Digital Transformation of the Automotive Industry: An Integrating Framework to Analyse Technological Novelty and Breadth

ABSTRACT

Research demonstrates that digital technologies stimulate industrial transformation by enabling new interdependencies with firms outside and across firm and industry boundaries. However, we know little about the degree of novelty and breadth of digital technologies that have the potential to transform industries. Understanding the degree of novelty (spanning from radical to incremental) and breadth (spanning from one sector to multiple sectors) of digital technologies is important for measuring their impact on industrial transformation. Through a topic modelling research approach on autonomous vehicle technology patents from firms operating in Sweden and a confirmatory survey with the inventors of those patents, this paper reveals 26 digital technology topics that are transforming the automotive industry. The digital technology topics are distributed across four ideal-typical technology categories for transformation: augmenting, spanning, transforming, and disrupting. This study illustrates the value of studying digital technologies’ transformative nature using an integrating framework; it reveals that digital technologies in the automotive industry have mainly incremental characteristics but that these characteristics provide necessary preconditions for the few more radical technologies to achieve their potential in transforming the industry.

KEYWORDS:

• Digital Technologies
• Industry digitalization
• Topic modelling
• Emerging technologies
• Artificial intelligence

JEL:

• D83
• O14
• O32
• O33
• R4

1. Introduction

Numerous scholars argue that the emergence of digital and artificial intelligence (AI) technologies and new abilities associated with implementing them are vital for firms’ activities in the digital age (Iansiti and Lakhani 2020; Menz et al. 2021; Dąbrowska et al. 2022; Bailey et al. 2022). However, the ‘novelty’ and ‘breadth’ of such technologies remain less explored. Debates are emerging around whether the novelty of digital technologies is more likely to be ‘radical’ by stimulating the creation of new market opportunities by dramatically changing the direction of industries (Nambisan, Wright, and Feldman 2019; Yoo et al. 2012) or ‘incremental’ by augmenting or automatising the capabilities of technologies (Lebovitz, Lifshitz-Assaf, and Levina 2022; Raisch and Krakowski 2021). Scholars also discuss whether the breadth of digital technologies is more likely to be for one sector or to span multiple sectors, arguing for potential complementarity to create modifications across several industries and products (Teece 2018). One example of digital technologies with a wide array of applications is movement recognition technologies, which can be applied in autonomous vehicles, allow robots to work alongside people, and improve wellbeing at home.

Although there are several well-documented examples of digital technologies transforming industries, including the health (Lebovitz, Lifshitz-Assaf, and Levina 2022; Coccia 2020), automotive (Svahn, Mathiassen, and Lindgren 2017), media (Anand 2016), and cosmetic (Lopez-Vega and Lakemond 2022) industries, researchers primarily explain the characteristics of single emerging digital technologies and do not study the impact of a whole range of digital technologies, which is required to transform an entire sector or industry (Röder, Both, and Hinneburg 2015; Kwon et al. 2019). Indeed, Bodrožić and Adler (2022, p. 105) highlight that ‘digital transformation is not simply a basket of individual innovations: it is an interdependent cluster of revolutionary technologies, where developments in each technology affect many others’. Hence, the impact of digital technologies’ novelty and breadth on industrial transformation cannot be accounted for by the characteristics of a single digital technology. Instead, the outcome in terms of an industry digitalisation must be studied in relation to a larger cluster of technologies working with each other (Arthur 2009; Coccia 2017; Bodrožić and Adler 2022; Bailey et al. 2022; Röder, Both, and Hinneburg 2015). The framework presented in this paper aims to address this limitation.

In this paper, we take a novel approach to study autonomous vehicle technologies by closely investigating their technological breadth (spanning from one sector to multiple sectors) and technological novelty (spanning from incremental to radical) to unravel the interplay of digital technologies across different digital technology categories and their impact on digital transformation in the automotive industry. Specifically, this paper addresses the following research question: What characteristics of digital technologies can be depicted across the novelty and breadth dimensions to explain their impact on industry digitalization?

Empirically, this paper presents an analysis of interdependent digital technologies captured as digital technology groups. We refer to these groups as digital technology topics, and we study the characteristics and impact of these topics in relation to technological breadth and novelty. In an aggregated patent-level analysis, we employed a computational research method for natural language processing (i.e. latent Dirichlet allocation [LDA]) (Blei, Ng, and Jordan 2003). Specifically, we employed topic modelling (Hannigan et al. 2019) to obtain emerging semantic topics from patents linked to autonomous vehicle technologies in Sweden. The dataset includes 455 patents, which reveal the emerging digital technology groups that automotive companies in Sweden developed between 1970 and 2019. This method allowed us to identify the probability that companies will create knowledge related to 26 unique digital technology topics. At an aggregated level, these 26 topics represent the most relevant digital technology areas for autonomous vehicle technologies. Each topic captures a set of words that recurs across all the patents. The identified topics provide an inductive semantic analysis of all patents included in the dataset. Furthermore, we conducted an expert survey to understand the novelty of the technology topics and the breadth of their application.

