Discovering competitive intelligence by mining changes in patent trends
Meng-Jung Shih, Duen-Ren Liu
*, Ming-Li Hsu
Institute of Information Management, National Chiao Tung University, Hsinchu, Taiwan
a r t i c l e
i n f o
Keywords: Change mining Patent analysis Competitive intelligence Knowledge discoverya b s t r a c t
Obtaining sufficient competitive intelligence is a critical factor in helping business managers gain and maintain competitive advantages. Patent data is an important source of competitive intelligence that enterprises can use to gain a strategic advantage. Under existing approaches, to detect changes in patent trends, business managers must rely on patent analysts to compare two patent analysis charts of differ-ent time periods. The discovery of change of trends currdiffer-ently still needs laborious human efforts and no efficient computer-based approaches are available for helping this task. In this paper, we propose a pat-ent trend change mining (PTCM) approach that can idpat-entify changes in patpat-ent trends without the need for specialist knowledge. The proposed approach consists of steps including patent collection, patent indica-tor calculation, and change detection. In change detection phase the approach firstly excavate rules between two different time periods, comparing them to determine the trend changes. These trend changes are then classified into four categories of change, evaluated with change degree and ranked by their change degree as the output information to be referred by decision makers. We apply the PTCM approach to Taiwan’s semiconductor industry to discover changes in four types of patent trends: the R&D activities of a company, the R&D activities of the industry, company activities in the industry and industry activities generally. The proposed approach generates competitive intelligence to help managers develop appropriate business strategies.
Ó 2009 Elsevier Ltd. All rights reserved.
1. Introduction
For a competitive organization, competence management is critical to organization development and even to survival issue. Complete competence management generally consists by pro-cesses including competence identification, assessment, acquisi-tion and knowledge usage (Berio & Harzallah, 2007). But before the four processes of competence management, a more perspective issue is to determine which competence to obtain. To accomplish this task, competitive organizations need to keep tracing the trends of competence change and find potential elements which may sub-stantially improve the organization competitiveness. Unfortu-nately, most competences—especially competitive intelligence, are neither structured nor quantifiable. So how to effectively dis-cover the trends of change among these abundant unstructured valuable data like intelligent properties, or more precisely say pat-ents, will be very essential to an organization to ‘‘lock on” the tar-get competences to obtain. For instance of patent data, they embody technological novelty and serve as important sources of competitive intelligence with which enterprises gain strategic
advantages (Stembridge & Corish, 2004). Patents directly represent the competitive intelligence of an industry. Any variation on patent trends in an industry as a whole will directly influence the research and development strategies of all involved enterprises. It emerges when a novel technique developed or when a revolutionary prod-uct (or parts) are invented. To maintain a leading position in the highly competitive business environment, enterprise managers need comprehend key intelligence properties of their own organi-zation, of their competitors, and of the environment in which they operate. By analyzing patent data, managers can evaluate and understand trends in the development of technologies and plan suitable strategies (Stembridge, 2005).
There has been a great deal of researches on patent data analy-sis, and several applications, such as patent map, patent citation analysis, and patent indicators, have been developed (Breitzman & Mogee, 2002; Brockoff, 1991; Chang, 2005; CHI-Research; Dou, Leveillé, Manullang, & Dou, 2005; Dürsteler, 2007; Kim, Suh, & Park, 2008; Reitzig, 2004; Yang, Akers, Klose, & Yang, 2008). Most of these studies and tools use statistical methods to analyze patent data in a specific period, and represent patent trends by visualiza-tion graphs and tables. However, these tools fail to express changes in patent trends over two time periods. A patent map visualization method proposed byKim et al. (2008) overcomes drawbacks of conventional patent maps; it enables user to understand the 0957-4174/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2009.09.001
*Corresponding author. Tel.: +886 3 5131245; fax: +886 3 5723792. E-mail address:dliu@iim.nctu.edu.tw(D.-R. Liu).
Contents lists available atScienceDirect
Expert Systems with Applications
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / e s w aprogresses of technologies, but it cannot provide a clear insight into the changes in patent trends for different periods. In real sce-nario, experts still have to identify changes in patent trends by comparing charts/tables for different periods. This task is laborious and there still has no corresponding automatic tools to help accomplish this work.
Changes in patent trends represent movements in the direction of technology development. For example, suppose a company X has its patents mainly on field A in 2003 and 2004. If the company’s main field of patents in 2005 and 2006 has became field B, we can say the technology development direction of company X has chan-ged from A to B. To capture changes in patent trends in different periods, this study proposes an approach which identifies patent trend changes with absence of specialist knowledge. These changes are ranked with change degree which is introduced in this paper. We combine association rule change mining (Song, Kim, & Kim, 2001) with patent indicators (Brockoff, 1991; CHI-Research) to de-velop a technique called patent trend change mining (PTCM), which transforms patent documents into a rule format and then identifies the most frequent rules. The frequent rules represent a patent trend in a specific period and thus, we can observe changes in patent trends by comparing the frequent rules of two time peri-ods. The patent trends of four different business levels are dis-cussed in this study: one in enterprise scope and three in industrial scope. We analyze each level of changes revealed by the proposed method, and these changes are classified, evaluated and ranked as the output.
