possible as physical and informational world continue to merge. And thus, additional information on
objects, processes, and individuals may be gathered, exchanged and proposed in a cost efficient way (Mu¨ller & Zimmermann, 2002). Ubiquitous computing triggers such functions as increased transparency, differential pricing and disintermediation of the value chain (Clemons & Hitt, 2000). According to these circumstances, a lot of patents are being invented all around world as a result of the advent of ubiquitous computing technology. To survive as a competitive leader in the market of ubiquitous computing technology, it is important to analyze patent information related to ubiquitous computing technology. Hence, in this paper we targeted ubiquitous computing technology as an emerging technology. And we applied our visualization method for building its patent map. Afterwards, we performed steps in Section 3 to apply our method and to build up a patent map. Their results for those steps are as follows. Related to Section 3.1, first we collected keywords recommended from experts related to ubiquitous computing technology. And then we searched patent documents related to ubiquitous computing technology using those
Table 1 The list of merged keywords for patent analysis Number
Keyword
Earliest filing date
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
RFID Universal PnP Trigger HAVI HTML M VXML interchangeability Fabrication Shop floor Magnetic memory device Logistics Automatic identification PDA, mobile, handheld device Intelligent Remote Control System GPS Ubiquitous computing Sensor network Smart Identification Manufacturing Distribution Lifecycle Healthcare Blue tooth Tracking Context awareness Inventory
1987-04-07 2002-10-01 2002-08-22 2002-08-22 2003-04-02 2002-08-22 1987-08-18 2003-04-30 1987-04-07 1987-04-07 2002-05-21 2002-12-27 2002-10-02 2002-05-21 2000-06-02 2001-01-31 2001-03-15 1987-04-07 1997-08-22 2000-06-02 2002-08-22 1987-08-18 2002-08-22 1991-12-24 2002-07-29 1987-04-07
Table 2 The list of keywords in each group of clustered patent documents Group
Keyword
1 2 3 4 5
1, 1, 2, 1, 1,
2, 3, 9, 10, 11, 13, 14, 15, 16, 18, 19, 20, 21, 24, 25, 26 3, 4, 6, 9, 15, 17, 18, 20, 21, 23, 25, 26 4, 5, 8, 10, 11, 12, 13, 14, 15, 17, 19, 20, 23, 24, 25 4, 7, 9, 10, 11, 15, 16, 17, 18, 19, 20, 22, 23, 24, 25, 26 11, 12, 13, 15, 16, 19, 20, 23, 24, 26
node 1 Ubiquitous Computing Distribution
node 2
node 3
node 4
node 5
node 6
node 7
Context awareness
RFID
Inventory
Blue Tooth
PDA Mobile device
Manufacturing Tracking
node 8
node 9
Logistics Identification
HAVI Smart
node 10 Sensor Network
node 11 Automatic Identification
node 13
node 14
Trigger Lifecycle
Universal PnP GPS
node 12 Remote Control System
node 16
node 17
node 18
Fabrication
Shop floor Healthcare
HTML VXML interchangeability Magnetic memory device
Fig. 10. A semantic network of nodes containing keywords.
node 15 Intelligent
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
recommended keywords. As a result, totally 96 patent documents were searched. And then, we investigated predefined keywords from the 96 patent documents. And then, we merged recommended and predefined keywords. The final list of merged keywords for ubiquitous computing technology is as shown in Table 1. As described in Section 3.2, we checked each keyword’s existence in the searched patent documents. Based on the result, we clustered 96 patent documents with the k-Means algorithm of ClementineTM. And then, we made semantic networks with increasing the number of groups, and selected a semantic network with five groups. The list of keywords for each group is shown in Table 2. Related to Section 3.3, we completed the final semantic network of keywords using the result of clustering (see Fig. 10). And then based on the semantic network, we investigated earliest filing date and frequency of each node in the semantic network (see Table 3). According to Section 3.4, with filing date and frequency of each node, the semantic network is transformed into the patent map for the ubiquitous computing technology (see Fig. 11). From the patent map, we see technologies related to ubiquitous computing technology have progressed towards HTML ! VXML interchangeability and magnetic memory devices in 2003 since the patents related to
1811
Table 3 Filing date and frequency for each node of the semantic network in Fig. 10 Node
Filing date
Frequency
Keyword (number)
1
2000-06-02
5
2 3 4 5 6 7
2007-07-29 1987-04-07 1987-04-07 2002-08-22 2002-05-21 1991-12-24
4 4 4 4 4 4
8
1987-04-07
3
9
2001-03-15
3
10 11 12 13
2001-01-31 1987-04-07 2002-10-02 2002-08-22
3 3 3 2
14
2002-05-21
2
15 16 17
2002-12-27 2002-08-22 1987-08-18
2 1 1
18
2003-04-02
1
Ubiquitous computing (15) Distribution (20) Context awareness (25) RFID (1) Inventory (26) Blue tooth (23) PDA, mobile, handheld device (11) Manufacturing (19) Tracking (24) Logistics (9) Identification (18) HAVI (4) Smart (17) Sensor network (16) Automatic identification (10) Remote Control System (13) Trigger (3) Lifecycle (21) Universal PnP (2) GPS (14) Intelligent (12) Fabrication (6) Shop floor (7) Healthcare (22) HTML M VXML interchangeability (5) Magnetic memory device (8)
frequency
5
2000-6-2 Ubiquitous Computing Distribution 2002-5-21 PDA Mobile device
1987-4-7 Inventory