Based on our findings, this paper offers two main contributions. First, this paper contributes to empirically untangling the characteristics of digital technology topics of varying technological novelty and breadth and thereby explains the pervasive characteristics associated with digital technologies in the automotive industry. Our work responds to the call to integrate solid findings about technological change and industry digitalisation (Hanelt et al. 2021; Warner and Wäger 2019; Bodrožić and Adler 2022; Coccia and Watts 2020). Particularly, our work connects to and extends the notion of creative accumulation (Bergek et al. 2013) by arguing that digital transformation is the outcome of introducing new creative processes to embed digital capabilities into objects as well as building new digital capabilities outside the main industry.

Second, this paper proposes an integrating framework to clarify the breadth and novelty of digital technologies and their impact on industrial transformation. The framework presents four digital technology categories: augmenting, spanning, transforming, and disrupting. The digital technologies in these four categories are not mutually exclusive but complementary and, in some instances, partially overlapping. By analytically distinguishing the novelty and breadth dimensions, we highlight the characteristics of and impact on industrial transformation in the four categories and thereby contribute to better understanding of emerging digital technologies (Daniele et al. 2015). The integrating framework also provides alternatives for firms to consider development and exploration in new digital technology categories.

Contrary to recent claims suggesting that digitalisation will transform or disrupt established firms (Anand 2016), our findings explain that digital technologies renew and extend the innovation capabilities across industries. However, the incremental renewal and improvement of the focal industry and associated technologies may provide crucial preconditions for realising the potential of radical changes. An illustrative example explored in this paper is collision prevention assistance systems, a radically new technology whose successful implementation also requires incremental improvements in other systems, such as navigation, speed control, and road friction analysis systems.

This paper is structured as follows: Section 2 discusses recent research on digital transformation and outlines a framework for addressing digital transformation based on the combined analysis of the novelty and breadth of digital technologies. Section 3 presents our computer-based text-analysis approach and methodological considerations. Section 4 presents our results. Section 5 discusses our integrating framework as well as our conclusions, study limitations, and avenues for further research.

2. Background literature and framework

This section begins with an overview of the recent literature on digital transformation and digital technologies. Section 2.2 integrates the two central dimensions of digital technologies: technological breadth and technological novelty. Finally, Section 2.3 describes four proposed categories of digital technologies based on the combination of these dimensions.

2.1. Digital transformation

The rise of digital technologies is triggering the transformation of several industries in which firms’ digital transformation activities are contingent on both the institutional conditions and firms’ innovation strategies (Vial 2019; Bodrožić and Adler 2022). So far, scholars have outlined distinct frameworks to design new digital innovation strategies and build digital innovation capabilities and forms of production for digital technologies (Bailey et al. 2022; Dąbrowska et al. 2022). They have also stressed the need to further identify new ways to comprehend the dimensions and characteristics of industry digitalisation (Menz et al. 2021; Nambisan et al. 2017; Bodrožić and Adler 2022). For example, it is necessary to understand how and what types of digital technologies enable industrial transformation that results in a fundamental renewal of firms’ established technology portfolios and innovation capabilities (Hanelt et al. 2021; McAfee and Brynjolfsson 2017; Utterback, Pistorius, and Yilmaz 2018).

Although extant research focuses on the application and benefits of digital technologies for business and innovation platforms (Anand 2016; Gawer 2021) as well as their benefits for labour productivity and sales (Cirillo et al. 2022), research partially disregards the challenges faced by firms in manufacturing industries and complex product industries that require the integration of digital capabilities in component- and system-related knowledge. In these industries, industrial transformation embodies innovation outside the range of expertise of the focal firm or industry and requires firms to undertake rapid technology development based on existing practices while pursuing distinct technology categories (Bergek et al. 2013).

Studying digital technologies at the industry level remains a large challenge because it requires new methods and theoretical lenses to explore the overarching categories and characteristics associated with these technologies. Indeed, only an aggregated meso-level perspective of digital and AI technologies and their characteristics can lead to a full explanation of industry digitalisation (Arthur 2009; Bailey et al. 2022). On this basis, digital transformation across different industries might depend on the aggregate and overlapping characteristics of individual digital technologies, which in turn form unique building blocks. The need for such meso-level studies may partially be due to the fact that ‘neither firms nor innovations, taken individually, can completely explain technology change. Instead, they must be viewed as parts of a larger system; [in which] various agents interact with each other, and institutions matter’ (Carlsson and Stankiewics 1991, p. 23). Bailey et al. (2022) also explain that studying the relationships across technologies and other industrial actors, such as users, equipment manufacturers and vendors, data analytic firms, governments, technology companies, etc., can help uncover the fundamental digital transformation of industries. Moreover, as early-stage digital technologies like the Internet of Things, AI, blockchain, and Metaverse have already matured and are ready for wide deployment, it is less clear how these technologies will be implemented across industries and appeal to a wider set of industries and activities (Bodrožić and Adler 2022; Daniele et al. 2015).

