Identify Technology Main Paths by Adding Missing Citations Using Bibliographic Coupling and Co-citation Methods in Photovoltaics

Identify Technology Main Paths by Adding Missing Citations Using Bibliographic Coupling and Co-citation Methods in Photovoltaics Mu-Hsuan Huang 1, Dar-Zen Chen 2, Huei-Ru Dong 1 1 Department of Library and Information Science, National Taiwan University, 10617 Taipei, Taiwan 2 Dept. of Mechanical Engineering and Institute of Industrial Engineering, National Taiwan University, 10617 Taipei, Taiwan Abstract--The objective of this research is to study the technology main paths by adding missing citations through the use of bibliographic coupling (BC) and co-citation (CC). Core patents are identified and used for patent citation network (PCN) analysis. In the citation network, patents with strong links are analyzed to construct the technology main paths. Previous researches have shown that both bibliographic coupling and co-citation may help in finding missing citation patents, so this study will utilize bibliographic coupling and co-citation to add missing citations into the original citation network and construct the technology main paths in the field of photovoltaic by connecting the strong links. The time duration of main paths from the original photovoltaic patent network has been relatively short. Furthermore, the number of patents has not exceeded three. A possible reason to explain the short time duration may be the nature of patents that is focused on technological innovation. On the contrary, the photovoltaic network after adding the missing citations has contained more links and presented a more complete trail for the main paths. Moreover, the network has sustained for longer in time duration. The results indicate that utilizing BC and CC methods to supply missing citations is an effective way in constructing the technology main paths. I. INTRODUCTION This study attempts to combine patentometrics and social network analysis to construct technology main paths for patents in the field of photovoltaics to track the current development of this technology. And this study takes one step further to locate missing links from indirect citations to explore if it is possible to strengthen the structure of these main paths, and to determine if these missing links can play a more important role for portraying a more accurate picture of the current technological trends. Patentometrics is the method that conducts systematic calculation and analysis on specific issues or features to obtain meaningful relational data and graphs. It is the process that turns bibliographic information of patents into systematic information and intelligence. The analysis provides a view to the direction of technological advancement as guiding indicators for technological and economic development as well as important reference points for product development. For industrial development, patentometrics is a diverse and useful tool that may pinpoint the latest technological trends or competitive product outputs [1], detect opportunities for collaborations or acquisitions [2], or sketch out blueprints for technological strategic alliances [3]. Social network analysis combines statistical calculations and the graph theory to answer the questions and issues in sociology. Citation network analysis is the use of citations and cited documents, combined with social network analysis, to construct the citation relationship network for the patents. Sternitzke, Bartkowski and Schramm [4] take the assignee s perspective to map out the citation network on III Nitrides technologies. The authors have grouped the citation relationships into three groups. Highly similar patent technologies are grouped together, indicating the technologies within the group are highly related to the assignees, and should be more advantageous in executing technological collaborations. In this case, patentometrics has utilized citation network analysis to locate assignees with competence for collaboration or assignees with similar technological development attributes. Bibliographic coupling (BC) [5] and co-citation (CC) [6] are methods currently used for retrieving relevant documents. BC is constructed by the citing relationship, while CC is constructed by the cited relationship. Cleverdon [7], Harter [8], Swanson [9], Small [6], Braam, Moed, and van Raan [10], and Chen, Sung, and Kuan [11] employ BC or CC to discover the relevant literatures that have not been found during ordinary studies. Small and Griffith [12], Garfield [13], Persson [14], Morris, Yen, Wu, and Asnake [15], and Jarneving [16] use BC or CC clustering to explore the research fronts. Comparisons between these two methods have been performed in several research works [15, 17]. BC is immediately available upon publication of the later-issued patent from a BC pair; however, it takes time to retrieve the CC between a pair of patents. Compared with CC, BC provides more current and immediate information about patents. Photovoltaic (PV) is