Scientometrics (2012) 91:645–670

DOI 10.1007/s11192-012-0635-7

A new instrument for technology monitoring: novelty

in patents measured by semantic patent analysis

Jan M. Gerken • Martin G. Moehrle

Received: 6 June 2011 / Published online: 31 January 2012

 Akadémiai Kiadó, Budapest, Hungary 2012

Abstract Given that in terms of technology novel inventions are crucial factors for
companies; this article contributes to the identification of inventions of high novelty in
patent data. As companies are confronted with an information overflow, and having patents
reviewed by experts is a time-consuming task, we introduce a new approach to the identi-
fication of inventions of high novelty: a specific form of semantic patent analysis. Subsequent
to the introduction of the concept of novelty in patents, the classical method of semantic
patent analysis will be adapted to support novelty measurement. By means of a case study
from the automotive industry, we corroborate that semantic patent analysis is able to out-
perform available methods for the identification of inventions of high novelty. Accordingly,
semantic patent information possesses the potential to enhance technology monitoring while
reducing both costs and uncertainty in the identification of inventions of high novelty.

Keywords Novelty measurement  Semantic patent analysis  Inventive progress 

Technology monitoring  Citation analysis  Classification analysis

Mathematical Subject Classification (2000) 68U15

JEL Classification O32  O34

Introduction

Technology proves to be a central driving agent of economic success, as technological

change is able to influence the competitive environment to a considerable extent as well as
in different directions. On the one hand, technological change is the most powerful driver
of growth (Sood and Tellis 2005), offering companies the opportunity to gain advantages
in competition. On the other hand, technological change may also result in the decline of
long-established institutions (Lichtenthaler 2004).

J. M. Gerken (&)  M. G. Moehrle

Institute of Project Management and Innovation (IPMI), University of Bremen,
Wilhelm-Herbst-Str.12, 28359 Bremen, Germany
e-mail: jgerken@uni-bremen.de

123
646 J. M. Gerken, M. G. Moehrle

Even though the definitions of technological change, e.g. as introduced by Achilladelis

et al. (1990), Christensen and Overdorf (2000), Chandy and Tellis (1998) and Anderson
and Tushman (1990), differ in some respects (e.g. as concerns customer benefits), they
share the notion that technological change is enforced by inventions which are charac-
terized by a high degree of novelty. For this reason, companies are confronted with the
challenge not only to create inventions of high novelty, but also to identify competitive
inventions of high novelty at an early stage, as they may provide the basis for a prospective
technological change in the competitive environment.
At present the greatest challenge does not lie in the availability of information about
inventions of high novelty, but rather in the assessment thereof, as companies are confronted
with the problem of information overflow (Bergmann et al. 2008). Information overflow
causes the monitoring of inventions by manual efforts to be extremely expensive. In order to
meet this challenge, several articles address the identification of inventions of high novelty
by way of a computer-based patent analysis (see Table 1 with their characteristics). Major
shortcomings are (i) the level of analysis, which is not often valid on the basis of a single
patent, and (ii) the use of linguistic variables for novelty measurement, which are mostly
based on citation links and keywords and neglect functional relationships in texts.
To overcome these shortcomings we will develop a specific form of semantic patent
analysis. Our approach focuses on functional structures in patents, which may be extracted by
using semantic knowledge. In our approach we are going to focus on three research queries:
• How can semantic patent analysis be adapted to measure novelty?
• How well does semantic patent analysis perform compared with existing methods in
the identification of inventions of high novelty?
• What implications are to be expected in theory and management practice?
In Theoretical background, we shall discuss the theoretical background of novelty mea-
surement. We will adapt semantic patent analysis for novelty measurement in Semantic patent
analysis for identifying inventions of high novelty. In particular, we will discuss issues of
variable measurement and novelty calculation. Using a case study from the automotive
industry in Data and research design, we will examine in detail, whether semantic patent
analysis is able to outperform available methods in identifying inventions of high novelty. The
results, which will be discussed in Results, may serve to confirm our assumptions: semantic
patent analysis offers an opportunity to identify inventions of high novelty. However, there is
still room for improvement. Limitations of this approach and the case study will be discussed
in Discussions and conclusions, which also refers to theoretical and managerial implications.

Theoretical background

Term and dimensions of novelty

Novelty is in fact a highly complex concept. In analogy to innovations, we differentiate

between recombinant and pioneering novelty. Recombinant novelty dates back to
Schumpeter, (Schumpeter 1934) who interpreted novelty as new combinations of more or
less known resources (see also Hargadon and Sutton 1997; Fleming and Sorenson 2001;
Fleming 2001; Fleming et al. 2007; Schoenmakers and Duysters 2010). In contrast, pio-
neering novelty can be found in innovations made up of elements, which are not drawn
from the ‘made’ world (Fleming 2001). We do not consider recombinant and pioneering
novelty as a dichotomy but as poles of a continuum, as the combinations of resources used

123
 Table 1 Prior methods for the analysis of novelty based on patents
Approach(es) Novelty indicator Novelty measurement Focus of measure Shortcomings

Achilladelis et al. (1987) Patents Counting of patents in S-curves Novelty measurement only on a macro level
technological field but not on the level of a single patent
Achilladelis et al. (1990) Patents Counting of patents in S-curves Inadequate for the identification of single
technological field patents with a high degree of novelty
Andersen (1999) Patents Counting of patents in S-curves
technological field
Frietsch (2007) Patents Counting of patents in S-curves
technological field
Haupt et al. (2007) Patents Counting of patents in S-curves
technological field
A new instrument for technology monitoring

Ahuja and Lampert (2001) Patents and US patent Counting of new US patent Novel technologies
classification classes entered by a firm in the
previous three years
Trajtenberg et al. (1997) Backward citation Counting of backward citations Basicness 70% of all patents are cited less often than
to scientific literature three times, citation analysis focuses
Trajtenberg et al. (1997) Backward citation Weighting of backward citation Distance in technology space exclusively on bibliographic information
by incorporating a three-digit and in many cases ignores the description
classification scheme section of patents (Lee et al. 2009a), on
average, the time lags between citing and
Ahuja and Lampert (2001) Backward citations Number of patents without Pioneering technologies cited patents are more than ten years (Hall
backward citations et al. 2001), the references stated in a
Dahlin and Behrens Backward citation Similarity structures between Novelty of radical inventions patent are subject to strategic decisions, as
(2005) backward citations patent owners can also file citations for
von Wartburg et al. (2005) Backward citations Multi-stage backward citations Technological value added their own patents (USPTO 2007)
(multi-stage) weighted by the inverse
number of backward citations
in a patent
Schoenmakers and Backward citations Number of backward citations Novelty of radical inventions
Duysters (2010)
647

123
 Table 1 continued
648

Approach(es) Novelty indicator Novelty measurement Focus of measure Shortcomings

123
Kim et al. (2008) Keywords (experts’ Clustering based on keyword Emerging technologies Dependent on experts’ judgement of
choice) existence matrix, patent keywords, limited to single words without
mapping contextual information
Lee et al. (2009b) Keywords (computer- Distinguishing between and Evolutionary directions of
based approach with counting of emerging, technologies
expert screening) declining, core and established
keywords in consumption with
patent mapping
Lee et al. (2009a) Keywords (computer- Distinguishing between and Discovering new technology
based approach with counting of adjacent patents opportunities
expert screening) with emerging and declining
keywords in consumption with
patent mapping
Lee et al. (2011) Keywords (computer- Similarity calculation between Monitor trends of technological
based approach with keyword vectors and analysis change
expert screening) of presence and absence of
keywords
Choi et al. (2011) SAO networks Density and cohesion of SAO Novel technology functions Evaluation of novel technology functions as
sub-networks elements of patents in a technological
domain but no evaluation on the level of
patents
Yoon and Kim (2011b) SAO structures Similarity measurement based on Novel patent clusters Novel inventions are identified on the level
SAO structures, construction of of sub-domains but not on the level of
patent networks based on single patents
similarity, calculation of
density and cohesion of patent
sub-networks
J. M. Gerken, M. G. Moehrle
A new instrument for technology monitoring 649

for recombinant novelty may come from a narrow field, but also from wider fields and then
approach pioneering novelty.
In this article, we are focussing on the measurement of novelty of patents for the
assessment of technologies. As we will analyse novelty especially in patents and their
respective inventions, it is necessary to highlight three particular aspects of novelty in this
context. First of all, Witt (2009) defines novelty ‘‘as something that was unknown before
a particular point in time that, hence, was discovered or created at that time’’. If novelty is
defined as something that had hitherto been unknown, this implies that novelty is also
directly related to what was known before (see also Dahlin and Behrens 2005). In the
context of patents ‘‘what was known before’’ is usually referred to as ‘‘state of the art’’ or
‘‘prior art’’. Comprehending what the novel impulse of an invention really is, relies on a
diligent selection of prior art. Secondly, in patents novelty is combined with so-called non-
obviousness (USPTO 2007). A patent should not be granted ‘‘if the differences between the
subject matter sought to be patented and the prior art are such that the subject matter as a
whole would have been obvious at the time the invention was made to a person having
ordinary skill in the art to which said subject matter pertains’’ (§ 103). In our article we will
consider novelty and non-obviousness as a joint phenomenon. In a more advanced
approach it could be useful to model non-obviousness as a moderating effect of the
relationship between a patent’s novelty and its impact on other variables like patent value.
Thirdly, in the context of technological change the novel item or method should be useful,
as we are not interested in detecting changes in fashion or similar phenomena. As concerns
patents we shall take this aspect for granted, because the patent law recommends that an
invention be new, inventive and useful in order to get filed as a patent (USPTO 2007).