2002-8-22 Blue Tooth
1991-12-24 Manufacturing Tracking
4
1987-4-7 RFID
2002-7-29 Context awareness
1987-4-7 Automatic Identification 3
2001-1-31 Sensor Network
1987-4-7 Logistics Identification
2002-10-2 Remote Control System
2001-3-15 HAVI Smart
2002-8-22 Fabrication 2
1
2002-5-21 Universal PnP GPS 2002-8-22 Trigger Lifecycle
1987-8-18 Shop floor Healthcare
1987
…
1991
…
2000
2001
2002
Fig. 11. A patent map based on a semantic network of Fig. 10.
2002-12-27 Intelligent
2003-4-2 HTML VXML interchangeability Magnetic memory device year 2003
1812
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
automatic identification, inventory, RFID, and logistics appeared in 1987. 5. Conclusions and further works In this paper, we proposed a new visualization method for patent analysis to overcome drawbacks of current patent maps. Comparing to the other methods in the literature of Section 2, our approach considered both sides of structured items and the unstructured items of patent documents. Thereby it provided an integrated and balanced approach to analyze patent information. Moreover, we suggested building up a patent map based on a semantic network of keywords from patent documents with the kMeans algorithm. So it could visualize a clear overview of patent information in a more comprehensible way. Finally, using the suggested framework of a visualization method for a patent map, we completed a patent map for ubiquitous computing technology as an emerging technology. From the patent map, we can find what kinds of patents on the ubiquitous computing technology have appeared and how those patents are merged and divided as time passes. Likewise, the proposed patent map gives a complete view of emerging technology’s advance. Also it helps us to have an insight to the technology field, thereby to avoid unnecessary investments and find the seeds for the next patent. As a further work, we had like to modify our approach into more sophisticated one with up-to-date data mining techniques. Especially, to make our visualization method more concrete, we plan to investigate methods which can determine the number of clustering groups in Section 3.3. Moreover, we plan to apply our approach to other emerging technologies in addition to ubiquitous computing technologies. References Black, D., & Ciccolo, P. (2004) Machine learning for patent classification. Camus, C., & Brancaleon, R. (2003). Intellectual assets management: From patents to knowledge. World Patent Information, 25(2), 155–159.
Clemons, E. K., & Hitt, L. M. (2000). The Internet and the future of financial services: Transparency, differential pricing, disintermediation. Discussion Draft, University of Pennsylvania, The Warton School. Fall, C. J., To¨rcsva´ri, A., Fie´vet, P., & Karetka, G. (2004). Automated categorization of German-language patent documents. Expert Systems with Applications, 26(2), 269–277. Fattori, M., Pedrazzi, G., & Turra, R. (2003). Text mining applied to patent mapping: A practical business case. World Patent Information, 25(4), 335–342. Fleisch, E. (2005). Ubiquitous network societies: Their impact on the telecommunication industry. Background Paper in ITU Workshop. Ganapathy, S., Ranganathan, C., & Sankaranarayanan, B. (2004). Visualization strategies and tools for enhancing customer relationship management. Communication of ACM, 47(11), 92–99. Japan Institute of Invention and Innovation. (2000). Guide book for practical use of patent map for each technology field. Lamirel, J. C., Shehabi, S. A., Hoffmann, M., & Francois, C. (2002). Intelligent patent analysis through the use of a neural network: Experiment of multi-view point analysis with the MultiSom model. In Proceedings of the ACL-2003 workshop on patent corpus processing. Liu, S. J. (2003). A route to a strategic intelligence of industrial competitiveness. The first Asia-Pacific Conference on Patent Maps, 2–13. Morris, S., DeYong, C., Wu, Z., Salman, S., & Yemenu, D. (2002). DIVA: A visualization system for exploring documents databases for technology forecasting. Computers & Industrial Engineering, 43(4), 841–862. Mu¨ller, C. D., & Zimmermann, H. D. (2002). Beyond mobile: Research topics for upcoming technologies in the insurance industry. In Proceedings of the 36th Hawaii international conference on system sciences. Ryoo, J. H., & Kim, I. G. (2005). Workshop H – What patent analysis can tell about companies in Korea, Far East Meets West in Vienna. Trappey, A. J. C., Hsu, F. C., Trappey, C. V., & Lin, C. I. (2006). Development of a patent document classification and search platform using a back-propagation network. Expert Systems with Applications, 31(4), 755–765. Tseng, Y. H., Wang, Y. M., Juang, D. W., & Lin, C. J. (2005). Text mining for patent map analysis. In IACIS Pacific 2005 conference proceedings (pp. 1109–1116). Weiser, M. (1991). The computer of the twenty-first century. Scientific America, 94–100. Westphal, C., & Blaxton, T. (1998). Data mining solution. New York: Wiley. WIPO. (2003). Patent map with exercises (related). WIPO-MOST intermediate training course on practical intellectual property issues in business, Theme 16. Yoon, B., & Park, Y. (2004). A text-mining-based patent network: Analytic tool for high-technology trend. The Journal of High Technology Management Research, 15(1), 37–50.