The remainder of this paper is organized as follows. In the next section, we review literature relevant to this research, including association rule mining, change mining, patent analysis, and patent indictors. Section 3 provides an overview of our patent trend change mining (PTCM) technique. In Section4, we describe the methods for mining changes in patent trends in detail. In Section 5, we investigate changes in patent trends in Taiwan’s semiconduc-tor industry. Then, in Section6, we present our conclusions and directions for future research.
2. Background and related work
We begin this section by reviewing the definition of association rule mining used to discover trends in patent documents, and then present an overview of state-of-the-art change mining techniques. The third subsection contains an introduction to patent analysis. Then, in the fourth subsection, we discuss commonly used patent indicators.
2.1. Association rule mining
Data mining techniques have been widely used in various fields of information science (Chang, Lin, & Wang, 2009; Chen & Liu, 2004; Kuo, Lin, & Shih, 2007; Ngai, Xiu, & Chau, 2009; Yen & Lee, 2006). Association rule mining is a data mining technique used in various applications, such as market basket analysis. The tech-nique searches for interesting associations or relationships among items in a large data set (Han & Kamber, 2001). Different associa-tion rules express different regularities that exist in a dataset; and two measures, support and confidence, are used to determine whether a mined rule is a regular pattern (Han & Kamber, 2001; Ian & Eibe, 2000). The support measure determines the probability that a transaction contains both the conditional and consequent parts of a rule, while the confidence measure is the conditional probability that a transaction containing the conditional part of a rule also contains the consequent part. The apriori algorithm (Agrawal & Skrikant, 1994) is typically used to find association rules by discovering frequent itemsets (sets of items), which are
considered to be frequent if their support exceeds a user-specified minimum support threshold. Association rules that meet a user-specified minimum confidence can then be generated from the fre-quent itemsets.
In this work, we apply association rule mining to patent data to find patent patterns (rule patterns).
2.2. Change mining
The objective of change mining is to discover changes in two datasets (e.g., about customer behavior) belonging to different time periods. Change mining approaches can be classified as follows:
(a) Decision Tree Models: this method constructs decision trees for two datasets, and then identifies the differences by com-paring the two decision trees (Liu & Hsu, 1996; Liu, Hsu, Han, & Xia, 2000).
(b) Association Rules: this method determines changes by com-paring the association rules mined from two datasets (Song et al., 2001; Chen, Chiu & Chang, 2005; Liu, Hsu, & Ma, 2001). Users can decide the type of rule changes according to the similarities and differences between the rules in the data-sets. There are several types of change mining patterns (Song et al., 2001; Chen, Chiu & Chang, 2005):
Emerging patterns: The concept of emerging patterns cap-tures significant changes between datasets. An emerging pattern is a rule pattern whose support increases signifi-cantly from one dataset to another.
Unexpected consequent changes: These changes are found in newly discovered association rules whose consequent parts differ from those of the previous rule patterns. Unexpected condition changes: These changes are found in
a newly discovered association rules whose conditional parts differ from those of previous rule patterns. Added rules: These are new rules that only exist in the
present dataset.
Perished rules: These are rules that only exist in the previ-ous dataset.
Association rule change mining techniques are used to analyze transaction data and discover changes in customer behaviour. In this work, we identify changes in patent trends from patent data. 2.3. Patent analysis
Rapid technological development has made it easier for compa-nies to search and access patent documents. Many patent offices already allow free download of the abstracts and complete texts of their patents [e.g., WIPO (WIPO, 2007), USPTO (USPTO, 2007) and EPO (EPO, 2007)].
Several software tools and services have been developed in the patent field (Breitzman & Mogee, 2002; Dou et al., 2005; Dürsteler, 2007; Huang, Ke, & Yang, 2008). These tools analyze patents by classification, clustering, and statistical methods to find the rela-tionships between patents with similar content/structure. The re-sults of patent analysis are usually presented as graphs or tables, and provided to specialists, researchers, and R&D practitioners to help them plan their strategies.
Patent information can be analyzed either quantitatively or qualitatively (Huang et al., 2003). Quantitative measures are based on statistical processing, and indicate the level of patenting activity of an analytical unit (e.g., the number of patents owned by an as-signee). Qualitative measures are calculated according to citation information and used to assess the quality of a patent.
In the literature, and in practice, several indicators are used to measure patents quantitatively or qualitatively. In the next subsec-tion, we introduce patent indicators.