a process of generating electrical power by converting solar radiation into direct current electricity using semiconductors that exhibit the photovoltaic effect. In the field of renewable energy, it is the fastest growing power-generating technology in the world. Photovoltaic patents in the utility number of USPTO database has increased rapidly, showing a drastic leap from 490 patents in 1990 to 1,611 in 2009. In the process of technological development, finding the technology main paths can be valuable by providing a glimpse of the direction of photovoltaic technology development. This study attempts to find main path for patents to investigate the development history of photovoltaic and possible direction for future development with the abovementioned analytical methods. First, the study utilizes patents with highly citations and clustering coefficient to find network clusters with strong links, identify the main paths 399

from the network clusters, and construct the technology main paths in the field of photovoltaic. But often missing links occur from failure to find, information overload, non-use policy, and the currency pattern in patents [18-20]. Therefore, the study supplements these missing citations through the use of bibliographic coupling and co-citation methods. After providing the missing citations, a further comparison is made between the modified main paths with the original ones to determine if the newly formed main paths can better portray the development process of the technology. II. METHODOLOGY A. Methods and tools This study employs patentometrics combined with social network analysis to identify technology main paths by supplementing missing citations, utilizing bibliographic coupling (BC) and co-citation (CC) methods in the field of photovoltaic. Patentometrics is a method that aggregates large quantity of patents and objective quantitative numbers to monitor the performance of study objects, while social network analysis is an analytic technique used extensively in sociology. Furthermore, for this study, UCINET and NodeXL are utilized to conduct social network analysis (SNA) and network graph illustrations. B. Data collection The patents in photovoltaics for this study are retrieved from the utility patents by United States Patent and Trademark Office (USPTO) for the period between January 1, 1990 and December 31, 2009, selecting photovoltaic patents based on keyword queries and USPC classifications. Relevant keywords including crystalline Si photovoltaics, thin-film photovoltaics, concentrated photovoltaics, photovoltaic module, photovoltaic related process, and other related technologies are used to search through the title, abstract and claim categories in the USPTO database to retrieve relevant patents. For USPC classification numbers, this study refers to reports on photocoltaics by OECD (Organisation for Economic Co-operation and Development), FEEM (Fondazione Eni Enrico Mattei), WIPO (World Intellectual Property Organization) and others, records the IPC or USPC classification numbers in these reports, and converts the IPC classification numbers into USPC numbers. The USPC classification numbers for photocoltaics include 136, 250, 257, 307, 318, 323, 326, 327, 349, 359, 362, 438 and 505. A total of 13,119 patents have been obtained from the above classification numbers, and 2,631 have been obtained from keyword searches. After eliminating the 958 overlapped patents, the final number of obtained patents in photovoltaics is 14,792. Since it is difficult to analyze the large quantity of patents by SNA, only patents with citation count (citing and cited count) higher than two are selected in the analysis resulting in 562 issue patents that include cited patents 404 (71.89%) and citing patents 174 (30.96%) that construct 468 patent pairs. C. Mining Main Path In order to mine the main paths among the patent numbers, threshold must be set. Two threshold conditions are as followed: (1) The citation (citing and cited) count for the patent must exceed the average citation count; (2) The clustering coefficient for the patent must be larger than 0, indicating the patent has enough influence and has high citation rate. The clustering coefficient is used here to measure the degree every node in the network can be clustered together. In assessing the degree of clustering, it is usually wise to compare the cluster coefficient to the overall density [21]. First, the patents in the main cluster are identified through the cluster coefficient. The citation relationship plus the application dates for the patents are important elements for finding the paths inside the cluster, and consequently the main paths to track the development of the patents. Apart from the original patent main paths, there are many missing links resulted from lack of citations for more recent patents. Reasons for such missing links may be failure to find, information overload, non-use policy or the consideration of current pattern in patents [18-20]. To compensate for