Keywords Patent portfolio; Patent cooperation; Topic identification; Correlation analysis, Social network analysis (SNA)

Transcription

1 推荐引用方式 : ZHANG Xian, XU Haiyun, FANG Shu, et al. Building potential patent portfolios: An integrated approach based on topic identification and correlation analysis[j]. Chinese Journal of Library and Information Science, 2015,8(2): Building potential patent portfolios: An integrated approach based on topic identification and correlation analysis Xian ZHANG 1,2, Haiyun XU 1, Shu FANG 1, Zhengyin HU 1,2 & Shuying LI 2 1 Chengdu Library of Chinese Academy of Sciences, Chengdu , China 2 University of Chinese Academy of Sciences, Beijing , China Abstract Purpose: This paper suggests a framework to identify important patents for building potential patent portfolios based on patents owned by different assignees so as to highlight the value of individual patents in technology transfer and identify potential collaborators for patent assignees. Design/methodology/approach: The analysis framework includes the following steps: 1) Co-classification analysis based on international patent classification (IPC) codes and Derwent manual codes (DMC) to detect sub-tech fields; 2) Keyword co-occurrence analysis aiming to understand the core technology information in each patent; 3) Social network analysis used for identifying important technologies and partnerships of key assignees. A case study is conducted with 27,401 chemistry patents filed by a Chinese national research institute. Findings: The results show that this framework is effective in building potential technological patent portfolios based on patents owned by different assignees and identifying future collaborators for patent assignees. This integrated approach based on topic identification and correlation analysis that combines network-based analysis with keyword-based analysis can reveal important patent technologies and their connections and help understand detailed technological information mentioned in patents. Research limitations: In keywords analysis, only titles and abstracts of patents were used and weights of keywords in different parts of patents were not considered. Practical implications: The analysis framework may provide valuable information for decision-makers of large institutions which have many patents with broad application prospects. Originality/value: Different from previous patent portfolio studies based on the use of a combination of patent analysis indicators, this study provides insights into a method of building patent portfolios to discover the potential of individual patents in technology transfer and promote cooperation among different patent assignees. Keywords Patent portfolio; Patent cooperation; Topic identification; Correlation analysis, Social network analysis (SNA) Corresponding author: Xian Zhang ( zhangx@clas.ac.cn). This paper is based on a presentation at the conference of The 4 th Global TechMining Conference, Stadsgehoorzaal, Leiden, Netherlands. 1

2 1 Introduction Companies apply for patents to obtain the exclusive right to make, use, sell and offer to sell the patented technologies for a limited period of time. Brockhoff [1] proposed the concept of patent portfolio, which is the collection of patents owned by an individual or a company. In order to enhance global competitiveness, companies have stepped up their efforts in patent management through patent portfolio analysis. For large enterprises or organizations which have a lot of subsidiary companies, however, it is a challenging task to build potential patent portfolios based on patents owned by different assignees and promote collaboration between these firms. Current studies on patent portfolio focused primarily on the analysis related to one patent holder and few considered the ways of building potential portfolios of patents held by different assignees [2]. Based on a case study of a large research institute in China, this paper intends to suggest a framework for enterprises and organizations to build potential patent portfolios based on patents of different assignees so as to reveal the value of individual patents in technology transfer and identify potential collaborators for these assignees. 2 Related study One of the major approaches into identification of important patents is co-occurrence analysis based on structured information such as patent classification codes, assignees, or citations. Researchers tried to use the co-occurrence relationships of patent classification codes to conduct co-classification analysis. Like co-citation and co-word analysis, it is assumed that classification codes represent cognitive elements associated with topics, specialities, or fields [3] and analysis of co-occurrence of patent classification codes can measure technological distance, with patents in a given patent category being considered more similar to one another than to those in other patent categories [4, 5].However, one limitation of co-classification analysis is that the classification may be too broad to fully capture all the technologies within a category and meet the goal for a particular analysis. Patent citations refer to the count of citations of a patent in subsequent patents, and thus citations of a patent can be used to measure the relative importance of the patent [6]. A lot of studies [7-13] used citation counts to evaluate the quality and impact of the cited technologies and the relationships between cited and citing technological areas. However, patent citation approach may lead to inaccurate results as patent citations are made by the applicants and examiners who have distinct purposes and motives in citing and thus may not reflect the association between technologies [14]. With the development of data mining techniques, analyzing the unstructured data in patent documents such as titles and abstracts have become possible. Yoon & Park [15] extracted keywords through text mining and analyzed the combination of keywords on major technologies by constructing the extracted keywords into vectors. Tseng et al. [16] discussed text mining techniques such as text segmentation, term association, and topic mapping technique and their empirical study found that patent abstract and general summary and summaries from each section in a patent document are the most topic-relevant sections. Social network analysis (SNA) is applied in identification of core technologies and detection of technological association. For instance, Yoon & Park [17] measured technology distance with Euclidian distance of keywords vectors and built patent association networks. Luan [18] built the evolution network of solar energy technologies and identified key technologies with betweenness centrality. However, relatively little research has been conducted in using SNA in building patent portfolios and finding collaborators for different patent assignees. 3 Methodology 3.1 Framework The suggested analysis framework is illustrated in Fig. 1, which involves three modules. Data preparation. This module includes selection of a patent technology field and appropriate databases for data retrieval. Keywords are first extracted from patent titles and abstracts with Thomson Data 2