Although existing patent analysis tools can provide various re-sults, analysts still need to compare the results of two periods to identify changes over time. For example,Fig. 1shows the distribu-tion of the technological fields of paper-making machinery in two periods, 1984–1989 and 1990–1995 (Breitzman & Mogee, 2002). Patent analysts can discover changes in the technological field by comparing the two distributions. In this case, R&D activities in-creased for Hard Rollers and Controls, dein-creased for Bearings, and remained stable for other areas (Breitzman & Mogee, 2002). Making such comparisons requires professional knowledge. More-over, changes cannot be ranked intuitively; the degree of change must be calculated and ranked by analysts.
The motivation of this study is to discover changes in the patent trends of different time periods without the need for expert knowl-edge, and report changes to business managers by ranking the de-gree of change.
2.4. Patent indicators
Patents are one of the major sources of technological and com-petitive information because such data are easy to access and the content is highly innovative. Since the value of patents is rarely ob-servable, scholars and research organizations have defined a num-ber of patent indicators to determine the value of patents (Brockhoff,1991; CHI-Research; Reitzig, 2004; Tuomo, Hermans, & Kulvik, 2007).
The common patent indicators are described below (Brockhoff, 1991; CHI-Research; Reitzig, 2004; Tuomo, Hermans, & Kulvik, 2007):
Patent age: The age of a patent (the patent’s age is calculated from the date the patent was applied for).
Citation made (backward citations): The number of patents cited by the target patent.
Citation index (forward citations): The number of citations received by the target patent. It is a measure of the impact of the target patent.
Originality: The originality of a target patent indicates the diver-sity of cited patents, i.e., the patents cited by the target patent. The measure is based on the distribution (ratio) of cited patents over classes, as expressed in Eq.(1).
Originality ¼ 1 X
j2SB
B2j
Bj¼
Number of cited patents belonging to Class j Number of cited patents
SB: the set of classes of cited patents
ð1Þ
Generality: The generality of a target patent indicates the diver-sity of citing patents, i.e., the patents that cite the target patent. The measure is based on the distribution (ratio) of citing patents over classes, as expressed in Eq.(2).
Generality ¼ 1 X
j2SF
F2j
Fj¼
Number of citing patents belonging to Class j Number of citing patents
SF: the set of classes of citing patents
ð2Þ
Technology Cycle Time (TCT): The TCT of a target patent is the median age of the patents cited by the target patent. It is a mea-sure of technological progress.
3. Methods
The proposed patent trend change mining (PTCM) approach comprises four components, as shown inFig. 2: a patent fetcher, a patent transformer, a patent indicator calculator, and a change detection module. The first three components are described in this section, and we have more detail discussion on change detection process in Section4.
3.1. Patent fetcher
With the rapid growth of computer and internet technologies, patent documents can now be accessed freely via the Internet. The patent fetcher module uses a keyword search strategy (e.g., As-signee and International Patent Classification Code, IPC) to retrieve patents for analysis. Patent fetcher acquires patent documents (in HTML format) from the patent website and stores them into the patent document pool.
3.2. Patent transformer
Initially, a patent document is in a semi-structured HTML for-mat. This module transforms the raw patent document from semi-structured HTML format into a text format, stores it in the database, filters out irrelevant content, and extracts required pat-ent contpat-ent, including the patpat-ent number, International Classifica-tion (IPC), ApplicaClassifica-tion Date, Assignee Name, and Assignee Country. The extracted content is stored in the database for further process-ing to compute patent indicators.
3.3. Patent indictor calculator
This module calculates the patent indicators for each patent to determine the patent’s value. In this study we use four patent indi-cators, which are defined in Section2.4, to analyze patent docu-ments: Citation Index (CI) of a patent reflects the technological significance of a patent—the higher the value of a patent’s CI, the greater the patent’s impact. Originality measures the innovation of a patent—the higher the value of a patent’s originality, the great-er the patent’s innovation value. Gengreat-erality measures the scope of cross-field applications on which a patent is applied—the higher the value of a patent’s generality, the greater the patent’s economic
value. A patent is interpreted as having more ‘‘generality” if the forward citations are spread over several technological fields. Tech-nology Cycle Time (TCT) measures the time between the previous patent and the target patent, which makes improvement on the previous one—shorter TCT means a faster technological progress of patents.
The values of patent indicators are discretized for further patent trend mining. We perform data discretization based on the normalized results derived by SPSS Visual Bander. The values of patent indicators are transformed into linguistic terms as shown inTable 1.
4. Change detection in patent trends
Patents indicate the technological competitiveness as well as the innovation strategy of a company in a given period. Business managers can observe changes in patent trends by comparing the trends of two periods. The process of detecting changes in patent trends is illustrated inFig. 3.
4.1. Patent trend mining
Before describing the patent trend mining module, we intro-duce the patent trends analyzed in this study. We define four kinds of patent trends and classify them into two levels for analysis: company-level and industry-level trends.
