Technology Mapping Using WebAI:

The Case of 3D Printing

Julian Schwierzy1 , Robert Dehghan4,6 , Sebastian Schmidt5,6 , Elisa Rodepeter1 ,
Andreas Stömmer1 , Kaan Uctum1 , Jan Kinne3,6 , David Lenz6,7 ,
Hanna Hottenrott1,2,3

1
Technical University of Munich, 2 Munich Data Science Institute, 3 Leibniz Centre for European Economic
Research (ZEW), 4 University of Mannheim, 5 University of Salzburg, 6 ISTARI.AI, 7 Justus Liebig University
Giessen; B [email protected]
arXiv:2201.01125v1 [econ.GN] 4 Jan 2022

Abstract—The diffusion of new technologies is crucial for the realization of social and economic returns to innovation.
Tracking and mapping technology diffusion is, however, typically limited by the extent to which we can observe
technology adoption. This study uses website texts to train a multilingual language model ensemble to map technology
diffusion for the case of 3D printing. The study identifies relevant actors and their roles in the diffusion process. The
results show that besides manufacturers, service provider, retailers, and information providers play an important role.
The geographic distribution of adoption intensity suggests that regional 3D-printing intensity is driven by experienced
lead users and the presence of technical universities. The overall adoption intensity varies by sector and firm size. These
patterns indicate that the approach of using webAI provides a useful and novel tool for technology mapping which adds
to existing measures based on patents or survey data.

Keywords: 3D printing, web mining, innovation, technology mapping

Data: An interactive visualisation of our data can be found under https://stories.istari.ai/3D/en/.

1 I NTRODUCTION useful, it applies only to patentable inventions,

whereas innovation and diffusion in other areas

N EW technologies are a key driver of sus- may not be captured [11].

tainable economic development and a Especially, for an interdisciplinary technology
crucial contributor to the substantial rise in liv- such as 3D printing, the identification of rel-
ing standards the world has seen since the first evant patents is a challenge, as these can be
industrial revolution [1]. Technological innova- filed in very different patent technology classes
tion is also important for the competitiveness [19]. Moreover, understanding the spread of
of companies and their performance [2], [3], new technologies require information on dif-
[6]. Policy-makers and practitioners are there- ferent actors relevant to diffusion processes. In
fore interested in understanding the emergence addition to the manufacturers and users of the
of new technologies. Yet, it is not only the technology, others also contribute to the dif-
invention that matters, but also the adoption fusion by simply providing information, thus
and diffusion of innovation [10]. Tracking the helping to raise customer awareness and re-
adoption of new technologies is, however, not duce the adoption resistance of potential users
straightforward. Researchers typically rely on and intermediaries. Alternatively to patent in-
information drawn from patent applications to formation, company surveys on new products,
measure, understand, and localize inventive ac- processes, and services are useful for collecting
tivity [18], [19]. While such data is extremely information that augments patent-based mea-
1
sures. Yet, surveys are typically restricted to appear to contribute to a more intense use of 3D
smaller samples and may potentially overlook printing as a new technology, possibly through
significant to the diffusion processes in certain different channels including access to human
areas [21]. In this study, we therefore propose capital, opportunities for applications, and cus-
a novel methodology for mapping technology tomers. These results are in line with earlier
diffusion in areas where patent data or com- research that stressed the importance of the co-
pany surveys may not be sufficient. We follow location of science and industry for spurring
the approach recently developed by Kinne et innovation especially in emerging technologies
al. [16] in the context of analyzing the reac- [5], [12].
tions of German companies to the COVID-19
pandemic. The webAI method builds on ad-
vances in research on natural language pro- 2 BACKGROUND ON 3D P RINTING
cessing to measure company activity (e.g. in- 3D printing, often used synonymously with the
novation) using web mining and deep learning term additive manufacturing, is an umbrella
[16]. In particular, we analyze text informa- term for a group of production technologies1 .
tion retrieved from company websites to map A key characteristic is the use of digital designs
the diffusion of 3D printing technologies while for creating a physical product layer by layer
distinguishing between different contributors. [8]. Compared to conventional manufacturing
For data collection, a scraper screens for topic- technologies such as drilling, milling, and injec-
specific keywords in text paragraphs on com- tion moulding, 3D printing provides substan-
pany websites. By manually labeling a subset tial advantages in terms of sustainability [9],
of the identified text paragraphs, we build a flexibility [7], and the freedom to design com-
training data set based on the context in which plex geometries [8]. Thus, it allows rapid idea
the keywords appear. A model ensemble is then iterations to create new goods [24], decentral-
trained on the labeled data and subsequently ized production [4], and it affects competition
applied to the companies with web domains [22].
in Germany listed in the ORBIS firm database The development of 3D printing technology
(approx. 1.3 million companies). Firms use their started in the 1980s, when the main applica-
website to promote their products and services, tions included the production of prototypes
but also to inform the public about important and spare parts. More recently, 3D printing has
events. Websites can therefore be used to read also become more important for the production
out the activities of companies, which is facili- of end-products [23]. Its increasing relevance is
tated by the fact that, depending on their size also reflected in the continuous growth of the
and industry, almost every relevant company 3D printing market including revenues from
has its own website [15]. 3D printing systems, materials, software, and
The model yields predictions for the degree services. A compound annual growth rate of
of engagement at a highly granular regional roughly 25% since 2014 led to a global market
level as well as a classification for the type size of 3.7 billion EUR in 2014, 8.9 billion EUR
of involvement of all included companies in in 2018, and is predicted to reach approximately
3D printing activities. We differentiate the four 30.2 billion EUR in 2024 [19]. Overall, actors rel-
types: Manufacturers, Service, Retailer and Infor- evant to the adoption and diffusion of 3D print-
mation. Based on this fine-grained data, we can ing technology comprise manufacturers of the
map the diffusion of 3D printing in Germany equipment (machines, materials, and software)
and study the adoption intensity across sectors on the supply side and technology adopters
and regions. We find several adoption hotspots on the demand side. In addition, information
and show that the local adoption intensity
seems to be related to two factors: the close- 1. While additive manufacturing is generally defined as a
process of combining materials to create objects layer by layer,
ness to technical universities and the presence 3D printing is specifically the fabrication of objects using a print
of traditional manufacturing companies. Both head, nozzle or other printer technologies [13].