Expert Systems with Applications Expert Systems with Applications 34 (2008) 1804–1812 www.elsevier.com/locate/eswa
Visualization of patent analysis for emerging technology Young Gil Kim, Jong Hwan Suh, Sang Chan Park
*
Department of Industrial Engineering, Korea Advanced Institute of Science and Technology, Guseong-dong, Yuseong-gu, Daejeon 305-701, South Korea
Abstract Many methods have been developed to recognize those progresses of technologies, and one of them is to analyze patent information. And visualization methods are considered to be proper for representing patent information and its analysis results. However, current visualization methods for patent analysis patent maps have some drawbacks. Therefore, we propose an alternative visualization method in this paper. With colleted keywords from patent documents of a target technology field, we cluster patent documents by the k-Means algorithm. With the clustering results, we form a semantic network of keywords without respect of filing dates. And then we build up a patent map by rearranging each keyword node of the semantic network according to its earliest filing date and frequency in patent documents. Our approach contributes to establishing a patent map which considers both structured and unstructured items of a patent document. Besides, differently from previous visualization methods for patent analysis, ours is based on forming a semantic network of keywords from patent documents. And thereby it visualizes a clear overview of patent information in a more comprehensible way. And as a result of those contributions, it enables us to understand advances of emerging technologies and forecast its trend in the future. Ó 2007 Elsevier Ltd. All rights reserved. Keywords: Visualization; Patent analysis; k-Means clustering; Semantic network; Ubiquitous computing technology
1. Introduction It has been a critical issue to understand technological trends not only to avoid unnecessary investment but also to gain the seeds for technological development. So, many methods have been developed to recognize those progresses of technologies, and one of them is to analyze patent information. However, it is hard for non-specialists to analyze patent information because patent information is enormous and rich in technical and legal terminology. Therefore, patent information needs to be transformed into something simpler and easier to understand. And visualization methods are considered to be proper for representing patent information and its analysis results. Generally, visualization methods are known as one of the best data mining ways to understand because graphical display methods often offer superior result compared to other
*
Corresponding author. Tel.: +82 42 8692920; fax: +82 42 8693110. E-mail address: [email protected] (S.C. Park).