Available methods for the identification of novel technologies in patents

Patents are generally considered to have various advantages over other information sources,
such as their coverage and extent (Fendt 1988; Trajtenberg et al. 1997; Debackere et al. 2002;
Knight 2004), availability (Debackere et al. 2002) and the precondition of industrial appli-
cability or usefulness (Granstrand 2000), to mention only a few. Due to the obvious merits of
patent information, several methods for determining novelty have already been established:
On the linguistic level, they are based on patent existence, citations and keywords.
The relationship between patent existence (measured in counts) and the technological life
cycle has been analysed in various studies, and in many cases an S-shaped curve or a double
S-shaped curve was observed (e.g. Achilladelis et al. 1987; Achilladelis et al. 1990; Andersen
1999; Frietsch 2007; Haupt et al. 2007). Even though S-shaped curves facilitate the anticipation
of technological change on a macro level, methods based on patent counts are not appropriate
for the identification of patents with a high degree of novelty, as the counting of patents relies
on the assumption that patents are equally significant This assumption is contradictory to the
existence of ‘key patents’ and ‘basic patents’ (Debackere et al. 2002). ‘Basic patents’ are patents
that protect a fundamental breakthrough in a technical field (Grupp 1997). Consequently, basic
patents are characterized by a high degree of novelty. Key patents’ constitute the basis for
further patents. They are of outstanding value (Debackere et al. 2002). Hence, patent counts are
inadequate for the identification of single patents with a high degree of novelty.
Patent citation analysis has often been considered able to overcome these drawbacks
and provide an appropriate index for estimating the value and the importance of patents
(Trajtenberg 1990). Patent citation analysis is frequently used for the analysis of patents
and gaining insight into a technological field (Debackere et al. 2002; Yoon and Park 2004).
The application of citation analysis in this context has its origin in the bibliometric studies

123
650 J. M. Gerken, M. G. Moehrle

of scientific publications (Trajtenberg 1990) and has been applied to analyse the knowl-
edge flow between technological sectors (e.g. Han and Park 2006; Park et al. 2005) and
geographical regions (e.g. Jaffe et al. 1993; Jaffe and Trajtenberg 1999), to discover
patterns in patent literature (Lee et al. 2009a) and to assess the value (Trajtenberg 1990) as
well as the basicness of patents (Trajtenberg et al. 1997). Backward citations are generally
applied as an indicator of novelty (Reitzig 2003b).
The available methods that are based on patent citation can be divided into three
categories: Methods, (i) which simply count backward citations, (ii) which qualify and
count backward citations and (iii) which compare the structures of backward citations.
(i) Ahuja and Lampert (2001) as well as Schoenmakers and Duysters (2010) refer to the
quantity of backward citations for a measurement of novelty. As patents are bound to
state their technological origins and limitations by citing all preceding patents on
which they are based (Ahuja and Lampert 2001) and which may have an impact on the
patentability of the patent (USPTO 2007), a quantity of zero backward citations
indicates that there are no closely related preceding patents. This also means that a low
number of backward citations suggests a high degree of novelty. In contrast, a high
number of backward citations indicates a close relation to preceding patents and to the
state of the art (Rost 2010), which may be an indication of a low degree of novelty. In
contrast to this, Wartburg et al. (2005) found no evidence to support a relationship
between counts of backward citations and the value added by a patent family.
(ii) An enhancement to the count of backward citation is to be found in methods which qualify
references. This can be achieved by distinguishing between certain types of references,
e.g. patents and scientific literature, through multi-stage citation analysis (von Wartburg
et al. 2005) and also by incorporating patent classifications (Trajtenberg et al. 1997).
Methods that distinguish between certain types of reference rely on the fact that patent
references do not only include backward citations to other patents but also to other types of
publications, such as journals, books and proceedings. Trajtenberg et al. (1997) suggest
that basic research is closely linked to scientific literature. Hence, scientific linkages of
patents may indicate a high degree of novelty. Wartburg et al. (2005) present a multi-stage
citation approach to the measurement of inventive progress. In this approach, the authors
introduce a weighting of references by the inverse number of references listed in a patent.
Based on this weighting, several measures were calculated, relating to the technological
value added. A qualification of references based on classification schemes can be found in
an approach by Trajtenberg et al. (1997). In this approach, ‘technological distance’
(TECHB) is computed by incorporating a three-digit classification scheme, as described
by Hall et al. (2001). Hall et al. (2001) summarized the US patent classes in a three-digit
classification scheme consisting of categories and sub-categories, which supports a
comparison of technological domains on several levels. Based on this classification
scheme, the average distance between a certain patent and the cited patents is calculated.
The distance between the citing patent and the cited patent is (a) set to zero if both patents
are in the same 3-digit class, (b) set to 0.33 if cited and citing patent class are in the same
2-digit class, (c) set to 0.66 if cited and citing patent class are in the same 1-digit class and
(d) set to 1 if citing and cited classes are not in the same class. TechB is calculated as the
average distance to all cited patents.
(iii) Dahlin and Behrens (2005) applied an approach which is closely related to
bibliographic coupling as introduced by Kessler (1963). Dahlin and Behrens (2005)
assume that inventions of high novelty differ from older patents in terms of their
citation structure. Dahlin and Behrens (2005) do not only count the overlapping

123
A new instrument for technology monitoring 651

references of two patents, instead they calculate the similarities in the citation
structures of two patents by dividing the number identical references by the total
number of references in both patents. Thus, the authors ensure that the results are not
influenced by the total number of references.
On the whole, citation analysis represents a valuable enhancement to the field of patent
analysis. However, it also has several drawbacks: (i) the scope of information is limited,
as citation analysis focuses exclusively on bibliographic information and ignores the
description section of patents (Lee et al. 2009a), (ii) patent citation merely indicates
individual links between patents and is thus not suitable for an analysis of overall rela-
tionship (Yoon and Park 2004) and, most importantly, (iii) the references stated in a patent
are subject to strategic decisions, as patent owners can also file citations for their own
patents (USPTO 2007). This may influence the number and choice of references.
With respect to the limitations of patent citation analysis, some keyword-based methods
for patent analysis have already been introduced by various authors (e.g. Kim et al. 2008;
Lee et al. 2009a, b). These methods involve a comparison of the occurrence of keywords in
patents, for example by means of measuring term frequency (Li et al. 2009) or extracting and
classifying keywords (Yoon and Park 2005). Most commercially available patent analysis
tools have long been relying on the measurement and comparison of term frequency (Trippe
2003). But especially in patent analysis, the extraction and comparison of single words is often
too unspecific for a detailed analysis. Therefore, the extraction of multi-words is preferable.
Multi-words are more precise in representing the contents of a patent (Tseng et al. 2007).

SAO-based patent analysis

To overcome the drawbacks of the preceding methods, approaches based on SAO-struc-

tures (Subject-Action-Object-structures) deserve special attention. SAO-structures are also
referred to as problem–solution-structures, as the combination of object and action
incorporates the problem or required function, while the subject provides the solution (see
Moehrle and Geritz 2007; Bergmann et al. 2008), and thus represent the functional rela-
tionship of an invention (Cascini et al. 2004; Choi et al. 2011; Yoon and Kim 2011b).
Accordingly, SAO-structures give insights into the relationship between components of
a technical system (Yoon and Kim 2011b). For example, the SAO structure ‘‘central
differential (S)—comprise (A)—sun gear (O)’’ clearly shows, that the central differential
involves a sun gear and consequently represent structural properties of the invention. Thus,
SAO-structures are expected to be a high-quality representation of patents’ technological
contents. SAO-based patent analysis, generally labeled as semantic patent analysis, has
already been used for several tasks, e.g. inventor profiling (Moehrle et al. 2005), M&A
decision support (Moehrle and Geritz 2007), patent infringement analysis (Bergmann et al.
2008; Park et al. 2011), technology monitoring (Gerken et al. 2010b) and the identification
of technological trends (Choi et al. 2011; Yoon and Kim 2011b). Yoon and Kim (2011a)
apply SAO-analysis to detect signals of new technological opportunities. For this purpose,
outlying patents are identified by calculating the distance to their k-nearest neighbours.