2
providers such as consulting companies, blogs, ORBIS database. This data set consists of a cor-
and associations contribute to the diffusion of porate URL as well as company characteristics,
knowledge about 3D printing. Despite these e.g. number of employees and incorporation
insights, we still know little about the diffusion date. The definition of types was performed in
of 3D printing technology across sectors and ge- several feedback loops. A first data set was gen-
ographic regions as well as the role of different erated in May 2021 by searching the corporate
actors in this process. websites for the selected keywords. This led to
103,000 data points, each containing the URL
of the website where the keyword was found,
3 M ETHODOLOGY the keyword itself, as well as the paragraph
The main objective of this study is to map the including the respective keyword. If a URL in-
diffusion of 3D printing technology using infor- cluded multiple keywords, multiple data points
mation that is readily available and by follow- were created. In the first step, a data subset of
ing a transparent and reproducible approach. 750 data points was selected randomly and di-
For this purpose, we make use of company vided evenly among five different researchers,
websites and general company-level informa- who then made an initial proposal for the type.
tion obtained from the ORBIS database. The For this first proposal, both the paragraphs and
web analysis of this study is conducted using the overall website itself were analyzed. Since
webAI, a cloud-based and artificial intelligence the resulting classification proposals strongly
web analysis tool set developed by ISTARI.AI. resembled each other, we agreed on seven types
With the help of this tool we aim at identi- in a first feedback round: Manufacturer, Service,
fying and clustering firms according to their 3D Printing for own Products, Consulting & Ed-
role in the innovation and diffusion process of ucation, Retail, Information, and Others. Those
the 3D printing technology. Our webAI based defined types were then used to manually label
approach comprises several steps, which we a training data set. For this purpose, 3,000 data
describe in the following. points were randomly selected from the initial
103,000 data points. Again, the labels were as-
3.1 Identification of Keywords signed on the basis of a keyword, a correspond-
In order to determine 3D printing technol- ing paragraph, and the URL. To avoid any bias
ogy engaged companies, we identify a list of due to prior knowledge about the companies,
keywords related to 3D printing. A uniform this data subset differed from the first data
standardized terminology of 3D printing tech- subset used to define the types. The decisive
nologies does not exist. However, companies factor for the respective label should be the
typically use the technical terms from indus- given paragraph. The labelling of this data set
trial standards, in this case from ASTM 52900 was again carried out by multiple researchers.
and VDI 3405. These lists of terms contain We used the labeled data set to train the model.
the nomenclature of different technologies and In several feedback rounds, the prediction re-
their applications. Due to the constant and sults of the model were assessed using a test
rapid change of 3D printing technology, we data set to gauge the prediction performance.
augment these keywords with terms identified It became clear that certain highly similar cat-
in current research papers and publications by egories and their associated keywords were
a consulting company. A complete list of all difficult to distinguish for the model. This led
keywords used for the search is attached in the to the adjustments of keywords and labels.
appendix (see Table 1). Eleven keywords that were not precise enough
for identifying 3D printing technologies, i.e.
they often appeared in a different context, were
3.2 Identification of Types removed. See Appendix Table A1 for a list of
Our initial data contains 1.3 million economi- initial and final keywords. Data points stem-
cally active companies in Germany listed in the ming from keywords that only appeared in
3
menus, signatures, or similar non-content parts text parts, we further encoded several com-
of websites were also removed from the data pany and website meta-information and con-
set. This resulted in a decrease of the data catenated them into a single high-dimensional
set from 103,000 to 32,000 data points. Further, semantic vector. In our specific case, a 1,920 di-
the models showed a low accuracy in distin- mensional semantic vector represented the rel-
guishing between the labels Service, Consulting evant information about each company.
& Education and 3D Printing for Own Products. Since a company’s website usually con-
Therefore, we decided to combine those labels. sists of several sub-webpages and each sub-
We dropped the label Others due to a low num- webpage is labeled individually, a company can
ber of allocations, resulting in four final types: be labeled with multiple types, e.g. Information
Manufacturer, Service, Retail, and Information. A and Retail. To classify the entire company, we
detailed description of those types can be found introduced a hierarchy system. The final clas-
in Table 1. sification of a company corresponded to the
highest ranking label assigned to that company.
The chosen hierarchy is shown in Table 1. For
3.3 Model Creation example, a company that provides 3D print-
Based on the manually labeled data set, we ing services but also manufactures 3D printers
trained a webAI model to automatically label would be assigned the final label Manufacturer.
the context of 3D printing related web text
content according to the four pre-defined types.
The final classification system was a model 4 R ESULTS
ensemble consisting of 10 single models. Each We present our findings in two steps. First,
of the 10 constituent models received different we describe the characteristics of the firms
data during training, thus learning distinct pat- involved in 3D printing and their respective
terns from the data. Consequently, each model types. Second, we study the geographical dis-
became an expert in differentiating the defined tribution of 3D printing companies and relate
types. However, each model used a slightly the occurrence of 3D printing hotspots to the
different reasoning. As each model cast a vote location of relevant manufacturing clusters and
during inference, the type with the most votes universities.
became the final prediction. Ensemble models
greatly increase the reliability of the results,
as single models can easily over- or under-fit 4.1 Company characteristics
the training data, leading to unwanted edge When differentiating between roles that the
cases and overall sub-par performance. This companies involved in 3D printing take in the
procedure also allows to calculate confidence adoption and diffusion of the technology, we
scores for the predictions, i.e. the more models find that the vast majority of companies (total:
favor the same outcome, the higher the con- 6,336) act as service providers (71.9%). 19.7%
fidence in the decision. We further optimized of companies provide information as the main
each constituent models architecture through activity. Manufacturers comprise 8.0% of com-
an extensive neural architecture search [14]. The panies and retail 0.4%.
constituent models received semantic vectors Dividing the engaged companies into six
as inputs. The semantic vectors resulted from age classes shows that both very young as
encoding the relevant paragraphs using pre- well as relatively old firms are active in the
trained sentence transformers [20]. The advan- area of manufacturing and service provision. In
tage of pre-trained models is that they already retail, mainly medium-old companies are most
possess fundamental language understanding active. Interestingly, information providers are
[17]. Thus, only comparatively little training relatively established in terms of age suggesting
data and minor adaptions are necessary to cope that experience and credibility may matter here
with new use cases. Along with the relevant (see Figure 1).
4
 TABLE 1: Labels for Website Classification
Definition Initial Classification Final Classification
Manufacturer of 3D printers or equipment (e.g. software, material) Manufacturer Manufacturer
Offering personalized 3D printing services Service Service
Using 3D printing technologies in own production 3D Printing for own Products Service
Offering consulting, training etc. for firms to adopt 3D printing Consulting & Education Service
Offering 3D printers, material, spare parts, etc. Retail Retail
Providing information about 3D printing technologies Information Information
Miscellaneous purposes Others −