Although some studies primarily explain the application of digital technologies in single industries, such as the health (Coccia 2020; Lebovitz, Lifshitz-Assaf, and Levina 2022), media (Anand 2016), and automotive industries (Svahn, Mathiassen, and Lindgren 2017), less attention has been devoted to the potential impact of such application across industries. Indeed, scholars lament this lack of attention paid to the benefits of the cross-sectorial application of digital technologies (Martynov 2020) as well as the mechanisms firms use to cross knowledge boundaries (Dougherty and Danielle 2012). Particularly for firms in manufacturing industries (Svahn, Mathiassen, and Lindgren 2017), digital technologies require both building capabilities in new technology areas that are outside existing knowledge bases and evolving existing technologies (Hanelt et al. 2021). Understanding the characteristics of digital technologies paves the way to study their aggregated effect on digital transformation as similar technologies can concurrently be developed in different industries and by different firms (Nambisan et al. 2017; Yoo et al. 2012).

2.2. Dimensions of digital technologies

2.2.1. Technological novelty

Scholars have recently begun to claim that the rise of digital and AI technologies, such as algorithms, sensors, and robots, has accelerated the transformation of industries and triggered a new industrial paradigm (c.f. Yoo et al. 2012; Bailey et al. 2022; Iansiti and Lakhani 2020; McAfee and Brynjolfsson 2017). However, not all digital technologies are uniformly disruptive across industries. For example, in the service and entertainment industries, digital technologies have had a profound influence on the associated industry structures and capabilities and have enabled the entrance of new players (Anand 2016). An example is Netflix, which changed from an online DVD-rental store to a film production company. In other sectors, particularly mature industries, such as the manufacturing and automotive industries, digital technologies can either be characterised by their radical changes or only represent an incremental product change (Benner and Tripsas 2012; Nambisan, Wright, and Feldman 2019).

Two streams of literature highlight the creation of radical technologies for technological change. One stream suggests it is large corporations with established capabilities that can launch new radical technologies, while the other stream points out the relevance of new industry entrants for radical technology development (Bergek et al. 2013). The first perspective primarily focuses on incremental technology evolution and learning from advances in technology – from the development of dominant design technologies to ‘incremental technology improvements which ultimately tap the full potential of the initial invention’ (Brynjolfsson and McAfee 2014, p. 77). For example, Dougherty and Danielle (2012) suggest digital technologies complement technology progress (e.g. drug discovery in the pharmaceutical industry), and Coccia (2020) highlights how artificial intelligence can be used for cancer detection. The second perspective proposes that radical innovations can be initiated by new industry entrants that tend to give rise to technological change and disrupt firms pursuing only incremental technological change (Benner and Tripsas 2012; Christensen et al. 2018). Although these research insights offer possible explanations for when technologies may have the potential to be radical or incremental within and across sectors and product markets (Malerba 1992), they have not been applied to settings with unspecified technology characteristics, such as digital technologies.

The literature also remains unclear on how firms develop digital technologies that could hinder or enhance established industry structures characterised by incremental changes and dominant technological architectures (Eggers and Francis Park 2018; Bodrožić and Adler 2022). One research stream suggests that technological change requires industry incumbents to pursue new business models in tandem with their old business models to maintain their market hegemony in the long term (Eklund and Kapoor 2019; Kapoor and Klueter 2021). Other streams of research focus on the antecedents (e.g. firm size and experience, complementary assets, managerial cognition, and top management characteristics) of technological change for incumbent firms (Eggers and Francis Park 2018; Bergek et al. 2013; Coccia and Watts 2020). According to this latter perspective, firms apparently compete by launching either radical technologies aiming to disrupt the market or incremental technologies seeking to extend the lifecycle of products (Utterback, Pistorius, and Yilmaz 2018; Bergek et al. 2013). Eggers and Francis Park (2018), p. 382) suggest that ‘future research needs to identify the nature of technological change and uncover the degree of radicalness of the innovation’ in firms pursuing digital technological change.

Right now, a point of contention is the novelty of digital technologies (Raisch and Krakowski 2021). Some researchers conceptualise digital transformation as ‘a process that aims to improve an entity [firm or technologies] by triggering significant changes to its properties through combinations of information, communication, and connectivity technologies’ (Vial 2019, p. 121) where the focus is on incorporating ‘digital capabilities into objects that previously had a pure physical materiality’ (Yoo et al. 2012, p. 1398). Others outline the need to study the novelty of digital technologies to transform industries’ established capabilities and knowledge bases (McAfee and Brynjolfsson 2017; Nambisan, Wright, and Feldman 2019) as well as the scope of digital technologies to generate a multi-industry impact for industrial transformation. The latter perspective also stresses that firms seek to build radical technologies to reinforce their existing technology portfolios and benefit from arising market opportunities as well as to cope with changes in the technological landscape and industry structure (Coccia 2017, 2015; Kwon et al. 2019).