Semantic patent analysis for identifying inventions of high novelty

In order to overcome the drawbacks of the available methods, we are going to introduce the
use of semantic patent analysis for identifying inventions of high novelty. In this approach,
we are relying on semantic patent analyses that have been developed for other tasks. We

123
652 J. M. Gerken, M. G. Moehrle

will extend and apply available methods of semantic patent analysis to calculate degrees of
novelty on the level of single patents.

The process of semantic patent analysis in general

Our method consists of four successive steps, which mainly focus on the extraction of
information and the analysis of the extracted information: First of all, semantic structures
have to be identified in and extracted from patent texts. Secondly, a specific domain- or
situation-related linguistic analysis has to be performed in order to ascertain the relevance
of the extracted semantic structures for the given field of interest. Thirdly, the similarity
measurement takes place, based on the consideration of the significance of specific
structures within the extracted structures. Fourthly, the resulting similarity matrices are
used for the calculation of the novelty of inventions in patents. These four steps will be
described in greater detail in the subsequent sections.

Natural language processing: the extraction of semantic structures

The first step of semantic patent analysis for identifying inventions of high novelty in patents
involves the extraction of semantic structures. The extraction of semantic structures aims at
the analysis of a vast quantity of textual data, which includes the preparatory processing of this
textual data, such as natural language text, for further analysis. Hence, the basis of semantic
patent analysis lies in a syntactic analysis of the patents’ full texts, which can be achieved by
use of part-of-speech tagging. This means that syntactic information is added to the text.
The gap between syntactical analysis and semantic analysis is bridged by focussing
primarily on the technical aspects of inventions, as represented by functional aspects of the
invention. In order to do this, we extract SAO-structure from patents. The extraction of SAO-
TM
structures1 can be achieved with the Knowledgist software by Invention Machine.2 As
characteristic SAO-structures might occur more frequently than less characteristic SAO-
structures, we extract SAO-structures and also count the frequency of these SAO-structures.

Specific domain- or situation-related linguistic analysis

In step two, the identification of semantic structures is followed by a linguistic analysis

of domain- and situation-related elements. The necessity of employing filter modules is
caused by the problem of (i) synonymy and (ii) stop words.
(i) The problem of synonymy arises from the fact that different words may represent the
same general meaning in a specific domain or situation. For example ‘motor vehicle’
and ‘automotive vehicle’ can be seen as general synonyms, as they both describe ‘‘a self-
propelled wheeled vehicle that does not run on rails’’ (Princeton University 2006). As a
matter of fact, general synonyms are very rare, as most synonyms arise from the context.
In addition to this, words are often related to each other in hierarchical terms. Hence, the

1
Although we will focus on SAO-structures in this article, it is noteworthy to show an alternative: Word
n-grams. Word n-grams can be extracted with or without regard to syntactical classes and functions.
Extracting n-grams regardless to syntactical functions cause loss of syntactical information. Nevertheless, n-
grams still have semantic information. n-grams take the co-occurrence of words into account and hence,
highlight a relationship between these words on the content level, as they show that n words co-occur close
together in a patent.
2 TM
For further information on Invention Machine and the Knowledgist see invention-machine.com.

123
A new instrument for technology monitoring 653

necessity to apply a domain- and situation related linguistic analysis is based on the fact
that words may be seen as synonyms in a specific context, though not in general.
(ii) Furthermore, some extremely common words are of little value for the selection and
the comparison of patents; they are thus referred to as stop words (Manning et al.
2008). Comparable to the problem of synonymy, some words may be identified as
stop words in a specific context but not in another. Hence, we distinguish between
domain-specific and general stop words. Domain-specific stop words are extremely
common in the respective technological field and thus of little value for the
discrimination of patents. In contrast, we will call words that occur in practically
every text or sentence, such as ‘the’ and ‘a’, general stop words.

Generating similarity matrices

Prior to the measurement of novelty, a similarity measurement is carried out to highlight

those aspects of the invention that have already been described in older patents. The
statistical comparison of the semantic structures of patents within a given patent set results
in a matrix of similarity values. According to Moehrle (2010) a two-step approach to
establishing textual similarities between patents can be applied: Firstly, a variable mea-
surement is concerned with a pairwise count of overlapping semantic structures between
patents and the overall count of semantic structures within each patent. Secondly, applying
similarity coefficients to the variables leads to similarity measures.
First of all, a variable measurement is required. For this purpose, complete linkage is
applied, as structures are compared without any modifications of their quantity, e.g.
removing of semantic structures which occur twice (see Moehrle 2010). The identical
structures of two or more patents are linked by means of directed connections. The count
of links between two patents represents the overlap of these patents, which is typical
of single-sided measurement. As we are focussing on novelty, we add a chronological
dimension to the variable measurement. Due to this, the variable measurement now
involves forward and backward looking variables—comparable to patent citation analysis.
Forward looking and backward looking overlapping semantic structures and the semantic
structures of each patent are measured according to their availability (see Fig. 1).
(i) As long as only patent i exists, the semantic structures in patent i (ci) can be counted.
(ii) Once patent j occurs, the semantic structures of patent j (cj) and the semantic
structures of patent i, which are also part of patent j (ci(j)), and the semantic structures
of patent j, which are also part of patent i (cj(i)), can be counted.
(iii) Once patent k occurs, the semantic structures of patent k (ck), the semantic structures
of patent i, which are also part of patent k (ci(k)), the semantic structures of patent j,
which are also part of patent k (cj(k)), the semantic structures of patent k, which are
also part of patent i (ci(i)), and the semantic structures of patent k, which are also part
of patent j (ck(i)), can be counted.
(iv) Analogous to steps i–iii, for each occurring patent, the semantic structures and
semantic structures overlapping with older patents can be counted.
ci(j), ci(k) and cj(k) are forward looking variables. Forward looking variables indicate the
share of semantic structures that have been retained throughout successive patents. For
example, ci(j) equal 3 indicates that three semantic structures of patent i also occurred in
patent j. In contrast, cj(i), ck(i) and ck(j) are backward looking variables. For example, cj(i) equal
4 indicates, that four semantic structures of patent j were already mentioned in patent i.

123
654 J. M. Gerken, M. G. Moehrle

Fig. 1 Example of single-sided variable measurement based on Moehrle (2010), with the addition of a
chronological dimension. A to G represent the semantic structures of patents i, j and k. Arrows link identical
semantic structures in patents i, j and k

On the basis of the introduced variables, single-sided-inclusion is a useful similarity

coefficient for measuring the share of a patent that is consistent with preceding patents (Eq. 1):
Backward looking single sided inclusion ¼ cjðiÞ =cj ð1Þ

where cj(i) is the number of semantic structures of patent j, which are also part of patent i,
and cj is the number of semantic structures of patent j (Moehrle 2010).

Novelty calculation

In this article, novelty measures will be calculated on the basis of a similarity matrix. As we
have already outlined that novelty can only be determined in relation to established tech-
nologies, a novelty calculation of the inventions described in the patents of the two patent
sets by means of semantic patent analysis is based on the similarity between a patent and
all preceding patents. The maximum similarity of a patent to the preceding patents can be
regarded as its ‘‘oldness’’: In contrast to this, we considered novelty as the particular share of
a patent that does not resemble the preceding patent in terms of maximum similarity.
Accordingly, we calculate the novelty of a patent in proportion to preceding patents by
subtracting the maximum similarity of a patent to its preceding patents from one (Eq. 2):
Ni ¼ 1  maxðsiðnÞ Þ for all n \ i ð2Þ

where Ni is the novelty of patent i and si(n) is the similarity of patent i to each patent n filed
prior to patent i.

123
A new instrument for technology monitoring 655

Data and research design

In order to test the method of semantic patent analysis for the identification of inventions of
high novelty, we compiled a data set from the automotive industry. For companies in this
industry, competitiveness depends upon the ability to innovate as much as on the ability to
defend their particular market niche. In this context, the four-wheel drive technology is a
noteworthy technology seen from a market perspective but also from a technological point
of view. Especially the leading manufacturer of four-wheel drive vehicles SUBARU3
seems to be an interesting example, as the market for four-wheel drives has expanded
steadily in recent years and is expected to grow still further (Kurmaniak 2008; Pope 2009;
Stockmar 2004). In the subsequent sections we will first describe the compilation of patent
data, followed by the analysis and preparatory processing of data.