Fig. 1: Age of 3D Printing Companies Fig. 2: Size of 3D Printing Companies

Distinguishing engaged firms of different optical products (3.2%) as well as in mechan-

sizes, we see that small (10-49 employees) and ical engineering (2.9%) and chemicals (2.6%).
micro firms (1-9 employees) account for the Some sectors are mainly active in 3D printing
highest shares in all four classes. This was service provision such as the pharmaceutical
to be expected, given that the overwhelming industry (100%), repair and installation services
majority of firms fall into this category. The (89.5%), and metal manufacturing (87.9%). This
highest share of large firms (>250 employees) finding is in line with the observation that 3D
is in manufacturing and the largest share of printing technology applications can be found
micro-sized firms is in retail (see Figure 2). in producing spare parts, appliances, medical
Figure 3 provides a detailed insight into devices, and drugs [19]. The largest relative pro-
which industrial sectors 3D printing technology portion of manufacturers can be found in min-
engagement is most prominent in.2 The share of ing (25.0%), textile/clothing (18.8%), and trans-
3D printing engaged firms is highest in mate- portation manufacturing (20.0%). The sectors
rial development such as synthetics (5.4%) and that show high shares of information providers
metal (4.8%). Relatively high shares can also related to 3D printing are the media & pub-
be seen in sectors producing electronics and lishing companies (56.1%) as well as interest
2. See Table A3 in the Appendix for details on the sector
groups (45.7%). This may reflect the value of in-
classification. We summarized sub-sectors into 31 main cate- formation diffusion as a mean to reduce adop-
gories aggregating related activities. For instance, the category tion barriers for end users. The prevalence of
’Electronics / optics’ refers to NACE class C26 and covers man-
ufacturing of electrical equipment including electric motors, 3D printing engagement in the retail category
machinery and equipment, appliances. is highest in textile/clothing (6.2%) and unsur-
5
prisingly in wholesale (1.8%) and retailing itself printers and equipment. Located in the south
(1.7%). and south-east of Munich are several other
high-tech companies including Airbus and In-
fineon Technologies.
4.2 Geographical diffusion Moreover, within the city district of Munich is
Figure 6a shows the location of the identified the headquarter of the car manufacturer BMW.
firms engaged in 3D printing technology using Other big corporations such as Siemens AG
the listed address in ORBIS as the location iden- (among the top 3 applicants for 3D printing
tifier. We see an agglomeration of engaged com- patents) and MTU Aero Engines (among the
panies in larger cities such as Berlin, Munich, top 10 patent applicants) are located in this area
and Hamburg. Yet, we also find relatively high (cf. [19], Table 4.1).
absolute numbers across North-Rhine West- Jena, a comparatively much smaller city than
phalia, particularly in the area around Cologne Munich, is a location where the engagement
and in the south-west of Germany (Baden- might be driven by users from the optical in-
Württemberg). The high overall agglomeration dustry, which has been traditionally very strong
near Cologne is likely based on the diverse in this region.
sectors located in the region ranging from min- Tuttlingen and Pforzheim as other high engage-
ing and consumer products to clothing and de- ment locations, on the other hand, have been
sign. Moreover, the areas around Hanover and traditionally strong in manufacturing sectors
Dresden stand out as having relatively many such as suppliers for the automotive sectors,
engaged companies. We assume that this geo- medical engineering, or the construction of
graphical pattern reflects the overall business parts and materials used in industrial produc-
activities in the vicinity of active locations. For tion including jewellery. A look at the overall
example, the appearance of Bremen and Kiel distribution of manufacturing firms (see Fig-
may reflect the presence of ship manufacturing, ure 6c) suggests that diffusion and adoption
whereas in the south there is a stronger focus seem to be related to the prevalence of tradi-
on automotive manufacturing. tional manufacturing firms. This hypothesis is
Even more informative, however, is the depic- supported by the hotspot Frankenthal (Pfalz),
tion in Figure 6b that shows the number of 3D located close to Ludwigshafen am Rhein, where
active firms relative to the overall number of the chemical corporation BASF, a holder of
firms in the same region. In this visualization, several 3D printing patents [19], has its head-
it becomes more evident that the relative inten- quarters.
sity is highest in the south and south-west. In Another pattern that becomes visible in Fig-
addition, there are adoption hotspots in central ure 6d is the presence of technical universities
Germany and in some areas in the north. The (TU) in and close to highly engaged locations.
red squares indicate the top 10 locations in This is also visible in Figure 4 where the most
terms of relative 3D engagement intensity. prominent downtown hotspot is the location of
The three most engaged regions are Jena the Technical University of Munich (TUM). In
(Thuringia), Tuttlingen (Baden-Württemberg), addition, the Munich University of the Federal
and Starnberg (Bavaria). Six out of the ten Armed Forces (Bundeswehruniversität) is located
biggest clusters are located either in Baden- in the south-east of the city where we also
Württemberg or in Bavaria. Only two of the observe high-intensity zones.
hotspots are not located in southern Germany This pattern is not unique to Munich, but can
(Olpe and Jena). Taking a closer look at the also be seen in Karlsruhe and Aachen, related
area around Munich (see Figure 4), we find to the Karlsruhe Institute of Technology (KIT)
several micro hotspots within and around the and the RWTH Aachen, respectively. Moreover,
city. Starnberg is one of top 10 hotspots and at there are plenty of smaller universities of ap-
the same time the location of the headquarter plied sciences, such as the Coburg University of
of EOS - one of the largest manufacturers of 3D Applied Science and Arts, in less densely popu-
6
 Fig. 3: Sector Distribution of 3D Printing Companies

lated areas that may also contribute to relatively the location choice.
high adoption intensities in their surroundings.
A more differentiated analysis can be made
4.3 Validation
when comparing the four types Manufacturer,
Service, Retailer, and Information. Regions with To check the validity of the webAI type iden-
a higher density of traditional manufacturing tification, we perform a consistency test that
firms show a stronger engagement in manufac- checks the overall innovativeness of the firms
turing (see Figure 7a) and retail (see Figure 7b). by category. If our identification of new technol-
Locations in close proximity to technical uni- ogy adopters and diffusers worked, we would
versities are stronger in 3D printing as a service expect that those firms are also more innova-
(see Figure 7c) and information provision (see tive. We make use of the InnoProb, a measure
Figure 7d). The latter seems to take place in lo- for firms overall innovativeness as developed
cations with a strong manufacturing base such in [16], which assigns an innovation probabil-
as Lichtenfels, where General Electric owns a ity to firms based on a model trained with
majority share in the company Concept Laser Community Innovation Survey data. As can be
and has since expanded its activities and at- seen in Figure 5, the 3D printing technology
tracts suppliers and service providers to the engagement correlates with the degree of in-
region. In the identified top 10 locations, we see novativeness of the respective types. We found
all types being active which suggests that clus- that 57.7% of companies engaged in 3D printing
ters emerge when there is both manufacturing were innovative (InnoProb ≥ 0.4). For manu-
and applications as well as information provi- facturers, as many as 74.9% were innovative
sion. Yet, there are areas with relatively high companies. This suggests that we were cap-
engagement intensity that are rather active in turing firms that have recently introduced a
retail. We find agglomerations north and south service or product innovation. This is true for
of Berlin as well as in the north of Germany all four types, but particularly for information
suggesting that inexpensive storage space and providers and manufacturers. Our results re-
accessibility to export markets may determine garding the use of 3D printing by sector are
in line with the findings presented by Pose-
7
 Fig. 4: 3D Printing Intensity in Munich