0957-4174/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2007.01.033
conventional techniques (Westphal & Blaxton, 1998). Especially to top managers who decide directions of technology investments, visualization methods are more useful than conventional ways such as textual, tabular, and list for quick and easy knowledge discovery (Ganapathy, Ranganathan, & Sankaranarayanan, 2004). Those visualization methods for patent analysis are called a patent map at large. A patent map is the visualized expression of total patent analysis results to understand complex patent information easily and effectively. And it is produced by gathering related patent documents of a target technology field, processing, and analyzing them (WIPO, 2003). In general, a patent document contains dozens of items for analysis which are classified into structured and unstructured item groups. Structured items are uniform in semantics and in format across a patent document such as patent number, filing date, or investors. On the other hand, the unstructured items are free texts and quite different in length and content for a patent document like claims, abstracts, or descriptions of the invention. The visualized analysis results of the former items are called
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
patent graphs, and those of the latter are called patent maps, although loosely a patent map may refer to both cases (Liu, 2003) Likewise, current visualization methods for patent analysis are based on mapping patent information and its analysis results. However, current patent maps have some drawbacks. First, most of them are time based, ranking based or matrix maps which consider only one aspect between structured items or unstructured items of each patent document. More integrated and balanced visualization approach is required to provide the overall structure of patent information effectively. Second, they are complicated networks of patent documents though they use different methods. As a result, they make patent analysis results incomprehensible and unclear to analyzer. Consequently, those deficient patent maps fail to provide an intuitive insight into the concerned technology field. And this is the third drawback we’d like to make mention of. Especially for an emerging technology like ubiquitous computing and bio informatics, it is essential to recognize its advance and make an estimate of the hereafter. Hence, we propose an alternative visualization method for patent analysis to overcome drawbacks of current patent maps. Our approach contributes to establishing a patent map which considers both structured and unstructured items of a patent document. We expect to keep the balance of analysis features by using filing dates as structured items and keywords and its frequency as unstructured items. Besides, differently from previous visualization methods for patent analysis, ours is based on forming a semantic network of keywords from patent documents. And thereby it visualizes a clear overview of patent information in a more comprehensible way. And as a result of those contributions, it enables us to understand advances of emerging technologies and forecast its trend in the future. The rest of the paper is structured as follows. In Section 2, we introduce related works with visualization methods for patent information and its analysis results. In Section 3, we explain an overview of our approach. In Section 4, we apply our visualization method to develop a patent map of the ubiquitous computing technology as an emerging technology field. And in Section 5, finally we conclude the paper with a discussion of the proposed patent map’s implications in the ubiquitous computing technology. 2. Literature review Related to researches using patent information, there are two mainstreams. One of them is to study visualization methods for patent information and its analysis results. And in this paper, we are interested in the former research area. This research area has attracted attention of the persons concerned. That’s because current technological development necessitates it to avoid unnecessary investment as well as gaining the seeds for technological develop-
1805
ment and the applicable fields. Also, the attention is increased to promote the efficient use of patent information by deepening related institutions’ understanding of patent information. On the basis of these awareness, the Japan Patent Office has been producing and providing more than 50 types of expressions and more than 200 maps for several technology fields since 1997 (JIII, 2000). In addition, many other countries such as Korea (Ryoo & Kim, 2005), Italy (Camus & Brancaleon, 2003; Fattori, Pedrazzi, & Turra, 2003) and the USA (Morris, DeYong, Wu, Salman, & Yemenu, 2002) also provide many kinds of patent maps. Currently, most patent analyses use patent citations to represent the meaningful relationship in patent information. But it is known that patent citation analysis has some serious drawbacks. According to Yoon and Park (2004), there are four drawbacks of patent citation analysis described. First, it is difficult to grasp the overall relationship among patent documents. Second, related to the first problem, the scope of analysis and the richness of potential information are limited. Third, citation has no capability of considering internal relationship between patent documents. Finally, citation analysis is a time-consuming task because it needs only an exhaustive search. So, Yoon and Park (2004) proposed a network-based patent analysis as an alternative method. But the network patent analysis still has some limitations. Researches on the intelligent methods for patent analysis have been made as well. The neural methods for mapping scientific and technical information (articles, patents) and for assisting a user in carrying out the complex process of analyzing large quantities of such information are concerned by Lamirel, Shehabi, Hoffmann, and Francois (2002). Based on text mining techniques, Tseng, Wang, Juang, and Lin (2005) created a real world patent map for an important technology domain: carbon nano-tube experimentally. And the other mainstream is concerned about patent classification. By Black and Ciccolo (2004), machine learning technology is applied to text classification on United States’ patent information to automatically differentiate between patents relating the biotech industry and those unrelated. Fall, To¨rcsva´ri, Fie´vet, and Karetka (2004) reported the results of applying a variety of machine learning algorithms for training expert systems in Germanlanguage patent classification tasks. And Trappey, Hsu, Trappey, and Lin (2006) developed a platform for patent document classification and search using a back-propagation network. 3. Methodology Our visualization method steps are summed up as follows. With colleted keywords from patent documents of a target technology field, we cluster patent documents by the k-Means algorithm. With the clustering results, we form a semantic network of keywords without respect of
1806
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
3.2. Clustering patent documents with keywords Extracting keywords from patent documents related to a target technology field
Clustering patent documents with keywords using k-Means algorithm
Forming a semantic network of keywords
Forming Patent Map
Fig. 1. Framework of visualization method for patent analysis.