Patent Fetcher Patent Document
Pool
Patent Indicator Calculator Patent Transformer
Patent Trend Mining
Patent Trend Comparison Patent Database Patent Trend Set (Association RuleSet) Patent Trend Reports Change Detection Managers USPTO
Fig. 2. An overview of the PTCM technique.
Table 1
Data discretization of patent indicators.
Patent indictor Linguistic term Numerical range
CI Low 60 Mid 1–4 High P5 Originality Low 0–0.39 Mid 0.40–0.65 High 0.66–1 Generality Low 0–0.44 Mid 0.45–0.65 High 0.66–1 TCT Short 0–5 Mid 6–7 Long P8 Patent Database in time period ti Patent Database in time period tj
Patent Trend Mining
Patent Trend Set (Association RuleSet)
in time period ti
Patent Trend Set (Association RuleSet)
in time period tj
Patent Trend Comparison
Evaluate the Degree of Change Patent Trend
Change Set
Patent Trend Report
(a) Company-level patent trends: These trends provide informa-tion about a company’s technological development. Trends in the R&D activities of a company: Changes in the R&D
activities of a company can be determined by comparing the relations between technological fields (IPC) and four patent indicators (the citation index, originality, generality and technology cycle time described in Section3.3) over two time periods.
(b) Industry-level patent trends: These trends provide informa-tion about the technological development of an industry. Trends in the R&D activities of an industry: Changes in the R&D
activities of an industry can be determined by comparing the relations between the technological fields (IPC) and four pat-ent indicators over two time periods.
Trends in the technological competitiveness of companies: We identify changes in technology competitiveness of companies by comparing the relations between a patent’s assignee (company) and the four patent indicators over two time peri-ods; the patent indicators reflect the technological competi-tiveness of a company.
Trends in the technological competitiveness of companies in a specific technological field: These changes can be observed by comparing the relations between both a patent’s assignee and technological fields (IPC) and four patent indicators over two time periods.
Table 2shows the four kinds of patent trends and their respective rule formats.
We apply association rule mining to patent data to identify patent trends (frequent association rule patterns). The mined fre-quent patterns can be regarded as trends extracted from patent documents. For example, if there are sufficient patents belonging to technological field B, whose assignee is X, and the CI value of those patents is high, the frequent association rule pattern ‘‘As-signee = X, IPC = B ? CI = high” can be identified. The rules iden-tify a patent trend in which the citation index of X’s patents in technological field B is relatively high. This information suggests that the quality of X’s patents in technological field B is high in the industry. Moreover, we may say that X is a pioneer company in technological field B.
4.2. Patent trend comparison
After the patent trends of different time periods have been dis-covered, the trends (in rule format) are compared to identify changes. We start with defining the types of change as follows and then discuss the process of trend comparison.
4.2.1. Types of change
Based on previous research (Song et al., 2001), four types of change in patent trends are defined:
(1) Emerging patent trends: an emerging patent trend is a rule pattern whose support increases significantly from one data-set to another.
(2) Unexpected changes in patent trends: unexpected changes in patent trends can be found in newly discovered patent trends whose consequent parts of the rule patterns are dif-ferent from those of the previous patent trend.
(3) Added patent trends: An added patent trend is a new rule, i.e., a rule not found in previous rule patterns.
(4) Perished patent trends: A perished patent trend is the oppo-site of an added rule, as it is only found in previous rule patents.
4.2.2. Rule matching
We use a rule matching method to compare the patent trends of different time periods. The method computes the similarity mea-sures and difference meamea-sures of the patent trends ruleti and
ruletþkj in time t and time t + k, respectively. The modified rule matching method comprises the following four steps (Liu, Shih, Liau, & Lai 2009; Song et al., 2001).
Step 1. Calculate the similarity degree of the conditional/conse-quent parts of two rules in different time periods.
Step 2. Calculate the similarity measure Sijbetween two rules.
The measure is derived by multiplying the similarity degree of the conditional parts (Cij) of the rules by the similarity degree
of the consequent parts (Qij).
Step 3. Calculate the difference measure oijbetween two rules.
The measure is the similarity degree of the conditional parts minus the similarity degree of the consequent parts.
Step 4. Determine the type of change according to the similarity measures and difference measures.
4.2.3. Identifying the type of change
Table 3shows the measures used to determine each type of pat-ent change; the measurempat-ents are adopted from (Liu et al. 2009; Song et al., 2001). The four types of patent change can be classified according to the two judged factors, i.e., the similarity measure Sij
and the difference measure oij, and three predefined thresholds:
hem for emerging patterns, hun for unexpected changes, and ha/p
for added and perished rules. Note that hem> hun> ha/p. The process
of identifying the types of changes follows a pre-determined sequence. First, we identify emerging patterns. If the similarity Table 2
Patent trends and their respective rule formats.