Technological foundation: SUBARU’s four-wheel drive patents

Patents corresponding to SUBARU’s four-wheel drive technologies were identified by

means of two different search threads, and two patent sets were established: a limited and a
comprehensive set. The limited patent set no. 1 is narrowly focussed on four-wheel drive
technology, especially regarding differentials. The comprehensive patent set no. 2 focuses
on four-wheel vehicles, in which case the four-wheel drive represents one technology
amongst others.
For both patent sets, patents were collected from the USPTO database. The study was
restricted to patents which had been granted between 1976 and 2006. The outset of the
timeframe was defined by the USPTO database itself, as it provides no full-text patents
dating back to any year before 1976. The end of the timeframe was not defined precisely,
but when the patent search was conducted in 2009, the latest relevant patent had been
issued in 2006, so this date served as a further boundary.
Concerning patent set 1, we focussed on US patent class 475, which is entitled
‘‘planetary systems and components’’. This class is highly significant for four-wheel
drive trains. For instance, sub-class 475/220 includes inventions with a differential gear
as part of the power train, and sub-class 475/230 refers to inventions with a differential
based on bevel gears. The patents that were to be analysed had to bear a close relation
to four-wheel drive technology. In addition to the classification and the company name,
it was decided that they should contain the keyword ‘‘four wheel’’. The patent search
according to the search thread produced a result of 57 patents in total (Table 2).
The second patent search was to produce a wider range of patents in patent set 2. The
basic idea behind this patent set was that there may be inventions related to four-wheel
drive trains that have been allocated to other classes, as complimentary technologies,
technological predecessors or for the purpose of concealment. Therefore, the search was
extended to all USPC classes. We collected additional patents from SUBARU with the aid
of the keyword ‘‘four wheel’’. As a result, we found 225 patents in total (Table 2).

3
For detailed information about SUBARU: http://www.subaru-global.com/.

123
656 J. M. Gerken, M. G. Moehrle

Table 2 Search strings and results for patent sets 1 and 2

Search string Results

Patent set 1 (AN/‘‘fuji jukogyo’’) AND (‘‘four-wheel’’ OR ‘‘four wheel’’) 57 Patents

AND CCL/475/$ AND ISD/19760101-[20061231
Patent set 2 (AN/‘‘fuji jukogyo’’) AND (‘‘four-wheel’’ OR ‘‘four wheel’’) 225 Patents
AND ISD/19760101-[20061231

Patent search was carried out in the USPTO database (uspto.gov)

Data analysis and preparatory processing

The data was analysed and processed in three steps: (i) Patents were evaluated in close
collaboration with experts from the FVA4, (ii) novelty indices were computed by means of
different methods, and (iii) computer-based results were compared with each other, using
the experts’ evaluation as a quality indicator.
First of all, the patents of both patent sets were evaluated manually by use of a qual-
itative scale. Starting with patent set 1, those patents where identified, in which certain
aspects of SUBARU’s four-wheel drive technology had been described for the first time. In
accordance with a four-wheel drive technology scheme by Naunheimer et al. (2007) the
type of transmission, e.g. manual and automatic transmission, the torque transfer between
the axes, the lock mechanism, and the transfer case, were distinguished. On the basis of
SUBARU’s four-wheel drive history, the factor of market relevance was also used as an
additional indicator of an inventions’ novelty (Table 3). All in all, eleven inventions of
high novelty were selected form patent set 1; the novelty of the inventions of the remaining
46 patents was classified as low.
According to this, patent set 1 can be considered a controlled patent set. Patent set 1
forms the core set of patent set 2. Based on the controlled patent set 1, we investigated,
whether any of the inventions had been published before in patents of patent set 2. Fur-
thermore, we identified product technologies, which had not been described in any patents
of patent set 1 (Fig. 2). All in all, thirteen inventions of high novelty were identified in
patent set 2 (Table 3).
Secondly, subsequent to the evaluation of the patents, novelty indices were computed. For
this purpose, we selected methods which analyse novelty on the basis of (i) citation counts,
(ii) qualified citations, (iii) set-related as well as (iv) non-set related citation structures and
(v) semantic patent analysis. In total, ten investigations were carried out for both patent sets.
(i) The analysis of citation counts is based on the method applied by Ahuja and Lampert
(2001). No or few backward citations point to a patent with a high degree of novelty;
a high number of backward citations indicates a relatively low degree of novelty.
Backward citations were thus counted for both patent sets.
(ii) For the analysis of qualified citations the method by Trajtenberg et al. (1997) was chosen.
In addition to analysing patent citations in terms of categories, sub-categories and patent
classes, US sub-classes were equally taken into account in this article, to enable a more
detailed analysis of novelty, or respectively the class-related distance between patents.
(iii) For the analysis of the citation structure, we adapted the approach of Dahlin and
Behrens (2005). We set the novelty of a patent to one minus the maximum of
4
For detailed information about the FVA: http://www.fva-net.de/. The FVA can be seen as the leading
innovation network in the field of drive train technology in Germany. The FVA enhance the collaboration
between industry and science in the field of drive train technology.

123
A new instrument for technology monitoring 657

Table 3 History of the market launch of SUBARU’s four-wheel drive technology, new aspects of these
technologies and their corresponding patents (according to SUBARU Deutschland GmbH (Ed.) 2005 with
additional information and slight modifications, own representation, for a detailed description of innovation
history see Gerken et al. 2010a, c)
Year of New aspects of the 4 9 4 system US- US- Vehicle
market patent patent
launch (Set 1) (Set 2)

1980 Mechanically insertable four-wheel drive and ‘‘dual range’’ 4,241,621 4,170,273 SUBARU
for manual transmission 1800/Leon II
1981 Automatic transmission with multi-platetransfer-4WD 4,480,505 4,480,505 SUBARU
1800/Leon II
1983 Electro-pneumatically insertable 4WD for manual – 4,545,457 Libero and Justy
transmission
1988 Permanent 4WD with central differential for manual 4,787,269 4,805,721 Coupé XT
transmission and automatic transmission (ACT-4)
1989 Permanent 4WD with self-locking viscous coupling and 4,819,506 4,819,506 Legacy
central differential
1989 ECVT and insertable 4WD 4,846,765 4,715,467 Justy
1991 Permanent 4WD with VTD: planetary gear-type central 5,066,268 5,066,268 Gran Turismo SVX
differential with a multi-disc-clutch
1998 Integration of VDC with optimized sensor system – 5,734,595 Premium models of
the legacy type
series
2004 Permanent 4WD with optimized VTD, VDC, l-estimator, 6,595,086 6,595,086 Legacy 3.0 since
automatic transmission and communication between 2004
different control units
2005 STi-AWD with Impreza WRX STi
(a) six-speed manual transmission 6,830,528 6,830,528
(b) variable dial 6,878,085 6,878,085
(c) DCCD, planet set—centre differential, electromagnetic 7,127,343 7,127,342
clutch
(d) helical geared front differential 7,029,415 7,029,415

similarity to preceding patents (Eq. 2). We calculated the similarity of citation

structures for a set-related citation structure.
(iv) On the basis of the same method, we also calculated the novelty in terms of a non-
set-related citation structure.
(v) Semantic patent analysis was conducted according to the process described in
Semantic patent analysis for identifying inventions of high novelty. In order to
overcome the problems arising from synonymy and the occurrence of stop words
(described in Specific domain- or situation-related linguistic analysis), certain filter
modules that have been devised by Forschungsvereinigung Antriebstechnik (FVA)
and are set to specific technical domains and situations (e.g. the purpose of the
semantic patent map), find application. We apply a synonym filter that serves to
replace more than 6,500 synonyms (e.g. ‘epicyclic gear’ and ‘satellite gear’ were
replaced by ‘planetary gear’). General and domain-specific stop word filter were
applied, which all in all eliminated almost 400 stop words (e.g. ‘figure’ and ‘example’
were deleted). For reasons of robustness we conducted three tests: (i) We applied two
different similarity coefficients, an Inclusion and a Jaccard coefficient (Moehrle
2010). (ii) We analysed the effect of speech filters in total. As can be seen in Table 4,
speech filters ensure that several SAO structures containing no content-oriented
information were deleted and the quality of the SAO analysis was enhanced. (iii) We