these missing parts, our study utilizes BC and CC methods to find the missing citations in order to observe if it is possible to find a better main paths after including indirect citations. III. RESULTS A. Original photovoltaics patent main path The average citation count for the 14,792 patents on photovoltaics is 1.01. First, this study takes 562 highly citation patents, patents with citation counts higher than 2 for analysis, resulting in 468 patent pairs. Since highly citation patents have higher number of links, these patents are considered strong linkages. Next, a threshold to limit the clustering coefficient to be higher than zero is set to identify patents of high clusters and close links. As the result, 33 patents and 48 patent pairs have been identified. Fig. 1 showed that 33 photovoltaic patents have formed 4 clusters. After grouping the patents in chronological order, the resulting four clusters labeled in colors from left to right are blue, brown, green, and red. Cluster (1) contained only 3 patents which are from the time period of 1990-1999. Its main path in patent number is: 5079426 (U of Michigan& Xerox Co.) 5262649 (U of Michigan& Xerox Co.) 5753921 (Eastman Kodak Co.& U of Michigan), without further patents after the year 2000. Cluster (2) from the period of 1990-2009, containing the highest number of patents, is the most important cluster. This cluster includes two main paths, one being patent numbers 5244509 (Canon) 6566594 (TDK Co.& Semiconductor Energy Lab Co.) 6960718 (TDK Co.& Semiconductor Energy Lab Co.) and the other being patent numbers 5348589 (Semiconductor Energy Lab Co.) 6444899 (TDK Co.& Semiconductor Energy Lab Co.) 6521823 (TDK Co.& Semiconductor 400

Energy Lab Co.). Combining the two main paths linked by patent number 6168968 reveals that the development of both links is highly influenced by the patent technologies of this patent. Cluster (3) from the time period of 1990-2004 includes patent numbers 5122685 (QuickLogic Co.), 5231312 (Atmel Co.), 5313119 (Crosspoint Solutions, Inc.), 5572148 (Altera Co.), and 5414377 (Xilinx, Inc.) which are all cited by patents 6020760 (Altera Co.& Quickturn Design Syst, Inc.) and 6285211 (Altera Co.& Quickturn Design Syst, Inc.). In addition, patent 6020760 (Altera Co.& Quickturn Design Syst, Inc.) is also cited by patent 6285211 (Altera Co.& Quickturn Design Syst, Inc.). The data shows the main path for this cluster ends at patent 6285211 (Altera Co.& Quickturn Design Syst, Inc.) and has not extended past year 2005. Apart from that, patent 6285211 (Altera Co.& Quickturn Design Syst, Inc.) has been influence by the patent technologies from all of the patents in this cluster. Cluster (4) only contains patents for the period of 1990-2004 with two main paths that are patents 5365355 (WAH-III Tech Co.) 5461501 (Hitachi, Ltd.& Hitachi Process Computer Engn, Inc.) 5990988 (Pioneer Electric Co.& Pioneer Video Co.) and patents 5365355 (WAH-III Tech Co.) 6034749 (Hitachi, Ltd.& Hitachi Process Computer Engn) 6686976 (Hitachi, Ltd.& Hitachi Process Computer Engn, Inc.), ending in 1999 and 2004 respectively. The two main paths are both based on the technologies for patent 5365355 (WAH-III Tech Co.) that splits into two paths, but neither path has been able to continue on later. For the patent network formed by 33 patents with strong linkage and clustering coefficient higher than zero, the paths have not been able to sustain beyond three. Furthermore, with the exception of the path by patent numbers 5244509 (Canon) 6566594 (TDK Co.& Semiconductor Energy Lab Co.) 6960718 (TDK Co.& Semiconductor Energy Lab Co.) that has extended beyond 20 years, the rest of the paths are all limited to less than 15 years, signifying patents are mostly innovation-based by nature and cannot sustain for a longer period of time. Another possible reason may be the implementation of a consistent set of threshold rules for all time period in selection of highly citation patents. As the result, many more recent patents in the period 2005-2009 cannot be included in these clusters. In Figure 1, there is only one patent between 2005 and 2009. If the threshold requirement for citation patents can be adjusted accordingly, better results for main paths may be obtained. Fig. 1. The original photovoltaic patent clusters and main paths 401