Analyzed level Patent trend Rule format
Conditional part ? Consequent part
Company level R&D activities of a company IPC ? CI/Originality/Generality/TCT
Industry level R&D activities of the specified industry IPC ? CI/Originality/Generality/TCT Technological competitiveness of companies Assignee ? CI/Originality/Generality/TCT Technological competitiveness of companies in a specific technological field Assignee, IPC ? CI/Originality/Generality/TCT
Table 3
Measurement for each type of change. Type of change (rt
i;rtþkj ) Measurement Emerging pattern SijP hem(Sij= Cij Qij)
(Cij: similarity degree of the conditional parts) (Qij: similarity degree of the consequent parts) Unexpected change Maxð1i;1jÞ < hem; @ij>hunð@ij¼ Cij QijÞ Added patent trend 1j<ha=pð1j¼ max
i SijÞ Perished patent trend 1i<ha=pð1i¼ max
measure Sijis greater than or equal to hem, it means that the two
rules are similar and rule rtþk
j can be regarded as an emerging
pat-tern. If the maximum similarity measure Maxð1i;
1
jÞ is less than hemand the difference measure oijis greater than hun, we regard rule
rtþk
j as an unexpected change. Note that
1
i¼ maxj Sij;
1
j¼ maxi Sij:Finally, if
1
jis less than ha/p, rule rtþkj is identified as an addedpat-ent trend; and if
1
iis less than ha/p, rule rtiis identified as a perishedpatent trend.
4.3. Evaluating the degree of change
As a large number of changes occur in a competitive business environment, managers need to focus on the most important changes. To do this, it is necessary to evaluate the degree of change, and rank the changed rules according to their importance.
Table 4shows the simple formulations for measuring the de-gree of change. The formulations, which are adopted from (Liu et al. 2009), measure the degree of change. The notations sup-portt(r
i) and supportt+k(ri) represent the support value of riat time
t and and rjat time t + k, respectively; while
1
iand1
jare themax-imum similarity measures of rt
i and rtþkj , respectively.
After calculating the degrees of change, the most important changes are reported to business managers, who then analyze the changes in patent trends over different time periods and use the information to understand the changing business environment and plan appropriate strategies.
5. Patent change mining in Taiwan’s semiconductor industry We now apply our proposed PTCM technique to Taiwan’s semi-conductor industry.
5.1. Data collection
The dataset of semiconductor-related patents was obtained from the USPTO (United States Patent and Trademark Office) pat-ent database. We select Taiwan semiconductor-related patpat-ents available online for the period 2001–2004 based on the IPCs belonging to the semiconductor industry, as identified by the Tai-wan Intellectual Property Office (seeAppendix A). We divided this dataset, which contains 4310 unique patents, into two periods: the first part contains 2352 patent documents for the period 2001– 2002, while the second part contains 1958 patent documents for the period 2003–2004.
5.2. Changes in the R&D activities of TSMC (Taiwan Semiconductor Manufacturing Co. Ltd)
Changes in a company’s R&D activities are identified by com-paring the relations between the technological field (IPC) of the target company and the citation index, originality, generality, and technology cycle time over two time periods. We chose TSMC as the target company, and divided its patents into two parts:
2001–2002 and 2003–2004. Table 5 lists some changes in the R&D activities of TSMC between 2001 and 2004.
From patent trend(1), we observe the rapid growth (57%) of the company in terms of high originality in H01L29/788. This informa-tion shows that, during the period under study, TSMC exhibited a high degree of inventiveness in the technological field H01L29/788. Meanwhile, patent trend (3) shows that the citation index of H01L27/108 decreased between 2001 and 2004. A reduction in the CI often indicates a decline in quality, although it can mean that the patent is fairly new. The added patent trends (5) and (6) inTable 6indicate that H01L21/336 and G01R31/26 are new tech-nological fields that TSMC invested in. The number of citations of these patents is relatively low. Finally, from perished patent trends (7) and (8), we observe that the innovativeness of TSMC declined gradually in terms of H01L21/336 and H01L21/44 in the period un-der study.
5.3. R&D activities of Taiwan’s semiconductor industry
Changes in the R&D activities of an industry are identified by comparing the relations between the technological fields (IPC) of the target industry and the citation index, originality, generality and technology cycle time over two time periods. Table 6 lists Table 4
Measuring the degree of change in patent trends.
Type of change Degree of change
Emerging patent trends Supporttþkðr jÞSupporttðriÞ Supporttðr
iÞ Unexpected changes in patent trends Supporttðr
iÞSupporttþkðriÞ Supporttðr
iÞ Support tþkðr
jÞ Added patent trend ð1 1jÞ Supporttþkðr
jÞ Perished patent trend ð1 1iÞ SupporttðriÞ
Table 5
Some changes in the R&D activities of TSMC.