To cluster patent documents with merged keywords, several steps are required (see Fig. 3). First, we check existence of each keyword within texts of a patent document. So we form a keyword existence matrix with a column index of keywords (1, . . ., j, . . ., n) and a row index of patent documents (1, . . ., i, . . ., m). If j-keyword exists within texts of i-patent document, then an element of (i, j) is filled with the number of 1. But if it does not, then the element of (i, j) is filled with the number of 0. As a result, we make a keyword existence matrix of which elements are filled with 0 or 1. Next step is to cluster patent documents by k-Means algorithm using the keyword existence matrix. Here each keyword’s value between 0 or 1 plays as a feature’s value for a patent document. So the keywords’ values are used to classify patent documents into k groups (k P 2). 3.3. Forming semantic networks of keywords
filing dates. And then we build up a patent map by rearranging each keyword node of the semantic network according to its earliest filing date and frequency in patent documents. An overview of our approach is described in Fig. 1. 3.1. Collecting keywords Our approach begins by targeting a domain technology which analyzers are interested in. And then initial keywords needs to be collected from experts to search related patent documents. After searching patent documents, we collect keywords from patent documents. And then, they are merged with the initial keywords. As a result, the list of merged keywords are completed to be used for next steps (see Fig. 2).
Now with clustered patent documents, we investigate what keyword each group has. For example, let’s assume that patent documents ‘A’ and ‘B’ belong to the group 1. According to the matrix in Fig. 4, patent document ‘A’ has keywords of ‘a’ and ‘c’. And patent document ‘B’ has keywords of ‘b’ and ‘c’. Then, the group 1 consists of three keywords of ‘a’, ‘b’, and ‘c’. Like this way, we investigate keywords for each group (see Fig. 4). And using the list of keywords for each group, we make a semantic network. How to form a semantic network is described in Fig. 5. According to it, group 1 has keywords of ‘a’, ‘b’, and ‘c’. On the other hand, group 2 has keywords of ‘c’ and ‘d’. Then two groups share ‘b’, and therefore relationship between two groups can be represented by three nodes: (a, b), (c), and (d). Here the shared node is higher than the others, so arrows are drawn from (c) to (a, b)
Target a concerned field of technology
Inquire recommendable keywords from experts
Search patent documents using the recommended keywords
Extract keywords predefined by inventors from the patent documents
Merge the recommended and Patent document databases
predefined keywords
Fig. 2. Extracting keywords from patent documents.
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
Check existence of keywords occurring in a patent document
Form a keyword existence matrix
1807
k-Means algorithm
Fig. 3. Clustering patent documents using k-Means algorithm.
a
b
c
d
A
1
0
1
0
B
0
1
1
0
C
0
0
1
1
D
0
0
0
1 a, b, c
A keyword existence matrix
c, d
Group 1
Group 2
Investigation of keywords for each group
A, B
C, D
Group 1
Group 2
Clustered patent documents: (A, B), (C, D)
Fig. 4. Investigating keywords belonging to clustered groups.
a, b, c a, b, c
c, d
Group 1
Group 2
c, d
a, b
c
d
Investigation of keywords for each group
keyword Group 1
a, b, c
Group 2
c, d
Level 0
c
List of keywords for each group
a, b
d
Level 1
Formation of a semantic network using keywords
Fig. 5. Forming a semantic network of keywords.
and (d). Like this way, we make a semantic network which consists of nodes with a keyword or more than two keywords.
The semantic network described in this section is based on the previous steps such as ‘clustering patent documents with the k-Means algorithm’ and ‘investigating key words
1808
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
for clustered patent documents’. Therefore, the semantic network is dependent on the number of groups which is set temporarily by the k-Means algorithm, and there can be so many semantic networks. There are many executable programs which can perform the k-Means algorithm. Using any of them, easily we can repeat the clustering with increasing the number of groups. And for each time based on the clustering result, we repeat both steps of ‘investigation of keywords for each group’ and ‘formation of a semantic network’. As a result of n-repetitions, we get n-semantic networks. Of course, the number of groups must be more than 1. Now we have to choose one of the n-semantic networks. Usually we select one which explains the most of the relations of keywords. And this is a manual operation. But usually as the number of groups in the chosen semantic network increases, it gets better to explain the relations of keywords by the semantic network. However, too big number makes it worse to form a semantic network therefore we have to find a point of comprise. Actually, there can be a number of ways to find the proper number of groups for the semantic network. However, in this paper, we do not include it as our concern and leave it as a further work. For the semantic network obtained, we have to investigate each node’s frequency in the clustered groups of patent documents. For each node, it is defined by counting the number of keywords’ existences in the clustered groups of patent documents. For example, in Fig. 6, ‘c’ of node 1 appears in group 1 and 2. So the frequency of node 1 is 2. The frequency of node 2 is 1 because both ‘a’ and ‘b’ appear in group 1. And the node 3’s is 1 because ‘d’ belongs
a
b
c
d
A
1
0
1
0
B
0
1
1
0
C
0
0
1
1
D
0
0
0
1
node 1 (c)
node 2 (a, b)
node 3 (d)
Fig. 6. The completed semantic network of keywords.