Rodriguez et al. [19]. This report is based on

patent application information (as recorded in
the PATSTAT database) and documents trends
in patenting by sector, which also confirms the
prevalence of 3D printing in transport manufac-
turing, industrial tooling, health, optics applica-
tions, and consumer goods. Our mapping also
overlaps geographically with some of the find-
ings in [19] (see for instance Figure 38 in their
paper), but also identified additional regions
not found in patent data. There is also a signifi-
cant overlap in the identified hotspots, as the re-
Fig. 5: Innovation Performance of 3D Printing
port also found Munich, Munich district, Berlin,
Companies
Hamburg and Starnberg. Yet, patent screening
alone may underestimate the role of locations
like Jena or regions that are less manufacturing-
intensive. which are not easily reflected in traditional
indicators such as patents. Moreover, it allows
to differentiate between the type of engagement
5 D ISCUSSION AND C ONCLUSION of companies capturing also activities that are
This study illustrates a website text-based related to the diffusion of new technology
approach to analyze new technology adoption. rather than its invention. Such novel indicators
WebAI utilizes machine learning to identify may be informative for policy makers as well
companies that engage in specific technologies as other stakeholders by providing additional
8
information about individual technologies. advantage of almost universal coverage, there
The mapping for the case of 3D printing may be very small or young companies that do
revealed interesting geographical patterns of not have a website (yet) and are therefore not
technology use and diffusion. Linking adoption included in our analysis [16]. Finally, it should
intensity to the characteristics of the companies be noted that the websites were analyzed at a
shows that young and small firms play an fixed point in time and that statements about
important role. Furthermore, we show that adoption dynamics will require repeating the
regions with a strong general manufacturing analyses in the future.
industry have higher levels of activities.
Interestingly, the great majority of 3D printing-
engaged companies is classified as the type
Service. Locations with technical universities
show higher adoption intensities suggesting
that academic research organizations and
basic research may play an important role for
pushing new technologies and the provision of
skilled employees trained in relevant fields. It
should be noted, however, that the presented
analysis is merely descriptive. Subsequent
analyses may use econometric models to
uncover statistically significant relationships.
Moreover, the analysis of causal relationships
between location factors, such as universities,
and 3D-printing technology adoption would
be interesting and could augment recent
research on that topic [12]. Comparing the
results to analyses based on patent data
could furthermore inform innovation research
and indicator development. Finally, a deeper
investigation of the link between webAI-based
measures and other innovation indicators at the
company-level would be desirable for a better
understanding of the potentially different
dimensions captured by both approaches for
mapping technology diffusion [15].

We encourage more research on the appli-

cation of webAI to other technologies as well
as the validation of our results by other re-
searchers. We made some efforts to validate
our results internally as well as externally, but
we acknowledge that the extent to which our
labels can be generalized may be limited. The
accuracy was significantly influenced by how
consistently the training data set could be la-
beled in the first step, i.e. by how evident the
allocation of each data point to the respective
cluster was for each labelling person. Moreover,
while relying on company websites has the
9
 (a) Absolute Intensity (b) Relative Intensity