3 Analyzer (TDA) 1. Standardization work is then conducted for the extracted keywords, which involves using a singular form, unification of synonyms, avoidance of hyphen and abbreviations and standardization of phrases. For example, fuel battery and fuel cell are standardized into fuel battery. Network construction and visualization. We first construct a technology network by co-occurrence of IPC codes and that of Derwent manual codes, respectively, and then a keyword concurrence network based on keywords extracted from patent titles and abstracts. The network-based patent analysis gives us the overall situation of a relevant technical field, including important patents and the relationships between patents from a viewpoint of the network, while the keyword-based patent analysis enables us to understand the core technology information within patents on the basis of analysis of patent content. To identify the unique characteristics of the patent technology network and keyword network, such SNA indicators as E-I index [19] and centrality measures [20] are employed. Girvan-Newman algorithm [21] is employed for clustering and visualization. UCINet 2 (Version 6.216) is used as the software tool for the analysis and visualization of network data. Data interpretation. With the participation of technological experts, the results are interpreted so as to refine the details of possible technological portfolios and potential cooperation opportunities for patent assignees. Fig. 1 The framework of the proposed method. 3.2 Key steps of analysis process 3.2.1Construction of patent technology network It is possible that each patent is given more than one classification code, which means this patent is involved in several technology fields. We used TDA tool (version 5.1) to build a co-occurrence matrix of IPC codes and Derwent manual codes, respectively. The reason why we use both IPC codes and Derwent manual codes is that IPC is more function-oriented rather than application-oriented, which means that IPC codes can be hardly matched with the classified industry technology in reality [22]. By comparison, Derwent manual codes provided by Derwent innovation index (DII) are application-oriented. As a result, using both IPC codes and Derwent manual codes to build the technology network will give us a comprehensive view of technology fields or areas. We refer to Tseng et al.'s study [14] and the similarity between patent J and patent K could be measured by calculating the cosine angle between them following Eq. (1). 1 Thomson Data Analyzer (TDA) can be used for patent data mining and visualization. Details are available at

4 Where each patent is regarded as a vector space with the collections of n IPC codes, IPC ij denotes the occurrence frequency of the ith IPC code in patent J, and IPC ik denotes the occurrence frequency of the ith IPC code in patent K Construction of keyword co-occurrence network As noun and noun phrases in a sentence express the main information of a sentence according to linguistics, the title and abstract of a patent document contain the main information of the patented technology [22]. Thus, keywords which are nouns and noun phrases are extracted from patent documents and are used to construct the keyword co-occurrence network. Each patent document is transformed into a vector by the frequency of the keywords occurrence. Words with co-occurrence frequency over 100 times are considered as those closely connected with a specific technology and adopted as keywords. The association strength between two nodes is calculated by applying the cosine measure following Eq. (2). (1) (2) Where Term ij denotes the occurrence frequency of the ith keyword in patent J, and Term ik denotes the occurrence frequency of the ith keyword in patent K Patent portfolio analysis with SNA indicators The external internal index (E-I index) [17], which is the ratio of the density of subgroup against that of the entire network, is used to measure the internal or external relationships between different technology fields. The possible values of E-I index range from -1 to 1. The value close to 1 indicates external relationships, while a value close to -1 indicates internal relationships [17]. In addition, betweenness centrality is used to identify important technologies in the keywords co-occurrence network. 4 Empirical study 4.1 Data collection We selected Derwent Innovation Index Database (DII) as the data source. DII merges the value-added patent information from Derwent World Patents Index with the patent citation information from Derwent Patent Citation Index. As the largest patent database in the world, DII covers over 14.3 million basic inventions from 40 worldwide patent-issuing authorities, including chemistry, electrics, electronics and mechanical engineering. Data in DII can be traced back to 1963, with over 30,000 patent records added in weekly. In DII, information can be found for patents according to patent code, assignee, assignee code and inventor and patent code includes IPC, Derwent class code, and Derwent manual code. According to WIPO s IPC8-Technology Concordance Table revised in January, 2013, the concept of technology classification is divided into 5 areas, respectively, in electric engineering, apparatus, chemistry, mechanical engineering and other fields, and also can be divided into 35 fields. We retrieved data on November 20, 2013 with IPC codes 3 of the chemistry patents filed by a Chinese national research institute and its branch institutes during 1985 and For this national research institute, chemistry is the area in which it owns the most patent applications. We retrieved totally 27,401 chemistry records. As enough data sample is necessary for building patent portfolios, this paper selects patent application in chemistry as our analytical sample. 3 IPC codes are listed in Appendix I. 4