Patent trend Change
degree Emerging patent trends
(1) IPC = H01L29/ 788 ? Originality = High 0.57 (2) IPC = H01L21/ 00 ? TCT = Short 0.21 Unexpected changes in patent trends
2001–2002 2003–2004
(3) IPC = H01L27/108 ? CI = Mid IPC = H01L27/ 108 ? CI = Low 0.02 (4) IPC = H01L21/ 311 ? TCT = Short IPC = H01L21/ 311 ? TCT = Long 0.02 Added patent trends
(5) IPC = H01L23/62 ? CI = Low 0.03
(6) IPC = G01R31/26 ? CI = Low 0.02
Perished patent trends
(7) IPC = H01L21/336 ? CI = High 0.05
(8) IPC = H01L21/ 44 ? Generality = High
0.03
Table 6
Some changes in the R&D activities of Taiwan’s semiconductor industry.
Patent trend Change
degree Emerging patent trends
(1) IPC = H01L29/ 76 ? CI = Low
1.31
(2) IPC = H01L21/00 ? CI = Low 1.07
Unexpected changes of patent trends
2001–2002 2003–2004 (3) IPC = H01L29/ 40 ? TCT = Short IPC = H01L29/ 40 ? TCT = Mid 0.02 (4) IPC = H01L21/ 48 ? TCT = Short IPC = H01L21/ 48 ? TCT = Mid 0.01 Added patent trends
(5) IPC = H01L29/ 788 ? CI = Low 0.03 (6) IPC = G11C16/ 04 ? CI = Low 0.02
some changes in the R&D activities of Taiwan’s semiconductor industry between 2001 and 2004.
InTable 6, the emerging patent trends (1) and (2) show that companies in the industry invested in H01L29/76 and H01L21/00 consistently throughout the period under study. The high growth rates (131% and 107%, respectively) indicate that companies focused their R&D activities on the two technological fields. How-ever, the low CI indicates that the companies lacked pioneer pat-ents and basic patpat-ents in these technological fields.
Patent trends (3) and (4) inTable 6 indicate that the TCT of H01L29/40 and H01L21/48 changed from a short-cycle time to a medium-cycle time, which implies that the speed of innovation in these technological fields slowed down. The added patent trends (5) and (6) indicate that H01L29/788 and G11C16/04 were new technological fields that Taiwan’s semiconductor companies in-vested during 2003–2004.
5.4. Technological competitiveness of companies in Taiwan’s semiconductor industry
Changes in the technological competitiveness of companies in an industry are identified by comparing the relations between the assignee of the target industry and the citation index, original-ity, generaloriginal-ity, and technology cycle time over two time periods. Table 7 lists some changes in the technological competitiveness of companies in Taiwan’s semiconductor industry between 2001 and 2004.
Patent trends (1) and (2) inTable 7show the consistent innova-tive power of TSMC and MIC. Specifically, the marked increase in MIC’s patents (263%) indicates the innovativeness of MIC and the direction of its R&D activities. However, the low CI indicates that
MIC was a technological follower between 2001 and 2004. Patent trend (4) inTable 8shows a decrease in the Originality of SPIC. The added patent trends (5) and (6) in the table show several new assignees of semiconductor patents, which means that new companies (AOC and NYT) entered the semiconductor industry during 2003–2004.
From the perished patent trends (7) and (8), we observe that the high value of CI and the Generality of TSMC’s patents decreased be-tween 2003 and 2004. This implies that the quality of TSMC’s R&D may have declined during 2003–2004, although the phenomenon may be due to new patents.
5.5. Technological competitiveness of companies in specific technological fields
Changes in the technological competitiveness of companies in specific technological fields are derived by comparing the relations between both the patent’s assignee and the technological field (IPC) of the target industry with the citation index, originality, gen-erality, and technology cycle time over two time periods.Table 8 lists some changes in Taiwan’s semiconductor industry between 2001 and 2004.
The frequent appearance of TSMC in emerging patent trends shows that the company played a leading role in Taiwan’s semi-conductor industry throughout the period under study. The per-ished patent trends (3) and (4) in Table 8 show that UMC’s technological competitiveness with medium CI and low Originality in H01L21/336 declined, which may imply a change in UMC’s inno-vative activities.
6. Conclusions
In this study we proposed a patent trend change mining (PTCM) technique that captures changes in patent trends without the need for specialist knowledge and reports changes to business managers by ranking the degrees of change. Competitive intelligence of ness is derived by an automatic change mining approach that busi-ness managers can modify and develop appropriate strategies according to their findings. The proposed approach mines changes in patent trends by analyzing the metadata in patent documents. We applied the proposed PTCM to Taiwan’s semiconductor indus-try for the period 2001–2004 to discover changes in four types of patent trends: the R&D activities of a company, the R&D activities of the industry, the technological competitiveness of companies and the technological competitiveness of companies in a specific technological field. The results obtained by the proposed approach Table 7
Some changes in the technological competitiveness of companies in Taiwan’s semiconductor industry.