(c) Total Manufacturing Industry (d) Technical Universities

Fig. 6: 3D Printing Intensity and Facilitators

10
(a) Manufacturer (b) Service Provider

(c) Retail (d) Information

Fig. 7: Relative 3D Printing Intensity Differentiated by Categories

11
R EFERENCES additive manufacturing: Trends in 3d printing technolo-
gies. 07 2020.
[1] Philippe Aghion, Céline Antonin, and Simon Bunel. The [20] Nils Reimers and Iryna Gurevych. Sentence-BERT: Sen-
Power of Creative Destruction: Economic Upheaval and the tence embeddings using Siamese BERT-networks. In
Wealth of Nations. Harvard University Press, 2021. Proceedings of the 2019 Conference on Empirical Methods
[2] Philippe Aghion, Richard Blundell, Rachel Griffith, Peter in Natural Language Processing and the 9th International
Howitt, and Susanne Prantl. The Effects of Entry on Joint Conference on Natural Language Processing (EMNLP-
Incumbent Innovation and Productivity. The Review of IJCNLP), pages 3982–3992, Hong Kong, China, November
Economics and Statistics, 91(1):20–32, 2009. 2019. Association for Computational Linguistics.
[3] David B. Audretsch and Maryann P. Feldman. R&D [21] Mónica Salazar and Adam Holbrook. A debate on inno-
spillovers and the geography of innovation and produc- vation surveys. Science and Public Policy, 31(4):254–266, 08
tion. The American Economic Review, 86(3):630–640, 1996. 2004.
[4] Avner Ben-Ner and Enno Siemsen. Decentralization and [22] Julian Schwierzy. Digitalisation of production: Industrial
localization of production: The organizational and eco- additive manufacturing and its implications for competi-
nomic consequences of additive manufacturing (3d print- tion and social welfare. Munich Papers in Political Economy,
ing). California Management Review, 59(2):5–23, 2017. 16, 2021.
[5] Wesley M. Cohen, Richard Nelson, and John P. Walsh. [23] Amanda Su and Subhi J Al’Aref. History of 3d printing. In
Links and impacts: The influence of public research on 3D Printing Applications in Cardiovascular Medicine, pages
industrial R&D. Management Science, 48(1):1–23, 2002. 1–10. Elsevier, 2018.
[6] Bruno Crepon, Emmanuel Duguet, and Jacques Mairesse. [24] Christian Weller, Robin Kleer, and Frank T Piller. Eco-
Research, Innovation And Productivity: An Econometric nomic implications of 3d printing: Market structure mod-
Analysis At The Firm Level. Economics of Innovation and els in light of additive manufacturing revisited. Interna-
New Technology, 7(2):115–158, 1998. tional Journal of Production Economics, 164(C):43–56, 2015.
[7] Daniel R. Eyers, Andrew T. Potter, Jonathan Gosling, and
Mohamed M. Naim. The flexibility of industrial additive
manufacturing systems. International Journal of Operations
and Production Management, 38(12):2313–2343, dec 2018.
[8] Andreas Gebhardt and Jan-Steffen Hötter. Additive man-
ufacturing: 3D Printing for Prototyping and Manufacturing.
Carl Hanser Verlag GmbH & Co. KG, Munich, 2016.
[9] Abby Ghobadian, Irene Talavera, Arijit Bhattacharya,
Vikas Kumar, Jose Arturo Garza-Reyes, and Nicholas
O’regan. Examining legitimatisation of additive manufac-
turing in the interplay between innovation, lean manufac-
turing and sustainability. International Journal of Production
Economics, 219:457–468, 2020.
[10] Bronwyn H. Hall. Innovation and Diffusion, chapter 17,
pages 312–326. Oxford University Press, Oxford, UK, 2006.
[11] Bronwyn H. Hall. Patents, innovation, and development.
Max Planck Institute for Innovation & Competition Research
Paper, No. 20-07, 2020.
[12] Naomi Hausman. University Innovation and Local Eco-
nomic Growth. The Review of Economics and Statistics,
pages 1–46, 03 2021.
[13] ISO/ASTM. Additive manufacturing - General principles -
Terminology (ISO/ASTM DIS 52900:2018). Beuth Verlag,
Berlin, 2018.
[14] Haifeng Jin, Qingquan Song, and Xia Hu. Auto-keras:
An efficient neural architecture search system. In Pro-
ceedings of the 25th ACM SIGKDD International Conference