2.2.2. Technological breadth

Research underlines the effect of digital technologies on industry structure and firm dynamics as well as firms’ efforts to experiment and spread the benefits of digital technologies across different markets and technological applications with the ultimate goal of defining new trajectories of technical progress (Benner and Tripsas 2012). In their pursuit to develop new digital technologies, firms build technologies across one-sector or multi-sector technology areas to position their innovation portfolios in relation to the breadth of industry technologies (Kneeland, Schilling, and Barak 2020; Lopez-Vega, Tell, and Vanhaverbeke 2016). Hence, because technological expertise is expanding and scattered among firms in different industries (Kwon et al. 2019), researchers have tried to provide an aggregated industry framework for digital technologies.

Recent studies provide insights into the pervasive characteristics of digital technologies, such as their openness, convergence, and generativity (Nambisan, Wright, and Feldman 2019), and highlight their potential complementarity to create new resource combinations and modifications across multiple industries and products (Teece 2018). For example, these characteristics are frequently associated with platform technologies, distributed innovation, and combinatorial innovation, which affect the patterns of technology development (Nambisan et al. 2017; Yoo et al. 2012). In particular, the system integration/generativity perspective (c.f. Thomas and Tee 2022) explains how firms use boundary mechanisms to shape the nature and extent of digital technologies by recombining them across multiple technology domains (Nambisan, Wright, and Feldman 2019; Brynjolfsson and McAfee 2014). For example, Lee and Berente (2012) explain that due to the increasing complexity of products and the need to integrate and coordinate work on digital technologies between original equipment manufacturers and suppliers, mature ‘digital control system’ technologies with generative capabilities have been developed to enable the transformation of multiple industries, including the manufacturing, automotive, and telecommunications industries, which can gain multiple benefits from a single digital technology.

Also, some studies highlight that digital technologies, such as 3D printers, sensors, and cameras, could have a broad range of applications in areas like robotics and computer vision (Yoo et al. 2012). In the automotive industry, Svahn et al. (2017) use the example of Volvo Cars’ Connected Car initiative (i.e. in-car entertainment system) to explain four competitive concerns – innovation capability, innovation focus, innovation collaboration, and innovation governance – that enhance the generative capabilities of digital technologies. This study of Volvo presents both tensions during the development of digital technologies and mechanisms to cope with potential digitalisation pitfalls. Despite the relevance of this particular digital technology, its potential for multi-sector application remains limited to the vehicle industry, and other potential digital technologies are not considered (e.g. speed control systems, cameras, sensors). Combining the technological novelty and breadth dimensions, the following section proposes a framework to understand different technology categories

2.3. An integrating framework for industry digitalization

The integrated study of technological breadth and novelty provides four ideal-typical categories of digital technologies’ impact on industry digitalisation: (1) augmenting, (2) spanning, (3) transforming and (4) disrupting (see Figure 1). The purpose of the proposed typology and the respective categories is to disentangle the main characteristics to which those digital technologies contribute. By highlighting the main objectives and characteristics of the respective categories, we arrive at more nuanced insights into their impact on transformation than the well-established one-dimensional distinctions between incremental vs. radical and one- sector vs. multi-sector. In turn, these insights help advance knowledge about digital technologies as drivers of innovation in different applications.

Figure 1. Digital Technology paths.

First, incremental change and upgrading by implementing new improved technologies within one sector characterise one of the most commonly recognised technology development categories – referred to as the augmenting category in this framework. This category represents what could be described as the digitalisation of previously non-digital technologies (i.e. physical products or analyses) (Raisch and Krakowski 2021; Lebovitz, Lifshitz-Assaf, and Levina 2022). Historically notable examples include the introduction of digital technologies in applications in the film, camera, and broadcasting industries. At first, this change could be described as the automatisation or augmentation of processes with the aim to improve quality and efficiency. However, gradually, this upgrading leads the focal industry to a tipping point at which those actors who have not embraced the new technology cannot survive in the new competitive landscape arising from the digital path becoming entirely dominant (Benner and Tripsas 2012). While we describe the technology breadth of this augmenting path as involving one sector, it should be clarified that the same or similar technologies – here referred to as platform technologies (Gawer 2021) – may develop several one-sector applications (e.g. platform applications).

Second, incremental change and upgrading by implementing improved technologies within several sectors characterise the category referred to as the spanning category in our framework. This ideal-typical category may be hard to distinguish from the augmenting category in empirical analyses in the sense that it does not represent a radical leap in technological novelty but rather gradual improvements in what already exists through the infusion of new digital technologies (Lopez-Vega, Tell, and Vanhaverbeke 2016). What mainly distinguishes this category from the augmenting category is that the new digital technologies are initially developed and implemented in other sectors and then spill over to new sectors, which, by definition, implies a wider scope of applicability than the technologies characterising the augmenting category (Thomas and Tee 2022). A wide range of information technology-based innovations fall into this ideal-typical category (e.g. movement sensors, vision technologies). As opposed to the augmenting category, which mainly entails gradual improvements in already existing applications in well-defined sectors, until the tipping point referred to above arises, the spanning category leads to the creation of new sectors with new markets and new types of factors, such as (de)centralised platforms and ecosystems that did not previously have a role in the respective system (Gawer 2021).