B. The photovoltaic patent main paths after adding missing citations This study attempts to determine the possibility of finding better main paths after supplementing missing citations. The occurrence of missing links may be partly caused by the failure to find, information overload, non-use policy, and the currency pattern in patents [18-20]. But missing links can be remedied by BC and CC methods. Bibliographic coupling (BC) and co-citation (CC) are methods currently used for retrieving relevant documents. BC is constructed by the citing relationship, while CC is constructed by the cited relationship. The 562 highly citation patents produce 932 BC pairs and 1,759 CC pairs. For this study, a 5% threshold for BC and CC has been set, resulting in 49 patent pairs with BC equal or larger than 7 and 101 patent pairs with CC equal or larger than 28. These, used as missing citations, are then added to the initial patent network. After that, an additional threshold of limiting clustering coefficient to be larger than zero is set. The result is patents with high clusters 95 patents that form 198 patent pairs. Fig. 2 showed that after adding missing citations, six clusters are formed. Patents lined in black and included patent numbers are the original patent clusters, while patents without number and are lined in grey are the patents and citations after adding the missing citations. For Cluster (1), five additional patents are included: 5313066 (E. I. Du Pont de Nemours and Co.) 5319206 (E. I. Du Pont de Nemours and Co.) 5381014 (E. I. Du Pont de Nemours and Co.) 5693947 (U of Surrey) and 5650626 (Eastman Kodak Co.). Although these patents have not affected the original cluster and the time period is only limited to 1990-1999, one new main path has occurred that includes five patents, revealing the change in technologies. Supplying missing citation has not affected Cluster (2) and Cluster (4). The patent paths remain the same, and no additional patents or links have occurred. For Cluster (3), links have increase after adding missing citations, but number of patents has not changed. The links between the patents have increased, resulting in a longer main path from 3 to 5 patents. In this cluster, addition of the missing citations has brought the links closer than before. Clusters (5) and (6) are formed after adding missing citations; therefore, they are all formed by missing citations. Cluster (5) is the cluster with the most close-knit links with main path the longest path that has extended from 1999 to 2009. The main path for Cluster (5) includes 13 patents, the highest number among the six clusters. The cluster has main path with the most complete network among all the clusters. For Fig. 2. Photovoltaics patent clusters and main paths after adding missing citations 402

Cluster (6), the links are also closely knit, but they have only extended for a short period of 2000-2009. It is a relatively newly developed cluster compared to the other clusters with seven patents for its main path that are far higher than the patents in the main paths for all four original clusters. A closer comparison of the changes in main paths after adding the missing citations shows that, for Cluster (2) that is more complete in the patent structure, the additional citations has not helped it to build up its main path. However, for Clusters (1), (3) and (4), the ones with less number of patents, the addition of missing citations have resulted in more links and longer trail for the main paths, increasing from three to five. For the clusters that are formed after adding the missing citations, the number of links becomes higher, and patents extends for a longer period of 1990 to 2009. The trail for main path is also more complete for these clusters with 13 patents. The data reveals the addition of missing citations is an effective way in the construction of the technology main path. IV. DISCUSSION AND CONCLUSIONS This study attempts to construct the technology main path in the field of photovoltaic by taking highly citation patents and clustering coefficient in order to identify network clusters with strong links and to identify main paths from these clusters. However, missing links often occur for patents due to reasons such as failure to find, information overload, non-use policy, and the currency pattern in patents [18-20]. To compensate for the missing links, this study further attempts to supply these missing citations through the use of bibliographic coupling (BC) and co-citation (CC) to identify a more complete main paths. Lastly, the study seeks to compare the differences in main paths to understand if addition of these missing links may indeed show a clear picture of the technology development process. The initial photovoltaic network includes four clusters. The main paths for all these clusters have not sustained for a very long time period. With the only exception of the main path of patent numbers 5244509 (Canon) 6566594 (TDK Co.& Semiconductor Energy Lab Co.) 6960718 (TDK Co.