123
658 J. M. Gerken, M. G. Moehrle

Fig. 2 Schematic illustration of the identification of patents related to inventions of high novelty. White
squares symbolize patents related to inventions of high novelty; black circles symbolize the remaining
patents, arrows indicate that a patent of patent set 2 includes earlier information than the corresponding
patent in patent set 1

conducted a semantic patent analysis without using a synonym filter but with a stop
word filter. The effect was that the mean value of similarities between all patents in a
set decreased by about 13% from 0.0310 to 0.0269 in patent set 1 and by about 14%
from 0.0137 to 0.0117 in patent set 2. Remarkably, the influence on the novelty
measure was considerably lower. The effect was that the mean value of novelty
increased by about 3% in patent set 1 and by about 5% in patent set 2. Obviously, the
synonym filter mainly increased the similarity between patents that are not closely
related to one another, but not of patents that are per se similar.
Thirdly, the results of our investigations were compared with each other, using the
experts’ evaluation as a quality indicator. For this purpose, we conducted a U test, we
calculated precision and recall and Spearman’s rank correlation.
In order to test, whether inventions of high novelty identified by experts differ in their
novelty indices from patents not related to inventions with high novelty, Mann–Whitney
U tests were conducted (see Mann and Whitney 1947; Buehl 2010).
Next, precision and recall were calculated. Precision and recall are quality measures in
the field of information retrieval, where precision refers to the fraction of relevant docu-
ments within the corpus of retrieved documents while recall refers to the total number of
relevant documents that have been retrieved (Manning et al. 2008). In other words, recall
can be seen as the completeness of the retrieved documents (Eq. 3), precision signifies the
accuracy of the retrieved documents (Eq. 4) (Stock 2007):
Recall ¼ jA \ Bj=jAj ð3Þ
Precision ¼ jA \ Bj=jBj ð4Þ
where A symbolizes relevant documents while B stands for retrieved documents (van
Rijsbergen 1981).
In research tasks, it is necessary to maintain a balance between precision and recall, as
these two influence each other (Manning and Schütze 2005): Reading all patents leads to a
recall equal one. Due to the fact, that not all patents are relevant, precision decreases. In

123
A new instrument for technology monitoring 659

Table 4 Results of SAO extraction with and without SAO filtering

Set Filtering Number of SAO structure Median Mean value Max. Min.

1 Yes 7,874 125 138.14 586 36

1 No 16,615 202 291.49 1,966 54
2 Yes 32,968 135 146.52 586 31
2 No 56,612 215 251.61 1,966 54

contrast, a selection of high-ranking patents leads to an increase in precision while the

recall is reduced, because some relevant patents do not belong to the high-ranking patent
set (Manning et al. 2008). In our concept, eleven recall levels form a basis for the com-
putation of corresponding precision values. Precision is calculated for those patents which
are the first to exceed recall levels between 0 and 100% in steps of 10% per level (see
Manning and Schütze 2005).
For the purpose of calculating precision we developed a specific algorithm (see
appendix A), in which we listed the patents according to their respective level of novelty.
For each recall level we then identified the one patent that exceeded a predetermined limit
and, if this appeared feasible, established the respective value of precision5.
Finally, rank correlations between the investigations were established. For this purpose,
Spearman’s correlation (Buehl 2010) was applied. Rank correlations highlight the simi-
larity between measures of novelty and constitute a basis for the characterization of novelty
measures.

Results

The Mann–Whitney U test serves to determine whether there is a significant difference

in the ranking of highly novel and less novel patents. In our case, the only significant
difference related to novelty calculated by means of semantic patent analysis (Table 5).
For patent set 1, the novelty indices calculated with TechB also pointed to a significant
difference. This suggests that semantic patent analysis may be more suitable for the
identification of inventions of high novelty than other methods.
Table 6 shows the results of the calculation of precision and recall for six novelty
indices, including one novelty index based on semantic patent analysis for reasons of
robustness, with respect to each patent of both patent sets. Two major findings deserve to
be mentioned first of all: Semantic patent analysis outperforms other available methods of
identifying inventions of high novelty. And novelty calculated by means of semantic patent
analysis is more differentiated than novelty calculated by other methods. In greater detail,
our findings are:
(i) In general, the smallest number of patents has to be read in order to reach a specific
recall level, if semantic patent analysis is applied. Consequently, precision reaches its
highest level, if semantic patent analysis is used for novelty assessment. Only in very
few cases, semantic patent analysis is outperformed by other methods. This can be
5
In the algorithm we took into account, that several patents may have the same level of novelty. In such
cases we assume that analysts read patents stepwise. Every novelty value has to be considered as one step.
Hence, if an analyst reads one patent with a novelty equal 0.2, he also reads all other patents with a novelty
of 0.2 independent of the relevance of the first patent he has read with a novelty equal 0.2. In some of these
cases it makes no sense to report a precision value on a specific level of recall.

123
660 J. M. Gerken, M. G. Moehrle

observed with respect to both patent sets. Assessing novelty by means of citation
counts performs well in both patent sets on a very low recall level. On higher recall
levels, citation counts are outperformed by TechB as well as by non-set related
citation similarity and set related citation similarity. On most recall levels, TechB has
a higher precision than non-set related citation similarity and set related citation
similarity. On high recall levels (0.9 and 1.0), set related and non-set related citation
similarity perform slightly better than TechB.
(ii) Some methods proved inapplicable for a calculation of precision on all recall levels.
In the first patent set precision could only be calculated on all recall levels for novelty
measured by means of semantic patent analysis. In the second patent set precision
could be ascertained on all recall levels for novelty established by means of semantic
patent analysis as well as by TechB. Accordingly, semantic patent analysis possesses
the greatest discriminatory power in novelty assessment.
(iii) The results of semantic patent analysis without synonym filter are comparable to the
results of semantic patent analysis with synonym filter. This may suggest that
semantic patent analysis is also a useful method for novelty calculation without
domain-specific modifications.
The results regarding rank correlation highlight three major findings (Table 7): novelty
calculated by means of semantic patent analysis has the strongest correlation with the experts’
choices. The performance of novelty indices is influenced by the extent of patent data. And
novelty indices do not only differ in terms of performance but also in terms of results.
The results of Spearman’s rank correlation predominantly confirm our findings con-
cerning recall and precision. Semantic patent analysis performs best in the identification of
inventions of high novelty, as it shows the strongest correlation with the experts’ choices
regarding both patent sets. TechB is also marked by a strong and significant correlation
with experts’ choices in regard to patent set 1, but less so regarding patent set 2.
Furthermore, we already observed that all methods perform better in the restricted
patent set 1. Here, the results of Spearman’s rank correlation differ in so far, as the set
related citation similarity correlates slightly more with the experts’ choice in reference to
patent set 2 than in reference to patent set 1. But the correlation between set related citation
similarity as well as non-set related citation similarity and the experts’ choice is of minor
significance for both patent sets.
Considering the correlations between different methods it becomes obvious that some
methods are similar in terms of results. But in some cases, the results also bear a minor

Table 5 Results of Mann–Whitney U Test

Semantic Semantic Non-set related Set related Citation TechB
(Jaccard) (Inclusion) citation citation counts
similarity similarity

Patent set 1
Mann–Whitney U 103.000 95.000 226.000 244.000 252.500 117.000
Z -3.033 -3.195 -0.555 -0.241 -0.010 -2.756
Asymp. Sig. (2-tailed) 0.002 0.001 0.579 0.810 0.992 0.006
Patent set 2
Mann–Whitney U 761.00 796.00 1200.50 1287.50 1238.00 1192.50
Z -2.71 -2.55 -0.79 -0.48 -0.62 -0.82
Asymp. Sig. (2-tailed) 0.01 0.01 0.43 0.63 0.54 0.41

123
 Table 6 Recall and precision for patent sets 1 and 2
Recall level (%) Patent set 1 Patent set 2

Semantic Semantic Non-set related Set related Citation TechB Semantic Semantic Non-set related Set related Citation TechB
(Jaccard) (Inclusion) citation citation counts (Jaccard) (Inclusion) citation citation counts
similarity similarity similarity similarity

Recall reached by top n patents

0.00 1 1 2 1 1 73 12 39
0.10 3 3 5 12 13 78 32 45
0.20 4 4 18 8 6 34 16 84 62 55
0.30 8 8 21 11 7 38 49 108 85 68
0.40 10 12 10 45 61 138 84
A new instrument for technology monitoring

0.50 13 15 24 32 13 64 67 139 86
0.60 17 16 32 38 14 69 83 154 150 93
0.70 18 17 36 43 43 23 94 89 166 170 114
0.80 25 22 41 45 50 24 98 103 158 185 139
0.90 29 23 44 53 56 33 109 105 168 178 203 198
1.00 41 40 47 56 57 52 157 160 169 182 217 209
Precision
0.00 1.00 1.00 0.50 1.00 1.00 0.01 0.08 0.03
0.10 0.67 0.67 0.40 0.17 0.15 0.03 0.06 0.04
0.20 0.75 0.75 0.17 0.38 0.50 0.09 0.19 0.04 0.05 0.05
0.30 0.50 0.50 0.19 0.36 0.57 0.11 0.08 0.04 0.05 0.06
0.40 0.50 0.42 0.50 0.13 0.10 0.04 0.07
0.50 0.46 0.40 0.25 0.19 0.46 0.11 0.10 0.05 0.08
0.60 0.41 0.44 0.22 0.18 0.50 0.12 0.10 0.05 0.05 0.09
0.70 0.44 0.47 0.22 0.19 0.19 0.35 0.11 0.11 0.06 0.06 0.09
661