5 4.2 Results Network based patent analysis Figure 2 displays the co-classification network of IPC codes. Based on Girvan-Newman algorithm, we found 27,401 patents in chemistry are related to 11 fields. The EI index value is , which means the 11 fields have maintained relative independent relationships. Figure 2 illustrates the internal and external relationships of the subgroups. The purple line indicates the internal relations and the green line the external relations. More internal relationships are observed from Fig. 2, especially in fields of chemical engineering, pharmaceuticals and organic fine chemistry. But the fields of food chemistry, micro-structural and nano-technology, and surface technology and coating have weak associations. Fig. 2 Patent co-classification network based on IPC codes. There are 188 subclasses in Derwent manual code classification, with clear hierarchy and highly longterm stability [23]. In our case study, this large research institute s patents in chemistry cover 182 subclasses. The clustering and visualization on the patents of 182 subclasses of manual codes based on the Girvan- Newman algorithm is shown in Fig. 3, which can be classified into 6 technological fields: 1) polymer/plastics, 2) general chemistry, 3) catalyst, 4) pharmaceuticals, 5) agriculture, 6) semiconductor, circuits, fireproof materials, ceramics, cement and electrochemistry. The threshold value of the network is set at 0.05 for satisfying visualization result. 5

6 Semiconductors / Electronic Circuitry /REFRACTORIES/ CERAMICS/CEMENT/ELECTROCHEMICALCEMENT Chemistry Polymerid/Plasdocer Agriculture Catalystser Pharmadoc Fig. 3 Patent co-classification network based on Derwent manual codes. Polymer/Plasdocer is located in the center of the network. The patents in this technology field are mainly about monomer, concentration, polymerization, natural polymer, addition polymer, condensation polymer, inorganic polymer, polymer blending, aqueous dispersion, additive, property, analysing, testing, controlling, polymerisation, polymer modification and polymer processing. Polymer processing contains devices, materials and preparation method of polymer application and plastics. Polymer application which has the highest betweenness centrality is identified as the core node of the whole network Keyword-based network construction Using the method in Section 3.2.2, we constructed the patent keyword network, and a part of the network is illustrated in Fig. 4. The keywords with high betweenness degrees such as polymer application, polymerization reaction and fermentation industry play a bridge role in connecting other patents in the network. These important technologies are usually in the core positions in the network and can be considered key technologies Portfolios possibility analysis In view of core nodes in the network, we analyzed the possibilities of building patent portfolios on the 409 patents in polymer application. Figure 4 shows the co-occurrence network of keywords with frequencies of occurrence greater than 100 times in patent documents of polymer application. Our findings indicate that this large research institute s polymer application patents can be roughly summed up into six subject areas: 1) biological polymers, 2) synthetic resin, 3) conductive polymers, 4) engineering plastics, 5) fertilizers and 6) polyamide system. We observed close relationships between the six subject areas, so a comprehensive protection solution is recommended. For example, when the same technology is applied to different fields, these different applications may be considered as a patent portfolio. Second, we suggest a patent portfolio protection for diversified technologies and their applications within one of the six subject areas. For instance, in synthetic resin area, the preparation of phenolic resin contains various solvents (acetone, ethyl acetate, component solvent, etc.) and strengthened materials (polycarbafil, activated charcoal, etc.). As a result, a variety of preparation and materials of phenolic resin can be considered as a portfolio of patented package, which highlights the value of individual patents in a package form. 6