only to group 1. Like this way, we add a frequency of each node in the semantic network of keywords. 3.4. Forming a patent map Since now, we have explained how to form a semantic network of keywords and their frequencies as unstructured items from patent documents related to the target technology. From now on, we explain how to make use of structured items in patent documents to complete a patent map on the basis of semantic networks. Let’s assume that finally we reached the semantic network as shown in Fig. 7. We have to investigate a filing date of each node in the semantic network. The filing date of each node is the earliest filing date among patent documents which have keywords of the node. For example, in Fig. 8, node 2 consists of keywords of ‘a’ and ‘b’. And if ‘a’ belongs to patent documents of ‘A’, and ‘b’ belongs to patent documents ‘B’, then the filing date which node 1 has is the earliest filing date among document ‘A’ and ‘B’. Therefore, the filing date of node 2 is ‘1997-11-27’. Similarly, the filing
c
a, b
d
Filing date A
1997-11-27
B
2000-05-14
C
2001-03-07
D
2000-08-18
node 1 (c) 1997-11-27 node 2 (a, b) 1997-11-27
node 3 (d) 2000-08-18
A semantic network with filing date information Fig. 7. Forming a semantic network with filing date information.
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
1809
c keyword Group 1
a, b, c
Group 2
c, d
a, b
d
keyword Node 1
c
node 1
Node 2
a, b
2
Node 3
d
node 2
node 3
1
1
A semantic network with frequency information
Fig. 8. Forming a semantic network with frequency information.
date of node 3 is ‘2000-08-18’. Like this way, we add a filing date of each node in the semantic network of keywords. By these two steps, the semantic network has both aspects of structured items – filing dates, and unstructured items – keywords and their frequencies in patent documents. And now we move on to the next stage for building up a patent map using the accomplished semantic network. The patent map is completed by arranging each node of the semantic network according to its filing date in an x-direction and their frequencies in a y-direction respectively. Fig. 9 shows an example of the proposed patent map.
4. Application to ubiquitous computing technology as an emerging technology The term, ubiquitous computing, was coined more than 10 years ago by Mark Weiser who, at that time, was the chief scientist at the Xerox Palo Alto Research Center. Weiser defined ubiquitous computing as the method of enhancing computer use by making many computers available throughout the physical environments, but making them effectively invisible to the user (Weiser, 1991).
frequency
node 1 (c) 2
2
1997-11-27 node 2 (a, b) 1 1997-11-27
1997-11-27
node 3 (d) 1 2000-08-18 1
A semantic network with filing date information
node 1 (c)
node 2 (a, b)
node 3 (d)
1997-11-27
2000-08-18
1997
2000
The proposed patent map
Fig. 9. Forming a patent map using a semantic network.
filing date
1810
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
In the long term, ubiquitous computing is expected to take on great economic significance (Fleisch, 2005). Numerous applications in the business environment will become possible as physical and informational world continue to merge. And thus, additional information on
objects, processes, and individuals may be gathered, exchanged and proposed in a cost efficient way (Mu¨ller & Zimmermann, 2002). Ubiquitous computing triggers such functions as increased transparency, differential pricing and disintermediation of the value chain (Clemons & Hitt, 2000). According to these circumstances, a lot of patents are being invented all around world as a result of the advent of ubiquitous computing technology. To survive as a competitive leader in the market of ubiquitous computing technology, it is important to analyze patent information related to ubiquitous computing technology. Hence, in this paper we targeted ubiquitous computing technology as an emerging technology. And we applied our visualization method for building its patent map. Afterwards, we performed steps in Section 3 to apply our method and to build up a patent map. Their results for those steps are as follows. Related to Section 3.1, first we collected keywords recommended from experts related to ubiquitous computing technology. And then we searched patent documents related to ubiquitous computing technology using those
Table 1 The list of merged keywords for patent analysis Number
Keyword