Patent trend Change
degree Emerging patent trends
(1) Assignee = Macronix International Co. Ltd ? CI = Low 2.63
(2) Assignee = Taiwan Semiconductor Manufacturing Co. Ltd ? Originality = High
0.01 Unexpected changes in patent trends
2001–2002 2003–2004
(3) Assignee = Advanced Semiconductor Engineering, Inc. ? CI = High Assignee = Advanced Semiconductor Engineering, Inc. ? CI = Low 0.32 (4) Assignee = Siliconware Precision Industries Co., Ltd ? Originality = High Assignee = Siliconware Precision Industries Co.,
Ltd ? Originality = Low
0.03 Added patent trends
(5) Assignee = Au Optronics Corp. ? CI = Low 0.04
(6) Assignee = Nan Ya Technology ? CI = Low 0.03
Perished patent trends
(7) Assignee = Taiwan Semiconductor Manufacturing Co. Ltd ? CI = High 0.07
(8) Assignee = Taiwan Semiconductor Manufacturing Co. Ltd ? Generality = Mid
0.07
Table 8
Some changes in the activities of Taiwan’s semiconductor industry.
Patent trend Change
degree Emerging patent trends
(1) IPC = H01L21/302, Assignee = Taiwan Semiconductor Manufacturing Co. Ltd ? CI = Low
1.4 (2) IPC = H01L21/44, Assignee = Taiwan Semiconductor
Manufacturing Co. Ltd ? CI = Low
0.78 Perished patent trends
(3) IPC = H01L21/336, Assignee = United Microelectronics Corp. ? CI = Mid
0.02 (4) IPC = H01L21/336, Assignee = United Microelectronics
Corp. ? Originality = Low
can be used as an important reference for decision makers to make more accurate strategies on research and development.
There remain several extended researches to do based on this study. The primary part of most patent document is textual con-tent which contains rich information to utilize (e.g., abstracts and claims). Through analyzing the textual part we can surely improve the quality of change detection and provide more comprehensive results. Therefore the next research will be a patent trend change mining approach which utilizes text mining techniques.
Acknowledgement
This research was supported in part by the National Science Council of the Taiwan under the grant NSC 96-2416-H-009-007-MY3.
Appendix A
IPCs belonging to the semiconductor industry identified by the Taiwan Intellectual Property Office.
IPC Description
C23C Coating metallic material; coating material with metallic material; surface treatment of metallic material by diffusion into the surface, by chemical conversion or
substitution; coating by vacuum evaporation, by sputtering, by ion implantation or by chemical vapor deposition
016/00 Chemical coating by decomposition of gaseous compounds, without leaving reaction products of surface material in the coating, i.e. chemical vapor deposition (CVD) processes
G01R Measuring electric variables; measuring magnetic variables
031/02 General constructional details
G03F Photomechanical production of textured or patterned surfaces, e.g., for printing, for processing of semiconductor devices 007/00 Photomechanical, e.g., photolithographic,
production of textured or patterned surfaces, e.g., printed surfaces
009/00 Registration or positioning of originals, masks, frames, photographic sheets, or textured or patterned surfaces
G05F Systems for regulating electric or magnetic variables
001/10 Regulating voltage or current G11C Static stores
007/00 Arrangements for writing information into, or reading information from, a digital store 016/04 Using variable threshold transistors, e.g.,
FAMOS
H01L Semiconductor devices; electronic solid state devices
021/00 Processes or apparatus specially adapted for the manufacture or treatment of
semiconductor or solid state devices or parts thereof
023/34 Arrangements for cooling, heating, ventilating or temperature compensation
Appendix A (continued)
IPC Description
023/48 Arrangements for conducting electric current to or from the solid state body in operation, e.g., leads, terminal arrangements
023/495 Lead-frames
023/52 Arrangements for conducting electric current within the device in operation from one component to another
023/58 Structural electrical arrangements for semiconductor devices
023/62 Protection against over-current or overload, e.g., fuses
027/108 Dynamic random access memory structures 029/00 Semiconductor devices specially adapted for
rectifying, amplifying, oscillating or switching and having at least one potential-jump barrier or surface barrier; capacitors or resistors with at least one potential-jump barrier or surface barrier, e.g. PN-junction depletion layer or carrier concentration layer; details of semiconductor bodies 029/40 Electrodes
029/76 Unipolar devices 029/788 With floating gate
029/94 Metal–insulator–semiconductors, e.g., MOS 031/062 The potential barriers being only of the
metal–insulator–semiconductor type 031/113 Being of the conductor–insulator–
semiconductor type, e.g., metal–insulator– semiconductor field-effect transistor 031/119 Characterized by field-effect operation, e.g.,
MIS type detectors
References
Agrawal, R., & Srikant, R. (1994). Fast algorithm for mining association rules. In Proc. of the int. conf. on very large data bases, Santiago de Chile (pp. 487–499). San Francisco, CA: Morgan Kaufmann.
Berio, G., & Harzallah, M. (2007). Towards an integrating architecture for competence management. Computers in Industry, 58(2), 199–209.
Breitzman, A. F., & Mogee, M. E. (2002). The many applications of patent analysis. Journal of Information Science, 28(3), 187–205.