Third, the transforming category refers to the development and application of radically new technologies in single sectors or within established technology paradigms, thereby transforming those sectors in a more fundamental manner than the augmenting category. Historically notable examples include the development and application of recombinant DNA, which radically changed parts of the pharmaceutical industry by opening a new path for implementing biotechnology-based drugs. While not necessarily revolutionary beyond the actual application, the development of this technology made it possible to address health challenges previously not within reach. Other examples are the development of biobased energy and speciality chemicals, which made those respective sectors partly abandon the temperature and pressure logic in favour of biological processes while maintaining largely the same range of products and applications and opening up fossil-free raw materials and cleaner production. In contrast to the augmenting category, in which such a shift happens through a gradual process leading up to a tipping point, the shift in this category appears at the very outset. This means that two distinct development logics exists in parallel without one necessarily becoming dominant over time.

Finally, radically new technologies with multi-sector application seem to be very rare and sometimes breakthroughs. In this ideal-typical framework, we refer to this type as the disrupting category. It resembles characteristics of the transforming category with regards to the degree of novelty and thereby impact in terms of change potential. However, the multi-sector scope of these radical technological advancements also implies significant differences from the transforming category. The disrupting category also resembles characteristics of the spanning category, particularly in regards to the simultaneous applicability in a variety of sectors. However, the combination of radical novelty and multi-sector scope makes this category unique. Historically notable examples of advancements that fall into this category include steam and combustion engines, semiconductors, and platform technologies based on bio- and nano- science. Notably, most other technologies sometimes referred to as disruptive due to their revolutionary impact fall into the transforming category according to our framework. It is the disruptive technologies’ ability to dramatically change the directionality of a multitude of sectors that makes them revolutionary, not their degree of novelty.

In the following, we present an analysis of change trajectories in the automotive industry in Sweden. The purpose of the analysis is to apply the framework presented above in an empirical field with the aim to further decompose the respective digital technology categories, thereby advancing our understanding of the contribution of individual firms and their technological advancements to industry digitalisation.

3. Research design and method

3.1. Sample selection and research setting

Our research process started with identifying digital innovations across multiple industries in Sweden. First, we accessed the SWINNO database¹, which provides an overview of firms’ innovations in Sweden from 1971 to 2017 (Sjöö et al. 2014). Although this dataset allows researchers to identify the market sector, collaborating actors, technological complexity, and novelty of the technology for the firm and market for multiple companies, we were mostly interested in the name and description of digital innovations. After reading the description of the innovations from the SWINNO database which is written in Swedish and collected from national magazines (i.e. Ny Teknik, Verkstäderna, Elektroniktidningen), we categorised all 4,507 innovations and identified 1,205 digital innovations for the 1971 to 2017 period. From the 1,205 digital innovations, we inductively distinguished six areas: (1) robotics, (2) cloud- based and connectivity, (3) Internet-of-things, (4) computers, (5) electrification and (6) autonomous vehicles.

We further categorised these innovations across the following industries: automotive, manufacturing, telecommunications, transportation and logistics, agriculture, forest, steel and metal, textile and confection, minerals, chemicals, plastics, food industry, and pharmaceuticals. These actions allowed us to identify numerous digital innovations from the automotive industry (e.g. vehicle connectivity, vehicle-to-vehicle infrastructure) as well as from other industries connected to the automotive industry (e.g. computer vision, machine learning, robotics). We also observed digital innovations in the telecommunications and manufacturing industries that have the potential to be used into the automotive industry, and vice-versa. This analysis provided an initial set of keywords to understand the major changes in the automotive industry in Sweden such as vehicle connectivity, driverless vehicles, and computer vehicles. See Table 1 for more details.

Table 1. Summary of Digital Technologies Reviewed from the SWINNO Database.

No. of inventions	Description	Examples of identified keywords
89	Robotic innovations, new and improved robots that enable automation in, for example, manufacturing	Robotics Automation
173	Cloud-based and communicative innovations that enable connectivity	Smart manufacturing Computer vision
168	Sensors, detectors, antennas, and cameras that enable automation; technology that might be used in infrared rays, and other lasers, etc.	Automation Internet of Things
529	All kinds of hardware and software development within the fields of personal computers, WIFI, operating systems, processors, etc., that may enable digitalisation	Digital manufacturing Smart manufacturing Internet of Things Machine learning
200	Electrification, computerisation, and automation of equipment, machines, and apparatuses	Automation Digital manufacturing Smart manufacturing
46	Self-driving, unmanned vehicles; self- functioning systems that increase safety and digitalisation.	Vehicle connectivity Automation and connectivity of cars and vehicles Self-driving and driverless vehicles Vehicle-to-infrastructure and vehicle-to- vehicle connection Computer vehicles Cloud-based services and safety Infrastructure for automated vehicles

Next, we reviewed publicly available consultancy reports and industry magazines to obtain a thorough understanding of the development and complexity of digital technologies and digital transformation in the automotive industry globally. We identified four industry trends: (1) autonomous vehicles, (2) electrification, (3) car sharing and connected vehicles, and (4) power plans. Finally, we concluded that autonomous vehicle technologies (see Table 2) represent the technologies with the largest number of inventions, disruption potential, and dynamism in Sweden, which is why we selected this area to focus on.