& Semiconductor Energy Lab Co.) that has extended for 20 years, the other main paths have not exceeded 15 years. Furthermore, all these main paths contain only three or less patents. Part of the reasons may be patents are mostly based on innovative technologies that cannot develop continuously by nature. Other possible reason may be the similar standards used in the selection of highly citation patents for all time periods, and citations between 2005 and 2009 are still too low to be included. In the future, if threshold values for citation counts can be adjusted for different time periods, better results in finding the main paths may be identified. After adding the missing citations, the new photovoltaic patent network includes six clusters. Among these, four are original clusters and two are newly formed ones. For the four original clusters, these additional citations have not affected the largest one; however, for the smallest cluster, more patents and links as well as longer trail for the main path have resulted from adding the citations. For the remained two mid-sized original clusters, the added citations have increased the links and the trails for the main paths but have not increased the number of patents. What's more, these newly added citations have not extended the time periods of the main paths. Contrary to the four original clusters, the two newly formed clusters after adding missing citations contain more links for longer period of time. More importantly, the trails for the main paths are more complete, including 13 patents for the time period between 1990 and 2009. The results suggest that implementation of BC and CC methods to supply missing citations may be successful in constructing the technology main paths. The abovementioned comparisons indicate that the addition of missing citations may not necessarily extend the time period for main paths. Therefore, in the selection of highly citation patents, adjustment in the threshold for number of citation count for each time period may lead to identifying a more complete main paths. Through the discovery of the technology main paths, it may then be possible to identify the technology trail in the field of photovoltaic in order to better plan for the development of the photovoltaic technology. REFERENCES [1] A. F. Breitzman and M. E. Mogee, The many applications of patent analysis, Journal of Information Science, vol. 28, no. 3, pp. 187-205, Jun. 2002. [2] Breitzman, P. Thomas, and M. Cheney, Technological powerhouse or diluted competence: Techniques for assessing mergers via patent analysis, R and D Management, vol. 32, no. 1, pp. 1-10, 2002. [3] P. Trott, Innovation management and new product development, 3rd ed. Pearson Education, 2005. [4] Sternitzke, A. Bartkowski, and R. Schramm, Visualizing patent statistics by means of social network analysis tools, World Patent Information, vol. 30, no. 2, pp. 115-131, Jun. 2008. [5] M. M. Kessler, Bibliographic coupling between scientific papers, American Documentation, vol. 14, no. 1, pp. 10-25, 1963. [6] H. Small, Co-citation in the scientific literature: A new measure of the relationship between two documents, Journal of the American Society for Information Science, vol. 24, no. 4, pp. 265-269, 1973. [7] Cleverdon, The Cranfield Tests On Index Language Devices, Aslib Proceedings, vol. 19, no. 6, pp. 173-194, 1967. [8] S. P. Harter, The Cranfield II Relevance Assessments: A Critical Evaluation, The Library Quarterly, vol. 41, no. 3, pp. 229-243, Jul. 1971. [9] Don R. Swanson, Some Unexplained Aspects of the Cranfield Tests of Indexing Performance Factors, The Library Quarterly, vol. 41, no. 3, pp. 223-228, Jul. 1971. [10] R. R. Braam, H. F. Moed, and A. F. J. van Raan, Mapping of science by combined co-citation and word analysis. I. Structural aspects, Journal of the American Society for Information Science, vol. 42, no. 4, pp. 233-251, 1991. [11] Z. Chen, Y. S. Sung, and C. H. Kuan, Identifying Core Patents by Citations, Bibliographic Coupling and Co-citation, in Eleventh International Conference on Science and Technology Indicators, pp. 69-70. [12] H. Small and B. C. Griffith, The Structure of Scientific Literatures I: Identifying and Graphing Specialties, Science Studies, vol. 4, no. 1, pp. 403

17-40, Jan. 1974. [13] Garfield, Research Fronts., Current Contents, vol. 41, pp. 3-7, 1994. [14] Persson, The intellectual base and research fronts of JASIS 1986-1990, Journal of the American Society for Information Science, vol. 45, no. 1, pp. 31-38, 1994. [15] S. A. Morris, G. Yen, Z. Wu, and B. Asnake, Time line visualization of research fronts, Journal of the American Society for Information Science and Technology, vol. 54, no. 5, pp. 413-422, 2003. [16] Jarneving, Bibliographic coupling and its application to research-front and other core documents, Journal of Informetrics, vol. 1, no. 4, pp. 287-307, 2007. [17] P. van den Besselaar and G. Heimeriks, Mapping research topics using word-reference co-occurrences: A method and an exploratory case study, Scientometrics, vol. 68, no. 3, pp. 377-393, 2006. [18] P. Wilson, The value of currency, Journal of the American Society for Information Science, vol. 41, no. 4, pp. 632-642, 1993. [19] P. Wilson, Unused relevant information in research and development, Journal of the American Society for Information Science, vol. 46, no. 1, pp. 45-51, 1995. [20] H. Hsieh, Constructing an enhanced patent citation network by adding unused relevant patents, M.S. thesis, Graduate Institute of Industrial Engineering College of Engineering, National Taiwan University, Taiwan, 2009. [21] P.W. Holland and S. Leinhardt, Transitivity in Structural Models of Small Groups, Small Group Research, vol. 2, 1971, pp. 107-124. 404