on Knowledge Discovery & Data Mining, pages 1946–1956.
ACM, 2019.
[15] Jan Kinne and Janna Axenbeck. Web mining for innova-
tion ecosystem mapping: a framework and a large-scale
pilot study. Scientometrics, 125, 2020.
[16] Jan Kinne and David Lenz. Predicting innovative firms
using web mining and deep learning. PLoS ONE,
16(4):e0249071, 2021.
[17] Aditya Malte and Pratik Ratadiya. Evolution of transfer
learning in natural language processing. arXiv preprint
arXiv:1910.07370, 2019.
[18] Petra Moster. Patents and innovation: Evidence from
economic history. Journal of Economic Perspectives, 27:23–
44, 2013.
[19] Javier Pose-Rodriguez, Judy Ceulemans, Yann Ménière,
Aliki Nichogiannopoulou, and Ilja Rudyk. Patents and

12
A PPENDIX
TABLE A1: 3D Printing Keywords
English German Source
Additive Manufacturing Additive Fertigung Norm: VDI 3405
Rapid Manufacturing - Norm: VDI 3405
Rapid Prototypingd - Norm: VDI 3405
Rapid Tooling - Norm: VDI 3405
Stereolithography Stereolithografie Norm: VDI 3405
Laser Sintering Laser-Sintern Norm: VDI 3405
Selective Laser Sintering Selektives Laser-Sintern Norm: VDI 3405
Laser Beam Melting Laser-Strahlschmelzen Norm: VDI 3405
Laser Forming Laser Forming Norm: VDI 3405
Selective Laser Melting Selective Laser Melting Norm: VDI 3405
LaserCUSING LaserCUSING Norm: VDI 3405
Direct Metal Laser Sintering Direktes Metall-Laser-Sintern Norm: VDI 3405
Electron Beam Melting Elektronen-Strahlschmelzen Norm: VDI 3405
Fused Layer Manufacturing Fused Layer Manufacturing Norm: VDI 3405
Fused Layer Modelling Fused Layer Modelling Norm: VDI 3405
Fused Deposition Modelling Fused Deposition Modelling Norm: VDI 3405
Filament Deposition Strangablegeverfahren Norm: VDI 3405
Multi-Jet Modelling Multi-Jet Modelling Norm: VDI 3405
Poly-Jet Modelling Poly-Jet Modelling Norm: VDI 3405
3D-Printing 3D-Druck Norm: VDI 3405
Layer Laminated Manufacturing Schicht-Laminat-Verfahren Norm: VDI 3405
Laminated Object Manufacturing Laminated Object Manufacturing Norm: VDI 3405
Digital Light Processing Digital Light Processing Norm: VDI 3405
Thermotransfer Sintering Thermotransfer-Sintern Norm: VDI 3405
Binder Jetting Freistrahl-Bindemittelauftrag Norm: ASTM 52900
Directed Energy Deposition Materialauftrag mit gerichteter Norm: ASTM 52900
Energieeinbringung
Material Extrusion Materialextrusion Norm: ASTM 52900
Material Jetting Freistrahl-Materialauftrag Norm: ASTM 52900
Powder Bed Fusion Pulverbettbasiertes Schmelzen Norm: ASTM 52900
Sheet Lamination Schichtlaminierung Norm: ASTM 52900
Vat Photopolymerization Badbasierte Photopolymerisation Norm: ASTM 52900
Metal Selective Laser Sintering - Consulting: AMPower
Laser Beam Powder Bed Fusion - Consulting: AMPower
Electron Beam Powder Bed Fusion - Consulting: AMPower
Powder Feed Laser Energy Deposition - Consulting: AMPower
Coldspray - Consulting: AMPower
Wire Arc d - Consulting: AMPower
Plasma Arc Energy Deposition - Consulting: AMPower
Wire Feed Laser - Consulting: AMPower
Electron Beam Energy Deposition - Consulting: AMPower
Resistance Welding d - Consulting: AMPower
Liquid Metal Printing - Consulting: AMPower
Ultrasonic Welding d - Consulting: AMPower
Friction Deposition - Consulting: AMPower
Nanoparticle Jetting - Consulting: AMPower
Metal Filament Fused Deposition - Consulting: AMPower
Modeling
Metal Pellet Fused Deposition Modeling - Consulting: AMPower
Metal Lithography - Consulting: AMPower
Powder Metallurgy Jetting - Consulting: AMPower
Mold Slurry Deposition - Consulting: AMPower
Electrographic Sheet Lamination - Consulting: AMPower
Thermal Powder Bed Fusion - Consulting: AMPower
Laser Powder Bed Fusion - Consulting: AMPower
Pellet Based Material Extrusion - Consulting: AMPower
Continuous Fiber Thermoplastic - Consulting: AMPower
Deposition
Continuous Fiber Material Extrusion - Consulting: AMPower
Filament Based Material Extrusion - Consulting: AMPower
Continuous Fiber Sheet Lamination - Consulting: AMPower