123
 Table 6 continued
662

Recall level (%) Patent set 1 Patent set 2

123
Semantic Semantic Non-set related Set related Citation TechB Semantic Semantic Non-set related Set related Citation TechB
(Jaccard) (Inclusion) citation citation counts (Jaccard) (Inclusion) citation citation counts
similarity similarity similarity similarity

0.80 0.36 0.36 0.22 0.20 0.18 0.38 0.11 0.11 0.07 0.06 0.08
0.90 0.34 0.43 0.23 0.19 0.18 0.30 0.11 0.11 0.07 0.07 0.06 0.06
1.00 0.27 0.28 0.23 0.20 0.19 0.21 0.08 0.08 0.08 0.07 0.06 0.06
Grey-coloured fields indicate that precision could not be calculated on a specific level of recall. Semantic patent analysis by use of the Jaccard coefficient was conducted for
robustness reasons
J. M. Gerken, M. G. Moehrle
A new instrument for technology monitoring 663

resemblance to those obtained by means of other methods. A strong correlation for both
patent sets can be found between set related similarity and non-set related similarity. Both
set related and non-set similarity also bear a resemblance to novelty measured by semantic
patent analysis. Furthermore, there is a weak correlation between citation counts and
TechB. This may indicate that citation counts marginally influence TechB, as TechB is also
based on citations. For both patent sets, the correlation between TechB on the one hand,
and set related citation similarity, non-set related citation similarity and semantic similarity
on the other, is insignificant.
A detailed analysis of the differing results obtained by TechB and semantic patent
analysis shows that in both patent sets TechB outperforms semantic patent analysis with
regard to US patent 4,819,506 and US patent 6,878,085. US patent 4,819,506 is primarily
assigned to US patent class 74, which is entitled ‘machine element or mechanism’. Only one
out of seven patents cited by US patent 4,819,506 is also assigned to US patent class 74.
Comparable to this, US patent 6,878,085 is primarily assigned to US patent class 475, but
only one out of nine of its cited patents is also assigned to US patent class 475. This indicates
that TechB performs well, if there is a swift from one technological domain, as represented
by certain patent classes, to another technological domain. In the case of US patent 4,819,506
we observed a swift from US patent class 180, which contains the majority of cited patents, to
US class 74, which is the primary patent class of US patent 4,819,506. Similar to that, we
observed that a swift from US patent class 180 to US patent class 475 was the most common.
In contrast, TechB works less well for US patents 6,830,528, 4,480,505 and 7,029,415
in both patent sets and for US patent 4,805,721 in the second patent set. In these patents,

Table 7 Spearman’s rank correlation coefficients between different types of novelty calculation for patent
set 1 and patent set 2
Semantic Semantic Non-set Set Citation TechB Expert
(Jaccard) (Inclusion) related related counts
citation citation
similarity similarity

Semantic Set 1 1.000 0.963** 0.446** 0.156 0.069 0.044 0.405**

(Jaccard) Set 2 1.000 0.981 0.557 0.295 -0.133 0.072 0.181
Semantic Set 1 0.963** 1.000 0.437** 0.190 0.087 -0.006 0.427**
(Inclusion) Set 2 0.981 1.000 0.549 0.293 -0.129 0.073 0.171
Non-set Set 1 0.446** 0.437** 1.000 0.527** -0.094 -0.056 0.074
related Set 2 0.557 0.549 1.000 0.537 -0.139 0.047 -0.053
citation
similarity
Set related Set 1 0.156 0.190 0.527** 1.000 -0.075 0.063 -0.032
citation Set 2 0.295 0.293 0.537 1.000 -0.036 0.093 0.032
similarity
Citation counts Set 1 0.069 0.087 -0.094 -0.075 1.000 0.092 -0.001
Set 2 -0.133 -0.129 -0.139 -0.036 1.000 0.137 0.041
TechB Set 1 0.044 -0.006 -0.056 0.063 0.092 1.000 0.373**
Set 2 0.072 0.073 0.047 0.093 0.137 1.000 0.056
Expert Set 1 0.405** 0.427** 0.074 -0.032 -0.001 0.373** 1.000
Set 2 0.181 0.171 -0.053 0.032 0.041 0.056 1.000

** The correlation is significant at the 0.01 level (two-sided)

Grey-coloured fields indicate that semantic analysis by use of the Jaccard coefficient was included for robustness
reasons

123
664 J. M. Gerken, M. G. Moehrle

only minor patents from other patent classes are cited. But they strongly differ from
preceding patents in semantic terms, as these patents involve new components, such as a
permanent four-wheel drive and a helical geared front differential in a four-wheel drive
train. Hence, a semantic patent analysis obviously outperforms TechB in comparing pat-
ents from identical technological domains. On the other hand, TechB seems to be a reliable
indicator of changing technological domains between a patent and its cited patents. In this
context, TechB focuses on trajectories, as represented by citation links. Semantic patent
analysis is fully independent of direct linkage between patents.

Discussions and conclusions

Semantic patent analysis, especially its application for measuring the novelty of inventions,
still is a fairly new field of research. In addition to a summary of our results the following
conclusions will refer to theoretical perspectives and to managerial and political impli-
cations as well as to certain limitations. Furthermore, we shall put a special emphasis on
scientific perspectives in the further advancement of semantic patent analysis.

Conclusions

In this article, we introduced a new method for the identification of patents related to
inventions of high novelty. For this purpose, we applied semantic patent analysis and
established each examined patent’s distance from prior art as represented by preceding
patents. In the course of a case study, we compared our method with three established
methods, based on patent citation data, with reference to two different patent sets and
analysed which method is the most adequate for the identification of patents related to
inventions of high novelty. We evaluated the results by use of measures borrowed from
information science: precision and recall. In both patent sets semantic patent analysis mostly
manages to outperform other methods; only under certain conditions TechB performs better.
Our findings contribute to (i) the field of novelty measurement, to (ii) the theory of
patent and technology management but also to (iii) the theory of technological change. Our
findings lead to an advancement of the theory of novelty measurement, as we have
introduced a semantic method, using functions within patent texts to generate measures of
novelty. Our findings help to advance the theory of technology and patent management,
because measuring the novelty of an invention protected by a patent may also be an
enhancement to patent valuation, as the value of a patent also depends upon its novelty
(Reitzig 2003a). These results contribute to the measurement of inventions’ novelty, which
is a basic element for the understanding of technological change. This may provide a
valuable support to scientists and practitioners alike.
Seen from a managerial perspective, derived from the phenomenon of information
overflow (Bergmann et al. 2008), the use of sematic patent analysis for the identification of
patents related to inventions of high novelty may have two implications, one for patent
management and one for technology monitoring. For patent management it considerably
expedites the reduction of efforts and uncertainty in monitoring patents, which has so far
been a particularly time-consuming task. The reliable identification of highly important
information opens up the opportunity to focus resources on this particular information
while less important information is only given a fast screening. The application of novelty
measures enables a ranking of new (and old) patents, directing the patent assessor’s

123
A new instrument for technology monitoring 665

attention to high-ranking patents which may be more significant than the low-ranking ones.
Novelty measures can also find application in the valuation of patents.
As concerns technology monitoring, our method is not a complete monitoring instrument
as it is not able to suggest a way of devising new technologies. But it can shed light on
technology fields where intensive progress seems to be emerging (measured in the devel-
opment of the related patents’ novelty) and thus help identify promising technology areas.
Regarded from a political point of view, a semantic patent analysis may provide
valuable support to patent offices. As novelty is a precondition for the granting of a patent,
patent offices are required to evaluate the novelty of each patent that has been filed.
Consequently, the use of semantic patent analysis for the identification of patents related
to inventions of high novelty and the appropriate assessment of patents’ novelty is also
capable of changing the process of patent assessment in patent offices. Even in cases where
the search has to be carried out across a range of patent classes, a semantic patent analysis
may be helpful: Separating the analysis over the classes (which represent different tech-
nological fields) can provide class specific measures for novelty and help the patent officer
evaluate the invention.
Our case study is subject to certain limitations concerning (i) the scope of the case study
as such, (ii) the scope of the patent searches, (iii) the fact, that the relationship between
semantic structures and functional relationships have not yet been explored in detail and
(iv) the necessity to develop different kinds of filters.
Firstly, the exploration of our method has so far been limited to the field of mechanical
engineering. This area appears suitable for semantic patent analysis, as the wording tends
to be simpler as well as more standardized than in other technological fields. For example,
patents from the pharmaceutical area seem to be more difficult, as they involve chemical
symbols and often highly complex chemical compounds.
Secondly, by comparing the results from both patent sets, we found that all approaches
showed a better performance with regard to the first patent set. Hence, the assessment of
novelty is not only dependent on the chosen method, but also on the data. In patent set 1 we
were unable to find a corresponding patent for each technology involved in SUBARU’s
innovations. Hence, this data set would be too small and incomplete for an effective tech-
nology monitoring in the field of four wheel drive trains. In contrast, patent set 2 rather seems
to be too extensive, as several patents of high novelty that are not directly related to the four
wheel drive train can actually be found in this patent set. Therefore, we have come to the
conclusion that there is still a gap concerning the unequivocal definition of technological
fields in patent data, even though patent classification serves to facilitate orientation.
Thirdly, we used semantic structures, i.e. subject-action-object structures, to analyse the
novelty of inventions. In this context, we presumed that these semantic structures represent
functional aspects of the inventions. Up to now, the relationship between SAO-structures
extracted from patents’ full texts and the functional aspects of inventions has not been
explored in detail.
Fourthly, we developed different kinds of speech filter in this case study. For robustness
reasons we tested the influence of domain-specific filters. Our results actually indicate that
semantic patent analysis for novelty measurement may be applied without such filters. But
one has to be aware of the fact that we analysed patents from a clearly defined patent set.
Elaborate speech filters may be more important in broader technological fields. The
development of filters may seem to be time-consuming. However, these filters merely have
to be generated once for a technological domain and can then be reused. Furthermore, we
see a perspective in approaches that deal with this problem by using available ontologies
(e.g. WordNet).