Table 2. Main Trends in Autonomous Vehicle Technologies.

Trends	Technologies	Keywords
Automated vehicle platforms	Perception, analysis, and decision making	Short-, medium-, long-range radar for adaptive cruise control Cameras for lane departure warning/control, traffic sign recognition, surrounding view Navigation and mapping systems Adaptive cruise control and platooning Scene perception and modelling Vehicle stability, dynamic chassis control, conjoint control of stability systems
	Computing	Bus systems Supervisory systems for fault recognition and recovery Artificial intelligence Computer security Diagnostics and fault management
	Vehicle handling	Automatic steering Vehicle suspension control Control systems for road vehicle drive control Powertrain: battery electric vehicles, hybrid vehicles, efficient internal combustion engine vehicles
Smart environments	Communication	5 G network MM wave antenna array technology Cloud for learning and updating high-definition maps and traffic data Cellular communication systems for vehicle applications Traffic signal arrangements Road-embedded sensors and signalling Connection management for emergency connections
	Smart logistics	Fleet management Traffic control systems for road vehicles Automated parking Inductive on-road recharging systems Smart grids in transport

The findings from our desktop research and the SWINNO database resulted in a set of keywords that would enable us to further search and identify digital technologies at the European Patent Office. In total, we identified 1,452 autonomous vehicle patents in Sweden. After reading the abstracts of all the patents, we eliminated patents that did not relate to autonomous vehicle technologies, duplicates (patents that are registered in several countries with different codes), and patents missing a detailed invention description (i.e. only the abstract was available). To study a homogeneous group of digital technologies (i.e. same institutional setting, requirements), we only included patents with at least one inventor working in Sweden for a Swedish or foreign company. This resulted in a total set of 455 patents for further analysis. Patent data were used to identify technology domains, company names, inventor names, (co-) citations, and – most importantly – the patent description.

3.2. Analytical approach

The data analysis for this research comprised using a computational research method – topic modelling (Hannigan et al. 2019) – and a survey to determine the technological novelty and breadth of the identified digital technologies.

3.2.1. Data analysis and content coding with topic modelling

Recently, the social sciences have benefited from advancements in data science methods (Salganik 2018; Grimmer, Roberts, and Stewart 2022), which, through the use of a large data scope and data granularity, have offered management scholars new ways to find answers to ‘new questions’ (George et al. 2016; Antons et al. 2020; Coccia 2020). Specifically, machine learning is employed for supervised learning and unsupervised learning. Unsupervised learning is used for inductive analysis when data categories are unknown. In this study, we used unsupervised machine learning to analyse the text of the 455 autonomous vehicle patents. To analyse the data collected, we opted for a topic-modelling procedure that has captured the interest of management scholars because it allows text to be analysed as data (Antons et al. 2020; George et al. 2016). Compared to other established qualitative methods, this method allows researchers to use machine learning to identify themes that are latent across a large set of documents and to identify the composition themes in each document (Hannigan et al. 2019).

Topic modelling was carried out in two steps: first, we created the text corpus based on the text files of the selected patents, and second, we generated the topic model and derived analytical results from it. These two main steps have several sub-steps that had to be performed to complete the topic modelling, resulting in meaningful topics and thus a good basis for subsequent analysis (see Appendix 1 for the topic model phases). All steps for creating the corpus and building the topic model, including some parts of the analysis, were performed using Python 3.7 and pre-developed public libraries, such as nltk and genism, which are commonly used to perform text-analysis tasks and are described in detail in the respective steps. For a detailed explanation about the dataframe, data corpus, and topic modelling, read Appendix 2.

3.2.2. Survey analysis: topic novelty and breadth

We conducted a survey with the inventors of the patents to understand their perceptions of the technological novelty and breadth of the 26 identified topics. First, we used the library WordCloud² to create a wordcloud for each topic, which is a cloud of words wherein the size of each word corresponds to the probability of the word occurring in the topic (See Appendix 3). We generated the wordclouds for the 26 topics to facilitate the inventors’ validation of the topics. Second, we designed a survey asking the inventors to assess the technological novelty and breadth of two wordclouds and respond to the following: (1) ‘Does the wordcloud capture (a) an incremental technology or (b) a radical technology?’ and (2) ‘Does the wordcloud capture a technology applicable (a) to a single technology sector or (b) to two or more technology sectors?’ For the topic assessments, each inventor received a unique link to two wordclouds in which her or his patents had the highest probability of appearing. No indication of the content of the topics or suggested names were provided to the inventors. This method allowed us to analyse experts’ assessments of the novelty and breadth of each topic. In total, we received 192 responses for the 26 topics.