13
Area-Wise Vat Polymerization - Consulting: AMPower
Fiber Alignment Area-Wise Vat - Consulting: AMPower
Polymerization
Thermoset Deposition - Consulting: AMPower
Continuous Fiber Thermoset Deposition - Consulting: AMPower
Elastomer Deposition - Consulting: AMPower
Vat Vulcanization - Consulting: AMPower
Laser Chemical Vapor Deposition - ISBN: 978-1-56990-582-1
Continuous Liquid Interface Production - ISBN: 978-1-56990-582-1
Laser Metal Fusion Selektives Laserschmelzen ISBN: 978-1-56990-582-1
Selective Mask Sintering Selektive Maskensintern ISBN: 978-1-56990-582-1
Laser Engineered Net Shaping Laserpulverformung ISBN: 978-1-56990-582-1
ARBURG Kunststoff-Freiformen - ISBN: 978-1-56990-582-1
Maskless Masoscale Material Deposition - ISBN: 978-1-56990-582-1
Foam Reaction Prototyping - ISBN: 978-1-56990-582-1
Tool-Less Fabrication Werkzeuglose Fertigung d DOI: 10.4337/9781781003930.00017
Generative Manufacturing Generative Fertigung DOI: 10.4103/jorr.jorr 9 17
Digital Composite Manufacturing - DOI: 10.3233/AOP-120022
- Laserstrahlschmelzen DOI: 10.25534/tuprints-00014474
Laser Consolidation Laserkonsolidierung DOI: 10.1533/9781845699819.6.492
Ultrasonic Consolidation Ultraschallkonsolidierung DOI: 10.1177/0892705714565705
Freeform Fabrication Freiformherstellung DOI: 10.1109/JPROC.2016.2625103
Layer Manufacturing Schichtbauverfahren DOI: 10.1109/JPROC.2016.2625102
Additive Layer Manufacturing Additive Schichtherstellung DOI: 10.1109/JPROC.2016.2625101
Additive Techniques d Additive Techniken DOI: 10.1109/JPROC.2016.2625100
Additive Process d Additive Prozesse DOI: 10.1109/JPROC.2016.2625099
Additive Fabrication - DOI: 10.1109/JPROC.2016.2625098
Digital Light Synthesis - DOI: 10.1101/2020.05.25.20112201
Continuous Digital Light Manufacturing - DOI: 10.1080/17452759.2012.673152
Two-Photon Polymerization - DOI: 10.1063/1.4916886
Hot Lithography - DOI: 10.1039/D0PY01746A
Direct Shell Production Casting - DOI: 10.1036/007142928X
Thermal Polymerization - DOI: 10.1021/acsapm.8b00244
Directed Light Fabrication - DOI: 10.1016/S0924-0136(97)00321-X
Light Initiated Fabrication Technology - DOI: 10.1016/j.synthmet.2017.11.003
Ultrasonic Additive Manufacturing Ultraschalladditivherstellung DOI: 10.1016/j.scriptamat.2019.10.004
E-Manufacturing d - DOI: 10.1016/j.pnsc.2008.03.027
Programmable Photopolymerization - DOI: 10.1016/j.matdes.2020.109381
Bioprinting - DOI: 10.1016/j.ajps.2019.11.003
Fused Filament Fabrication Schmelzschichtung DOI: 10.1016/j.addma.2019.100861
Masked Stereolithography - DOI: 10.1016/j.addma.2014.08.003
Laser Metal Deposition Laserauftragsschmelzen DOI:
10.1016/B978-0-12-818411-0.00012-4
Lithography-Based Ceramic - DOI: 10.1007/s11665-017-2843-z
Manufacturing
- Robocasting DOI: 10.1007/s00170-015-7576-2
Direct Metal Deposition - DOI: 10.1007/s00170-015-7576-2
Beam Interference Solidification - DOI: 10.1007/s00170-015-7576-2
Ballistic Particle Manufacturing - DOI: 10.1007/s00170-015-7576-2
Holographic Interference Solidification - DOI: 10.1007/s00170-015-7576-2
Solid Ground Curing - DOI: 10.1007/s00170-015-7576-2
Solid Foil Polymerisation - DOI: 10.1007/s00170-015-7576-2
Film Transfer Imaging - DOI: 10.1007/978-3-662-49866-8
Cold Spraying d Kaltgasspritzen d DOI: 10.1007/978-3-319-16772-5 12
Selective Deposition Lamination - DOI: 10.1007/978-3-030-58960-8
Digital Manufacturing d Digitale Fertigung d DOI: 10.1007/978-1-4939-2113-3
- Direct Ink Writing DOI: 10.1002/adfm.200600434
Stereolithography Apparatus - DOI: 10.1.1.564.3157
d Keyword was eliminated