Earliest filing date
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
RFID Universal PnP Trigger HAVI HTML M VXML interchangeability Fabrication Shop floor Magnetic memory device Logistics Automatic identification PDA, mobile, handheld device Intelligent Remote Control System GPS Ubiquitous computing Sensor network Smart Identification Manufacturing Distribution Lifecycle Healthcare Blue tooth Tracking Context awareness Inventory
1987-04-07 2002-10-01 2002-08-22 2002-08-22 2003-04-02 2002-08-22 1987-08-18 2003-04-30 1987-04-07 1987-04-07 2002-05-21 2002-12-27 2002-10-02 2002-05-21 2000-06-02 2001-01-31 2001-03-15 1987-04-07 1997-08-22 2000-06-02 2002-08-22 1987-08-18 2002-08-22 1991-12-24 2002-07-29 1987-04-07
Table 2 The list of keywords in each group of clustered patent documents Group
Keyword
1 2 3 4 5
1, 1, 2, 1, 1,
2, 3, 9, 10, 11, 13, 14, 15, 16, 18, 19, 20, 21, 24, 25, 26 3, 4, 6, 9, 15, 17, 18, 20, 21, 23, 25, 26 4, 5, 8, 10, 11, 12, 13, 14, 15, 17, 19, 20, 23, 24, 25 4, 7, 9, 10, 11, 15, 16, 17, 18, 19, 20, 22, 23, 24, 25, 26 11, 12, 13, 15, 16, 19, 20, 23, 24, 26
node 1 Ubiquitous Computing Distribution
node 2
node 3
node 4
node 5
node 6
node 7
Context awareness
RFID
Inventory
Blue Tooth
PDA Mobile device
Manufacturing Tracking
node 8
node 9
Logistics Identification
HAVI Smart
node 10 Sensor Network
node 11 Automatic Identification
node 13
node 14
Trigger Lifecycle
Universal PnP GPS
node 12 Remote Control System
node 16
node 17
node 18
Fabrication
Shop floor Healthcare
HTML VXML interchangeability Magnetic memory device
Fig. 10. A semantic network of nodes containing keywords.
node 15 Intelligent
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
recommended keywords. As a result, totally 96 patent documents were searched. And then, we investigated predefined keywords from the 96 patent documents. And then, we merged recommended and predefined keywords. The final list of merged keywords for ubiquitous computing technology is as shown in Table 1. As described in Section 3.2, we checked each keyword’s existence in the searched patent documents. Based on the result, we clustered 96 patent documents with the k-Means algorithm of ClementineTM. And then, we made semantic networks with increasing the number of groups, and selected a semantic network with five groups. The list of keywords for each group is shown in Table 2. Related to Section 3.3, we completed the final semantic network of keywords using the result of clustering (see Fig. 10). And then based on the semantic network, we investigated earliest filing date and frequency of each node in the semantic network (see Table 3). According to Section 3.4, with filing date and frequency of each node, the semantic network is transformed into the patent map for the ubiquitous computing technology (see Fig. 11). From the patent map, we see technologies related to ubiquitous computing technology have progressed towards HTML ! VXML interchangeability and magnetic memory devices in 2003 since the patents related to
1811
Table 3 Filing date and frequency for each node of the semantic network in Fig. 10 Node
Filing date
Frequency
Keyword (number)
1
2000-06-02
5
2 3 4 5 6 7
2007-07-29 1987-04-07 1987-04-07 2002-08-22 2002-05-21 1991-12-24
4 4 4 4 4 4
8
1987-04-07
3
9
2001-03-15
3
10 11 12 13
2001-01-31 1987-04-07 2002-10-02 2002-08-22
3 3 3 2
14
2002-05-21
2
15 16 17
2002-12-27 2002-08-22 1987-08-18
2 1 1
18
2003-04-02
1
Ubiquitous computing (15) Distribution (20) Context awareness (25) RFID (1) Inventory (26) Blue tooth (23) PDA, mobile, handheld device (11) Manufacturing (19) Tracking (24) Logistics (9) Identification (18) HAVI (4) Smart (17) Sensor network (16) Automatic identification (10) Remote Control System (13) Trigger (3) Lifecycle (21) Universal PnP (2) GPS (14) Intelligent (12) Fabrication (6) Shop floor (7) Healthcare (22) HTML M VXML interchangeability (5) Magnetic memory device (8)
frequency
5
2000-6-2 Ubiquitous Computing Distribution 2002-5-21 PDA Mobile device
1987-4-7 Inventory
2002-8-22 Blue Tooth
1991-12-24 Manufacturing Tracking
4
1987-4-7 RFID
2002-7-29 Context awareness
1987-4-7 Automatic Identification 3
2001-1-31 Sensor Network
1987-4-7 Logistics Identification
2002-10-2 Remote Control System
2001-3-15 HAVI Smart
2002-8-22 Fabrication 2
1
2002-5-21 Universal PnP GPS 2002-8-22 Trigger Lifecycle
1987-8-18 Shop floor Healthcare
1987
…
1991
…
2000
2001
2002
Fig. 11. A patent map based on a semantic network of Fig. 10.