123
666 J. M. Gerken, M. G. Moehrle

Scientific perspectives

Although semantic patent analysis worked very well in our particular patent sets, there still
are some drawbacks and a need for enhancement, especially (i) in testing further full-text
based methods for patent analysis, (ii) in profiling available patent-based methods, (iii) in
SAO weighting, (iv) in similarity calculation, (v) in answering the question, what is new
and, maybe most importantly, (vi) in differentiating between types of novelty.
(i) In this article, SAO-based patent analysis was tested for novelty measurement, which
was mainly motivated by drawbacks of patent citation analysis and keyword analysis.
Comparable to SAO-based patent analysis, the analysis of patents by use of n-grams
(word chains with n words) (for details see e.g. Egghe 2000; Manning and Schütze
2005) might also be a promising approach and remains to be tested in detail.
(ii) As mentioned before, TechB performs better with regard to patents citing patents
form different patent classes. This phenomenon calls for meticulous investigation,
and we are challenged to find ways of combining both approaches, as they both have
their respective advantages and disadvantages.
(iii) In our approach, we used SAO structures all with the same weights. Prospectively,
different weights may be assigned to the SAO structures. For instance, the weights
may be calculated using an SAO structure’s frequency within one patent in
proportion to the frequency in a textual corpus.
(iv) Similarity calculation has been discussed in respect to various areas (Sternitzke and
Bergmann 2009; Jeong et al. 2008; Kangasabai and Pan 2008) and there are several
methods of linking similar objects and calculating the similarity between these
objects (Moehrle 2010). Hence, similarity calculation has to be approached from a
more theoretical direction, as in our case study we applied a very common similarity
measure. We defined novelty as the distance between a patent and the most similar
preceding patent. But there possibly are more complex relationships between patents,
as novelty may be limited to a certain aspect in one patent, and to completely
different aspects in another. Hence, limitations of novelty are not restricted to the
most similar patent, but can, in different forms, also be traced back to the second,
third and fourth most similar patents.
(v) Our approach could be usefully complemented by co-word-analysis (see for example
Callon et al. 1991; Engelsman and van Raan 1994; An and Wu 2011). This would not
only facilitate a quantitative measure of novelty, but would also allow for gaining
insights into textual structures that might characterize a technological field.
(vi) In this article, we focussed on a very general understanding of novel inventions, even
though there are various types of novelty (see Theoretical background). Indeed, it is
possible to distinguish between different types of novelty, especially recombinant
and pioneering novelty, by way of patent information.

Acknowledgments The cited case study was produced in the course of a joint project with the For-
schungsvereinigung Antriebstechnik (FVA). We wish to thank the FVA and all industrial members for their
contributions and their support. Furthermore, we would like to thank Dipl.-Ing. (FH) Jens Potthast for
extensive programming efforts on the PatVisor, Dr. Lothar Walter for commenting an earlier version of
this paper and two anonymous reviewers for their constructive and helpful comments.

Appendix A: algorithm for the calculation of precision

See Fig. 3.

123
A new instrument for technology monitoring 667

Fig. 3 Flow chart of the algorithm for the calculation of precision

123
668 J. M. Gerken, M. G. Moehrle

References

Achilladelis, B., Schwarzkopf, A., & Cines, M. (1987). A study of innovation in the pesticide industry:
Analysis of the innovation record of an industrial sector. Research Policy, 16(2–4), 175–212.
Achilladelis, B., Schwarzkopf, A., & Cines, M. (1990). The dynamics of technological innovation: The case
of the chemical industry. Research Policy, 19(1), 1–34.
Ahuja, G., & Lampert, C. M. (2001). Entrepreneurship in the large corporation: A longitudinal study of how
established firms create breakthrough inventions. Strategic Management Journal, 22(6–7), 521–544.
An, X. Y., & Wu, Q. Q. (2011). Co-word analysis of the trends in stem cells field based on subject heading
weighting. Scientometrics, 88(1), 133–144.
Andersen, B. (1999). The hunt for S-shaped growth paths in technological innovation: A patent study.
Journal of Evolutionary Economics, 9(4), 487–526.
Anderson, P., & Tushman, M. L. (1990). Technological discontinuities and dominant designs: A cyclical
model of technological change. Administrative Science Quarterly, 35(4), 604–633.
Bergmann, I., Butzke, D., Walter, L., Fuerste, J. P., Moehrle, M. G., & Erdmann, V. A. (2008). Evaluating
the risk of patent infringement by means of semantic patent analysis: The case of DNA-chips. R&D
Management, 38(5), 550–562.
Buehl, A. (2010). PASW 18: Einführung in die moderne Datenanalyse (12th ed.). München: Pearson
Studium.
Callon, M., Courtial, J. P., & Laville, F. (1991). Co-word analysis as a tool for describing the network of
interactions between basic and technological research: The case of polymer chemistry. Scientometrics,
22(1), 155–205.
Cascini, G., Fantechi, A., & Spinicci, E. (2004). Natural language processing of patents and technical
documentation. In S. Marinai & A. Dengel (Eds.), Document analysis systems VI (pp. 89–92). Berlin:
Springer.
Chandy, R. K., & Tellis, G. J. (1998). Organizing for radical product innovation: The overlooked role of
willingness to cannibalize. Journal of Marketing Research, 35(4), 474–487.
Choi, S., Yoon, J., Kim, K., Lee, J. Y., & Kim, C. (2011). SAO network analysis of patents for technology
trends identification: A case study of polymer electrolyte membrane technology in proton exchange
membrane fuel cells. Scientometrics, 88(3), 863–883.
Christensen, C. M., & Overdorf, M. (2000). Meeting the challenge of disruptive change. Harvard Business
Review, 78(2), 66–77.
Dahlin, K. B., & Behrens, D. M. (2005). When is an invention really radical? Defining and measuring
technological radicalness. Research Policy, 34(5), 717–737.
Debackere, K., Verbeek, A., Luwel, M., & Zimmermann, E. (2002). Measuring the progress and evolution
in science and technology—II: The multiple uses of technometric indicators. International Journal of
Management Reviews, 4(3), 213–231.
Egghe, L. (2000). The distribution of N-grams. Scientometrics, 47(2), 237–252.
Engelsman, E. C., & van Raan, A. F. J. (1994). A patent-based cartography of technology. Research Policy,
23(1), 1–26.
Fendt, H. (1988). Technische Trends rechtzeitig erkennen—Patentschriften gewähren Blicke hinter die
Kulissen von F&E. Havard Manager, 10(4), 72–80.
Fleming, L. (2001). Recombinant uncertainty in technological search. Management Science, 1, 117–132.
Fleming, L., Mingo, S., & Chen, D. (2007). Collaborative brokerage, generative creativity, and creative
success. Administrative Science Quarterly, 52(3), 443.
Fleming, L., & Sorenson, O. (2001). Technology as a complex adaptive system: Evidence from patent data.
Research Policy, 30(7), 1019–1039.
Frietsch, R. (2007). Patente in Europa und der Triade: Strukturen und deren Veränderung. Karlsruhe:
Fraunhofer Institut für System- und Innovationsforschung.
Gerken, J. M., Moehrle, M. G., & Walter, L. (2010a). Patents as an information source for product fore-
casting: Insights from a longitudinal study in the automotive industry. R&D Management Conference
2010 Proceedings. Manchester.
Gerken, J. M., Moehrle, M. G., & Walter, L. (2010b). Semantische Patentlandkarten zur Analyse techno-
logischen Wandels: Eine Längsschnittstudie aus der Allradtechnik. In J. Gausemeier (Ed.). 6. Sym-
posium für Vorausschau und Technologieplanung (pp. 325–349). Paderborn: Heinz Nixdorf Institut
Gerken, J. M., Walter, L., & Moehrle, M. G. (2010c). Semantische Patentlandkarten. Einsatz semantischer
Patentlandkarten im Anwendungsfeld der Antriebstechnik—Eine explorative Analyse am Beispiel der
Planentengetriebe. Heft Nr. 924 der Forschungsvereinigung Antriebstechnik. Frankfurt/Main: VDMA.