14
 TABLE A2: Full names of the technical universities
Abbreviation Full Name
BTU Brandenburgische Technische Universität
Cottbus-Senftenberg
HCU HafenCity Universität Hamburg
KIT Karlsruher Institut für Technologie
LUH Leibniz Universität Hannover
OVGU Otto-von-Guericke-Universität Magdeburg
RWTH Aachen Rheinisch-Westfälische Technische Hochschule Aachen
TU Berlin Technische Universität Berlin
TU Braunschweig Technische Universität Braunschweig
TU Chemnitz Technische Universität Chemnitz
TU Darmstadt Technische Universität Darmstadt
TU Dortmundt Technische Universität Dortmund
TU Dresden Technische Universität Dresden
TU Freiberg Technische Universität Bergakademie Freiberg
TU Ilmenau Technische Universität Ilmenau
TUC Technische Universität Clausthal
TUHH Technische Universität Hamburg
TUK Technische Universität Kaiserslautern
TUM Technische Universität München
U Stuttgart Universität Stuttgart

15
 TABLE A3: Industry classification
Industry Name NACE range Description
Mining B Mining and quarrying
Textiles/clothing C13-C15 Manufacturing of textiles, wearing apparel, fur, carpets, leather
Manufacturing of wood and of products of wood and cork, except furniture; manufacture of articles of straw
Wood/paper/print C16-C18
and plaiting materials
Chemicals C20 Manufacturing of chemicals and chemical products
Pharmaceuticals C21 Manufacturing of basic pharmaceutical products and pharmaceutical preparations
Synthetics C22 Manufacturing of rubber and plastic products, packaging
Manufacturing of glass and glass products, refractory products, clay building materials, tiles and flags, porcelain,
Glass/ceramics C23
cement, concrete
Metal C24 Manufacturing of basic metals (iron, steel, ferro, lead, zinc and tin)
Metalware C25 Manufacturig of fabricated metal products (incl. cuttlery, tools), except machinery and equipment
Electronics/optics C26 Manufacturing of computer, electronic, optical products, and peripheral equipment
Electrical engineering C27 Manufacturing of electrical equipment (incl. electric motors, machinery & equipment, appliances)
Mechanical engineering C28 Manufacturing of machinery and equipment (incl. tools)
Automotive manufacturing C29 Manufacturing of motor vehicles, trailers and transport equipment
Transportation manufacturing C30 Manufacturing of boats, trains, aircraft, spacecraft
Other manufacturing C31-C32 Manufacturing of furniture, coins, games & toys, jewellery, medical and dental instruments and supplies
Repair/installation C33 Repair and installation of machinery and equipment
Construction F Construction of buildings, roads, railways, bridges, civil engineering, demolition, drilling, roofing, other installations
Wholesale on a fee or contract basis (including raw materials, live animals, textile raw materials, household goods,
Wholesale G46
pharmaceuticals, equipment and supplies)

16
Retail G47 Retail trade of food, fuel, ICT, electronics, cultural goods, media, toys, flowers, plants
Publishing of books, periodicals and other publishing activities, motion picture, video and
Media/publishing J58-J60
television programme production, sound recording and music publishing activities, radio broadcasting
Telecommunications, computer programming, information service activities, data processing, hosting and
ICT services J61-J63
related activities; web portals
Finance/insurance K Financial and insurance activities, monetary intermediation, activities of holding companies,
Management services M69-M70 Legal and accounting activities, including accounting, bookkeeping and auditing activities; tax consultancy
Engineering/scientific services M71-M72 Architectural and engineering activities; technical testing and analysis, scientific R&D
Creative Services M73 Advertising and market research, media representation, public opinion polling
Other professional, scientific and technical activities; Specialised design activities; Veterinary activities; Administrative
Other services M74-N82
and support service activities; Rental and leasing activities; Activities of employment placement agencies
Public administration O Public administration and defence; compulsory social security; foreign affairs, defence, fire service
Education P Education, driving schools, higher education, educational support activities
Interest groups S94 Activities of membership organisations, business associations, trade unions, political and religious organisations
Personal services S95-S96 and T Repair of computers and personal and household goods, repair of computers and communication equipment
Source: https://nacev2.com/en