2002-12-27 Intelligent
2003-4-2 HTML VXML interchangeability Magnetic memory device year 2003
1812
Y.G. Kim et al. / Expert Systems with Applications 34 (2008) 1804–1812
automatic identification, inventory, RFID, and logistics appeared in 1987. 5. Conclusions and further works In this paper, we proposed a new visualization method for patent analysis to overcome drawbacks of current patent maps. Comparing to the other methods in the literature of Section 2, our approach considered both sides of structured items and the unstructured items of patent documents. Thereby it provided an integrated and balanced approach to analyze patent information. Moreover, we suggested building up a patent map based on a semantic network of keywords from patent documents with the kMeans algorithm. So it could visualize a clear overview of patent information in a more comprehensible way. Finally, using the suggested framework of a visualization method for a patent map, we completed a patent map for ubiquitous computing technology as an emerging technology. From the patent map, we can find what kinds of patents on the ubiquitous computing technology have appeared and how those patents are merged and divided as time passes. Likewise, the proposed patent map gives a complete view of emerging technology’s advance. Also it helps us to have an insight to the technology field, thereby to avoid unnecessary investments and find the seeds for the next patent. As a further work, we had like to modify our approach into more sophisticated one with up-to-date data mining techniques. Especially, to make our visualization method more concrete, we plan to investigate methods which can determine the number of clustering groups in Section 3.3. Moreover, we plan to apply our approach to other emerging technologies in addition to ubiquitous computing technologies. References Black, D., & Ciccolo, P. (2004) Machine learning for patent classification. Camus, C., & Brancaleon, R. (2003). Intellectual assets management: From patents to knowledge. World Patent Information, 25(2), 155–159.
Clemons, E. K., & Hitt, L. M. (2000). The Internet and the future of financial services: Transparency, differential pricing, disintermediation. Discussion Draft, University of Pennsylvania, The Warton School. Fall, C. J., To¨rcsva´ri, A., Fie´vet, P., & Karetka, G. (2004). Automated categorization of German-language patent documents. Expert Systems with Applications, 26(2), 269–277. Fattori, M., Pedrazzi, G., & Turra, R. (2003). Text mining applied to patent mapping: A practical business case. World Patent Information, 25(4), 335–342. Fleisch, E. (2005). Ubiquitous network societies: Their impact on the telecommunication industry. Background Paper in ITU Workshop. Ganapathy, S., Ranganathan, C., & Sankaranarayanan, B. (2004). Visualization strategies and tools for enhancing customer relationship management. Communication of ACM, 47(11), 92–99. Japan Institute of Invention and Innovation. (2000). Guide book for practical use of patent map for each technology field. Lamirel, J. C., Shehabi, S. A., Hoffmann, M., & Francois, C. (2002). Intelligent patent analysis through the use of a neural network: Experiment of multi-view point analysis with the MultiSom model. In Proceedings of the ACL-2003 workshop on patent corpus processing. Liu, S. J. (2003). A route to a strategic intelligence of industrial competitiveness. The first Asia-Pacific Conference on Patent Maps, 2–13. Morris, S., DeYong, C., Wu, Z., Salman, S., & Yemenu, D. (2002). DIVA: A visualization system for exploring documents databases for technology forecasting. Computers & Industrial Engineering, 43(4), 841–862. Mu¨ller, C. D., & Zimmermann, H. D. (2002). Beyond mobile: Research topics for upcoming technologies in the insurance industry. In Proceedings of the 36th Hawaii international conference on system sciences. Ryoo, J. H., & Kim, I. G. (2005). Workshop H – What patent analysis can tell about companies in Korea, Far East Meets West in Vienna. Trappey, A. J. C., Hsu, F. C., Trappey, C. V., & Lin, C. I. (2006). Development of a patent document classification and search platform using a back-propagation network. Expert Systems with Applications, 31(4), 755–765. Tseng, Y. H., Wang, Y. M., Juang, D. W., & Lin, C. J. (2005). Text mining for patent map analysis. In IACIS Pacific 2005 conference proceedings (pp. 1109–1116). Weiser, M. (1991). The computer of the twenty-first century. Scientific America, 94–100. Westphal, C., & Blaxton, T. (1998). Data mining solution. New York: Wiley. WIPO. (2003). Patent map with exercises (related). WIPO-MOST intermediate training course on practical intellectual property issues in business, Theme 16. Yoon, B., & Park, Y. (2004). A text-mining-based patent network: Analytic tool for high-technology trend. The Journal of High Technology Management Research, 15(1), 37–50.