123
A new instrument for technology monitoring 669

Granstrand, O. (2000). The Economics and management of intellectual property: Towards intellectual
capitalism. Cheltenham: Edward Elgar.
Grupp, H. (1997). Messung und Erklärung des technischen Wandels: Grundzüge einer empirischen Inno-
vationsökonomik. Berlin: Springer.
Hall, B. H., Jaffe, A. B., & Trajtenberg, M. (2001). The NBER patent citations data file: lessons, insights and
methodological tools. NBER Working Paper 8498.
Han, Y., & Park, Y. (2006). Patent network analysis of inter-industrial knowledge flows: The case of Korea
between traditional and emerging industries. World Patent Information, 28(3), 235–247.
Hargadon, A., & Sutton, R. I. (1997). Technology brokering and innovation in a product development firm.
Administrative Science Quarterly, 42(4), 716–749.
Haupt, R., Kloyer, M., & Lange, M. (2007). Patent indicators for the technology life cycle development.
Research Policy, 36(3), 387–398.
Jaffe, A. B., & Trajtenberg, M. (1999). International knowledge flows: Evidence from patent citations.
Economics of Innovation and New Technology, 8(1/2), 105–136.
Jaffe, A. B., Trajtenberg, M., & Henderson, R. (1993). Geographic localization of knowledge spillovers as
evidenced by patent citations. The Quarterly Journal of Economics, 108(3), 577–598.
Jeong, B., Lee, D., Cho, H., & Lee, J. (2008). A novel method for measuring semantic similarity for XML
schema matching. Expert Systems with Applications, 34(3), 1651–1658.
Kangasabai, R., & Pan, H. (2008). Method of text similarity measurement. US-Patent 7,346,491 B2.
Kessler, M. M. (1963). Bibliographic coupling between scientific papers. American Documentation, 14(1),
10–25.
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.
Knight, H. J. (2004). Patent strategy for researchers and research managers, (2nd ed.). Chichester: Wiley.
Kurmaniak, C. (2008). Electromagnetics comes through in the clutch. ANSYS Advantage, 2(3), 30–31.
Lee, C., Jeon, J., & Park, Y. (2011). Monitoring trends of technological changes based on the dynamic
patent lattice: A modified formal concept analysis approach. Technological Forecasting and Social
Change, 78(4), 690–702.
Lee, S., Yoon, B., Lee, C., & Park, J. (2009a). Business planning based on technological capabilities: Patent
analysis for technology-driven roadmapping. Technological Forecasting and Social Change, 76(6),
769–786.
Lee, S., Yoon, B., & Park, Y. (2009b). An approach to discovering new technology opportunities: Keyword-
based patent map approach. Technovation, 29(6–7), 481–497.
Li, Y., Wang, L., & Hong, C. (2009). Extracting the significant-rare keywords for patent analysis. Expert
Systems with Applications, 36(3), 5200–5204.
Lichtenthaler, E. (2004). Technological change and the technology intelligence process: A case study.
Journal of Engineering and Technology Management, 21(4), 331–348.
Mann, H. B., & Whitney, D. R. (1947). On a test of whether one of two random variables is stochastically
larger than the other. Annals of Mathematical Statistics, 18(1), 50–60.
Manning, C. D., Raghavan, P., & Schütze, H. (2008). Introduction to information retrieval. Cambridge:
Cambridge University Press.
Manning, C. D., & Schütze, H. (2005). Foundations of statistical natural language processing. Cambridge:
MIT Press.
Moehrle, M. G. (2010). Measures for textual patent similarities: A guided way to select appropriate
approaches. Scientometrics, 85(1), 95–109.
Moehrle, M. G., & Geritz, A. (2007). Developing acquisition strategies based on patent maps. In T. Khalil &
Y. Hosni (Eds.), Management of technology: New directions in technology management (pp. 19–29).
Oxford: Elsevier.
Moehrle, M. G., Walter, L., Geritz, A., & Müller, S. (2005). Patent-based inventor profiles as a basis for
human resource decisions in research and development. R&D Management, 35(5), 513–524.
Naunheimer, H., Novak, W., & Ryborz, J. (2007). Fahrzeuggetriebe: Grundlagen, Auswahl, Auslegung und
Konstruktion. Berlin: Springer.
Park, H., Yoon, J., & Kim, K. (2011). Identifying patent infringement using SAO based semantic techno-
logical similarities. Scientometrics, 90(2), 515–529.
Park, Y., Yoon, B., & Lee, S. (2005). The idiosyncrasy and dynamism of technological innovation across
industries: Patent citation analysis. Technology in Society, 27(4), 471–485.
Pope, B. (2009). All-wheel-drive suppliers get grip on changing market. Resource document. Accessed
March 8, 2010 http://wardsauto.com/ar/suppliers_grip_market_090427/.
Princeton University. (2006). WordNet 3.0. Resource document. Accessed March 23, 2011, from http://
wordnetweb.princeton.edu/perl/webwn.

123
670 J. M. Gerken, M. G. Moehrle

Reitzig, M. (2003a). What do patent indicators really measure? A structural test of ‘novelty’ and ‘inventive
step’ as determinants of patent profitability. LEFIC Working paper 2003–1. Copenhagen, DK.
Reitzig, M. (2003b). What determines patent value? Insights from the semiconductor industry. Research
Policy, 32(1), 13–26.
Rost, K. (2010). The strength of strong ties in the creation of innovation. Research Policy. doi:
10.1016/j.respol.2010.12.001.
Schoenmakers, W., & Duysters, G. (2010). The technological origins of radical inventions. Research Policy,
39(8), 1051–1059.
Schumpeter, J. A. (1934). The theory of economic development: an inquiry into profits, capital, credit,
interest, and the business cycle. Cambridge: Harvard University Press.
Sood, A., & Tellis, G. J. (2005). Technological evolution and radical innovation. Journal of Marketing,
69(3), 152–168.
Sternitzke, C., & Bergmann, I. (2009). Similarity measures for document mapping: A comparative study on
the level of an individual scientist. Scientometrics, 78(1), 113–130.
Stock, W. G. (2007). Information retrieval: Informationen suchen und finden. München: Oldenbourg.
Stockmar, J. (2004). Das große Buch der Allradtechnik. Stuttgart: Motorbuch-Verl.
SUBARU Deutschland GmbH (Ed.). (2005). 33 Jahre SUBARU-Allradantrieb. Resource document.
Accessed May 15, 2009, from http://www.subaru-presse.de/fileadmin/templates/downloads/awd/
PressemappeSubaruAllrad-Technologie04-2005SUB.doc.
Trajtenberg, M. (1990). A penny for your quotes: Patent citations and the value of innovations. The Rand
Journal of Economics, 21(1), 172–187.
Trajtenberg, M., Henderson, R., & Jaffe, A. (1997). University versus corporate patents: A window on the
basicness of invention. Economics of Innovation and New Technology, 5(1), 19–50.
Trippe, A. J. (2003). Patinformatics: Tasks to tools. World Patent Information, 25(3), 211–221.
Tseng, Y., Lin, C., & Lin, Y. (2007). Text mining techniques for patent analysis. Information Processing &
Management, 43(5), 1216–1247.
USPTO (Ed.). (2007). Manual of patent examining procedure (8th ed.). Alexandria.
van Rijsbergen, C. J. (1981). Information retrieval (2th ed.). London: Butterworth.
von Wartburg, I., Teichert, T., & Rost, K. (2005). Inventive progress measured by multi-stage patent citation
analysis. Research Policy, 34(10), 1591–1607.
Witt, U. (2009). Propositions about novelty. Journal of Economic Behavior & Organization, 70(1–2),
311–320.
Yoon, J., & Kim, K. (2011a). Detecting signals of new technological opportunities using semantic patent
analysis and outlier detection. Scientometrics. doi:10.1007/s11192-011-0543-2.
Yoon, J., & Kim, K. (2011b). Identifying rapidly evolving technological trends for R&D planning using
SAO-based semantic patent networks. Scientometrics, 88(1), 213–228.
Yoon, B., & Park, Y. (2004). A text-mining-based patent network: Analytical tool for high-technology trend.
The Journal of High Technology Management Research, 15(1), 37–50.
Yoon, B., & Park, Y. (2005). A systematic approach for identifying technology opportunities: Keyword-
based morphology analysis. Technological Forecasting and Social Change, 72(2), 145–160.

123