4. Findings

We used the library pyLDAvis³ to visualise an inter-topic distance map and an interactive and multidimensional scale representation of all topics. This web-based interactive visualisation tool also allowed us to capture the most relevant terms associated with each individual topic as well as understand the topic-term relationships in a fitted LDA model, including their proximity to each other as well as how they differ from each other. Figure 2 provides a global view of the topic model. Figure 2 plots the topics as circles in the two-dimensional plane whose centres are determined by computing the distance between topics. In Figure 2, each topic’s overall prevalence (i.e. marginal topic distribution) is calculated using the areas of the circles (i.e. the size of the bubble) which is presented in decreasing order of prevalence from 10% for topics 1 and 2 to 1% for topic 26. The pyLDAvis library allowed us to make each topic human interpretable as they enabled us to see which technology characteristics made up a certain topic (see Appendix 3) as well as their relation, proximity to other topics and clusters of topics.

Figure 2. Clusters of Autonomous Vehicle Technologies.

We analysed the 26 most significant topics covering all patents as well as calculated the development and relevance of the topics over time to understand their evolution (see Appendix 4). We noticed that 26 topics captures the highest distribution of the patents (read Appendix 2 for more details). Next, we asked three independent practitioners with experience in the automotive industry to help name the 26 topics (see Figure 3 and Table 3). Thereafter, the practitioners also named the recognisable digital technology clusters – driving assistance systems; routing, positioning, and environmental monitoring; and data, connectivity, and in- car technology (see Figure 2).

Figure 3. Topics of the autonomous vehicle technologies.

Table 3. Novelty and Breadth of Autonomous Vehicle Technologies.

Topic No.	Topic Name	Topic Cluster	Topic Characteristics	Total No. of Responses	No. of responses: Incremental Tech.	Prob. of Incremental Topic	No. of Responses: Radical Tech.	Prob. of Radical Topic	No. of Responses: Single Tech. Sector	Prob. of Single Tech. Sector	No. of Responses: Multiple Tech. Sector	Prob. of Two or More Tech. Sectors	Total No. of Patents Included in the Topic
1	Autonomous speed control assistance	Driving assistance systems	Enhancing technologies for autonomous driving	19	18	94.74%	1	5.26%	14	73.68%	5	26.32%	43.00
2	Environmental sensor technology	Routing, positioning, & environmental monitoring	New sensor technologies	19	17	89.47%	2	10.53%	3	15.79%	12	63.16%	43.00
3	Smart navigation systems	Routing, positioning, & environmental monitoring	New navigation systems	9	9	100.00%	0	0.00%	2	22.22%	7	77.78%	6.00
4	Sensor technology for observing road conditions & obstacles​	Driving assistance systems	New sensor technologies	10	9	90.00%	1	10.00%	4	40.00%	6	60.00%	7.00
5	Collision prevention assistance	Driving assistance systems	Prevention technologies for autonomous driving	7	2	28.57%	5	71.43%	3	42.86%	4	57.14%	11.00
6	Connected system to transfer data about fleet mobility	Routing, positioning, & environmental monitoring	Connected data transfer systems for fleet mobility	6	6	100.00%	0	0.00%	3	50.00%	3	50.00%	16.00
7	Lane keeping/ overtake/traffic assistant​	Driving assistance systems	Enhancing technologies for autonomous driving	9	6	66.67%	3	33.33%	5	55.56%	4	44.44%	18.00
8	Wireless entertainment & communication system	Data, connectivity, & in-car technology	Enhancing entertainment & communication systems	7	6	85.71%	1	14.29%	1	14.29%	6	85.71%	14.00
9	Distance control assistance	Driving assistance systems	Enhancing control assistance technologies	8	7	87.50%	1	12.50%	2	25.00%	6	75.00%	3.00
10	Steering & braking control system​	Driving assistance systems	Enhancing control assistance technologies	8	6	75.00%	2	25.00%	4	50.00%	4	50.00%	13.00
11	Communication module for platooning (of aircrafts)	Routing, positioning, & environmental monitoring	Enhancing platooning support technologies	7	6	85.71%	1	14.29%	4	57.14%	3	42.86%	8.00
12	Sensor to identify & detect objects in the air	Routing, positioning, & environmental monitoring	Enhancing sensors	9	5	55.56%	4	44.44%	3	33.33%	6	66.67%	11.00
13	Range prediction route assistance	Routing, positioning, & environmental monitoring	Enhancing route assistance systems	8	6	75.00%	2	25.00%	1	12.50%	7	87.50%	15.00
14	Electronic control system for parking​	Driving assistance systems	Enhancing parking assistance technologies	8	7	87.50%	1	12.50%	3	37.50%	5	62.50%	31.00
15	Jaw & pitch estimation sensor technology	Routing, positioning, & environmental monitoring	Enhancing route assistance systems	9	7	77.78%	2	22.22%	4	44.44%	5	55.56%	6.00
16	Positioning system for navigation, incl. direction	Routing, positioning, & environmental monitoring	Enhancing route assistance systems	7	7	100.00%	0	0.00%	0	0.00%	7	100.00%	12.00
17	Failure detection system	Routing, positioning, & environmental monitoring	Enhancing failure detection systems	6	5	83.33%	1	16.67%	0	0.00%	6	100.00%	2.00
18	Device-connecting function for communication​	Data, connectivity, & in-car technology	Connection systems for communication	2	1	50.00%	1	50.00%	0	0.00%	2	100.00%	2.00
19	Camera sensor assistance for classifying the environment	Routing, positioning, & environmental monitoring	Enhancing camera sensor technologies	5	3	60.00%	2	40.00%	1	20.00%	4	80.00%	10.00
20	Drive mode assistance	Routing, positioning, & environmental monitoring	Enhancing drive mode assistance technologies	5	4	80.00%	1	20.00%	2	40.00%	3	60.00%	41.00
21	Fuel/energy consumption monitoring system	Routing, positioning, & environmental monitoring	Enhancing consumption monitoring systems	8	7	87.50%	1	12.50%	5	62.50%	3	37.50%	6.00
22	Optical track detection device​	Data, connectivity, & in-car technology	Enhancing track detection systems	4	4	100.00%	0	0.00%	0	0.00%	4	100.00%	23.00
23	Road friction analysis technology ​	Driving assistance systems	Enhancing road assistance technologies	5	4	80.00%	1	20.00%	5	100.00%	0	0.00%	15.00
24	Device to recognise light sources​	Data, connectivity, & in-car technology	New light source recognition technologies	2	1	50.00%	1	50.00%	0	0.00%	2	100.00%	4.00
25	Display & control system for seat positioning ​	Data, connectivity, & in-car technology	Enhancing seat positioning systems	3	3	100.00%	0	0.00%	0	0.00%	3	100.00%	23.00
26	Curve profile analysis technology	Routing, positioning, & environmental monitoring	Enhancing road assistance technologies	2	2	100.00%	0	0.00%	2	100.00%	0	0.00%	13.00