Brockhoff, K. K. (1991). Indicators of firm patent activities. In Technology management: the new international language (pp. 476–481).
Chang, M. C. (2005). Quantum computation patent mapping – A strategic view for the information technique of tomorrow. In Proc. of the Int. conf. on service systems and service management (pp. 1177–1181).
Chang, C.-W., Lin, C.-T., & Wang, L.-O. (2009). Mining the text information to optimizing the customer relationship management. Expert Systems with Applications, 36(2), 1433–1443.
Chen, M. C., Chiu, A. L., & Chang, H. H. (2005). Mining changes in customer behavior in retail marketing. Expert Systems with Applications, 28(4), 773–781. Chen, S. Y., & Liu, X. (2004). The contribution of data mining to information science.
Journal of Information Science, 30(6), 550–558. CHI-Research. Available fromhttp://www.chiresearch.com.
Dou, H., Leveillé, V., Manullang, S., & Dou, J. J. (2005). Patent analysis for competitive technical intelligence and innovative thinking. Data Science Journal, 4, 209–237.
Dürsteler, J. C. (2007). Patent analysis. Available from http://www.infovis.net/, visited 2007/5.
EPO (2007). European Patent Office. Available from http://www.european-patent-office.org/index.en.php.
Han, J., & Kamber, M. (2001). Mining association rules in large databases. Data mining-concepts and techniques. San Francisco, CA: Morgan Kaufmann.
Huang, Z., Chen, H., Yip, A., Ng, G., Guo, F., Chen, Z. K., et al. (2003). Longitudinal patent analysis for nanoscale science and engineering: country, institution and technology field. Journal of Nanoparticle Research, 5(3–4), 333–363.
Huang, S.-H., Ke, H.-R., & Yang, W.-P. (2008). Structure clustering for Chinese patent documents. Expert Systems with Applications, 34(4), 2290–2297.
Ian, H. W., & Eibe, F. (2000). Output: knowledge representation. Data mining. San Francisco, CA: Morgan Kaufmann.
Kim, Y. G., Suh, J. H., & Park, S. C. (2008). Visualization of patent analysis for emerging technology. Expert Systems with Applications, 34(3), 1804–1812. Kuo, R. J., Lin, S. Y., & Shih, C. W. (2007). Mining association rules through
integration of clustering analysis and ant colony system for health insurance database in Taiwan. Expert Systems with Applications, 33(3), 794– 808.
Liu, B., Hsu, W., & Ma, Y. (2001). Discovering the set of fundamental rule changes. In Proc. of the 7th ACM int. conf. on knowledge discovery and data mining (pp. 335– 340), San Francisco, CA.
Liu, B., & Hsu, W. (1996). Post-analysis of learned rules. In Proc. of 13th natl. conf. on artificial intelligence (pp. 828–834), Menlo Park, CA.
Liu, B., Hsu, W., Han, H. S., & Xia, Y. (2000). Mining changes for real-life applications. In Proc. of the 2nd int. conf. on data warehousing and know discovery (pp. 337– 346), London.
Liu, D. R., Shih, M. J., Liau, C. J., & Lai, C. H. (2009). Mining the change of event trends for decision support in environmental scanning. Expert Systems with Applications, 36(2), 972–984.
Ngai, E. W. T., Xiu, L., & Chau, D. C. K. (2009). Application of data mining techniques in customer relationship management: a literature review and classification. Expert Systems with Applications, 36(2), 2592–2602.
Reitzig, M. (2004). Improving patent valuations for management purposes – Validating new indicators by analyzing application rationales. Research Policy, 33(6–7), 939–957.
Song, H. S., Kim, J. K., & Kim, S. H. (2001). Mining the change of customer behavior in an internet shopping mall. Expert Systems with Applications, 21(3), 157–168. Stembridge, B. (2005). Sorting the wheat from the chaff- the use of patent analysis
in evaluating portfolios. Available from http://www.scientific.thomson.com/ newsletter.
Stembridge, B., & Corish, B. (2004). Patent data mining and effective patent portfolio management. Intellectual Asset Management(Oct./Nov.), 30–35.
Tuomo, N., Hermans, R., & Kulvik, M. (2007). Patent citations indicating present value of the biotechnology business. Available fromhttp://www.etla.fi/. USPTO (2007). United States Patent & Trademark Office. Available fromhttp://
www.uspto.gov/.
WIPO (2007). World Intellectual Property Organization. Available from http:// www.wipo.int/portal/index.html.en.
Yang, Y. Y., Akers, L., Klose, T., & Yang, C. B. (2008). Text mining and visualization tools – Impressions of emerging capabilities. World Patent Information, 30(4), 280–293.
Yen, S. J., & Lee, Y. S. (2006). An efficient data mining approach for discovering interesting knowledge from customer transactions. Expert Systems with Applications, 30(4), 650–657.