7 Fig. 4 The possible patent portfolios of polymer applications Potential cooperation analysis This paper selected top 10 assignees on polymer patent applications of this national institute and the keywords of these applications to perform a two-mode co-occurrence network as shows in Fig. 5. We detected multiple technology cooperation areas for these patent assignees. Fig. 5 Potential partners in polymer application. In Fig. 5, the blue square nodes represent the top-ranked patent assignees. The red circle nodes represent the key patent technologies. The nodes size is proportional to the value of the betweenness centrality of the nodes. The lines between these nodes indicate the relations between the assignees and the patent technologies. The thicker the line, the stronger is the linkage. As revealed in Fig. 5, A, B and C possess multiple important patents and technologies in polymer application. As a result, there is a potential cooperation opportunity for these institutes. We discuss some of the most promising cooperation opportunities in details: A, B, D and F all focus their research on synthesis and application of phenolic resin and other resins; A, B, C, G and F have filed many patents on the preparation and application of polyethylene glycol; 7

8 B, D and E have patent filings in preparation and application of chitosan; A and E both have applied patents in aluminium metal materials and preparations; B and C are both involved in the preparation and application of coating materials; A, C and F all have applied patents in polyvinyl alcohol preparation and application. 5 Discussions and conclusions The previous studies on patent portfolio were based on the use of a combination of patent analysis indicators and focused primarily on an integration of multi-dimensional indicator system for exploring the patent assets of a certain assignee. This paper suggests a framework to provide insights into building potential patent portfolios on the technology level, especially focusing on the patents owned by different assignees and identifying future collaborators among them. More specifically, the integrated approach based on topic identification and correlation analysis that combines network-based analysis with keyword-based analysis can reveal the overall situation of a relevant technological field and the content of the patent technologies at the same time. This research needs to be improved in several aspects. For instance, in order to reveal the value of individual patents for technology transfer, we need to perform analysis at a more detailed level of granularity. To this end, we may consider the weight of words extracted from patent titles and abstracts or extract keywords from patent summary section in addition to titles and abstracts for keyword analysis. In addition, more case studies are needed to be conducted in different technological fields in order to verify the effectiveness of the suggested framework. We will address these issues in our future research. Author contributions X. Zhang (zhangx@clas.ac.cn, corresponding author) was responsible for the overall research design, proposed the analysis framework, wrote the paper outline and wrote and revised the paper. H.Y. Xu (xuhaiyun@mail.las.ac.cn) designed the patent analysis framework, performed data analysis, made the figure and tables and revised the paper. S. Fang (fangsh@clas.ac.cn) proposed the research topic and revised the paper. Z.Y. Hu (huzy@clas.ac.cn) completed the details of patent analysis framework and S.Y. Li (lisy@mail.las.ac.cn) searched information for related study and edited the paper. References 1Brockhoff, K.K. Indicators of firm patent activities. In Technology Management: The New International Language, 1991: Yue, X.P. Inner firm patent portfolio scale strategy: Based on cournot model. Journal of Intelligence (in Chinese), 2012, 31(11): , Spasser, M.A. Mapping the terrain of pharmacy: Co-classification analysis of the international pharmaceutical abstracts database. Scientometrics, 1997, 39(1): Retrieved on May 14, 2015, from DOI: /BF Jaffe, A. Technological opportunities and spillovers of R&D: Evidence from firms patents, profits, and market value. American Economic Review, 1986, 76(5), Kauffman, S., Lobo, J., & Macready, W.G. Optimal search on a technology landscape. Journal of Economic Behaviour and Organization, 2000, 43: Lee, S.J., Yoon, B.G., & Park, Y.T. An approach to discovering new technology opportunities: Keywordbased patent map approach, Technovation, 2009, 29 (6/7): Trajtenberg, M.A. Penny of your quotes: Patent citations and the value of innovations. RAND Journal of Economics, 1990, 21(1): Narin, F. Patents as indicators for the evaluation of industrial research output. Scientometrics, 1995, 34 (3): Retrieved on May 14, 2015, from 9 Lanjouw, J.O., & Schankerman, M. Characteristics of patent litigation: A window on competition. RAND Journal of Economics, 2001, 32 (1): Harhoff, D., Scherer, F.M., & Vopel, K. Citations, family size, opposition and the value of patent rights. Research Policy, 2003, 32:

9 11 Harhoff, D., & Reitzig, M. Determinants of opposition against EPO patent grants - the case of biotechnology and pharmaceuticals. International Journal of Industrial Organization, 2004, 22: Haupt, R., Kloyer, M., & Lange, M. Patent indicators for the technology life cycle development. Research Policy, 2007, 36: Retrieved on May 14, 2015, from DOI: /j.respol Wang, X., Zhang, X., & Xu, S. Patent co-citation networks of Fortune 500 companies. Scientometrics, 2011, 88(2): 761:770. Retrieved on May 14, 2015, from x. DOI: /s x. 14 Kraslawski, A. Semantic analysis for identification of portfolio of R&D projects - Example of microencapsulation. The 16th European Symposium on Computer Aided Process Engineering and 9th International Symposium on Process Systems Engineering, 2006, 21: Yoon, B.G., & Park, Y.T. Development of new technology forecasting algorithm: Hybrid approach for morphology analysis and conjoint analysis of patent information, IEEE Trans. Eng. Manag, 2007, 54 (3): Retrieved on May 14, 2015, from 16 Tseng, Y.H., Lin, C.J., & Lin, Y.I. Text mining techniques for patent analysis, Information Processing & Management, 2001, 43: Retrieved on May 14, 2015, from 17 Yoon, B.G., & Park, Y.T. A text-mining- based patent network: Analytical tool for high-technology trend. The Journal of High Technology Management Research, 2004, 15(1): Retrieved on May 14, 2015, from DOI: /j.hitech Luan, C.J. Mapping the evolution of technology network and identifying key technologies in the field of solar energy technology via co-occurrence analysis. Journal of the China Society for Scientific and Technical Information (in Chinese), 2013, 32(1): Retrieved on May 14, 2015, from /j.issn Krackhardt, D., & Stern R. Informal networks and organizational crises: An experimental simulation. Social Psychology Quarterly, 1988, 51: Retrieved on May 14, 2015, from DOI: / Social network analysis. Field manual 3-24: Counterinsurgency. Retrieved on May 14, 2015, from 21 Newman, M E J, & Girvan, M. Finding and evaluating community structure in networks. Physical Review E, 2004, 69: Wang, B., Liu, S., & Ding, K., et al. Identifying technological topics and institution-topic distribution probability for patent competitive intelligence analysis: A case study in LTE technology. Scientometrics, 2014, 101: Retrieved on May 14, 2015, from DOI: /s Thomson Reuters. DWPI Classification System. Retrieved on May 14, 2015, from Appendix I: WIPO technology concordance in chemistry Area Field IPC code Chemistry Organic chemistry fine Biotechnology Pharmaceuticals A61K-008,A61Q,C07B,C07C,C07D,C07F,C07H, C07J,C40B C07G,C07K,C12M,C12N,C12P,C12Q,C12R,C12S A61K-006,A61K-009,A61K-031,A61K-033,A61K-035, A61K-036,A61K-038,A61K-039,A61K-041,A61K-045, A61K-047,A61K-048,A61K-049,A61K-050,A61K-051, A61K-101,A61K-103,A61K-125,A61K-127,A61K-129, A61K-131,A61K-133,A61K-135,A61P 9

10 Area Field IPC code Macromolecular chemistry, polymers Food chemistry Basic materials chemistry Materials, metallurgy Environmental technology C08B,C08C,C08F,C08G,C08H,C08K,C08L B81B,B81C,B82B,B82Y A01H,A21D,A23B,A23C,A23D,A23F,A23G,A23J, A23K,A23L,C12C,C12F,C12G,C12H,C12J,C13B- 010,C13B-020,C13B-030,C13B-035,C13B-040,C13B- 050,C13B-099,C13D,C13F,C13J,C13K A01N,A01P,C05B,C05C,C05D,C05F,C05G,C06B, C06C,C06D,C06F,C09B,C09C,C09D,C09F,C09G, C09H,C09J,C09K,C10B,C10C,C10F,C10G,C10H, C10J,C10K,C10L,C10M,C10N,C11B,C11C, C11D, C99Z B22C,B22D,B22F,C01B,C01C,C01D,C01F,C01G, C03C,C04B,C21B,C21C,C21D,C22B,C22C,C22F B05C,B05D,B32B,C23C,C23D,C23F,C23G,C25B, C25C,C25D,C25F,C30B Surface technology, coating Micro-structural and nanotechnology Chemical engineering B01B,B01D-001,B01D-003,B01D-005,B01D-007,B01D- 008, B01D-009, B01D-011, B01D-012, B01D-015,B01D- 017,B01D-019,B01D-021,B01D-024,B01D-025, B01D- 027,B01D-029,B01D-033,B01D-035,B01D-036, B01D- 037,B01D-039,B01D-041,B01D-043,B01D-057, B01D- 059,B01D-061,B01D-063,B01D-065,B01D-067, B01D- 069, B01D-071, B01F, B01J, B01L, B02C, B03B, B03C,B03D,B04B,B04C,B05B,B06B,B07B,B07C, B08B,C14C,D06B,D06C,D06L,F25J,F26B,H05H A62C, B01D-045, B01D-046, B01D-047, B01D-049, B01D-050, B01D-051, B01D-052, B01D-053, B09B, B09C,B65F,C02F,E01F-008,F01N,F23G,F23J,G01T 10