As computer-based research methods require careful interpretation and validation of the results (McKenny et al. 2018), our survey results helped to further validate and categorise the independent practitioners’ analysis. The survey results confirm that most digital technology topics are incremental and equally applicable to one sector (i.e. the automotive) or multiple sectors. Our respondents categorised only Topic 5 (new collision prevention technology for autonomous driving), which is part of the driving assistance systems cluster, as radical and applicable to multiple sectors. Topics 18 and 24 (device-connecting function for communication and device to recognise light sources) were equally radical and incremental and applicable to multiple sectors. Finally, surprisingly, our survey results indicate that none of the topics were radical and applicable only to the automotive industry. See Figure 4 for the typology with the results.

Figure 4. Digital Technology Categories for Autonomous Vehicle Technologies.

To further understand the technology category of each topic in relation to the potential breadth and novelty of application, we integrated the patent-level data with the 26 topics. As part of our analysis, in Appendix 5., we calculated the average of the total number of Cooperative Patent Classification (CPC) codes from all the patents for each topic, the average of the total unique CPC codes (maximum of seven CPC categories and minimum of one CPC category), the average for each CPC category, and the average number of backwards citations for each topic. CPC Code B is most related to the automotive industry. Appendix 6 shows the percentages for all the unique CPC codes.

For incremental topics, our results highlight a convincing difference between topics with one- sector application and multi-sector application. In the augmenting category (incremental and one-sector application), the topics have a low average number of total unique CPC codes, which confirms that these topics are concentrated among a few distinct technology sectors (i.e. the transportation sector and one other sector), as well as a low average number of backwards citations, indicating that the respective patent might not be based on knowledge from multiple domain areas. For example, Topics 1, 11, 21, and 26 have a low average number of unique CPC codes, which hints at the narrow range of application for these topics compared to other topics. Moreover, for the operations and transportation technology domain (i.e. CPC Code B), the aforementioned topics have large technology-domain depth in the transportation sector compared to topics in other paths. These topics correspond to different topic clusters – driving assistance systems and routing, positioning, and environmental monitoring – which confirms the validity of our results.

The topics in the spanning category (incremental and multi-sector application) have a higher average number of total unique CPC codes overall, which confirms that they cover multiple technology domains. The technology domains correspond to all the technology classifications (i.e. seven CPC codes but a lesser degree to CPC Code A, human necessities, and CPC Code E, fixed construction) and do not exclusively emerge from the transportation sector (i.e. CPC Code B) as the main technology domain. Our results also highlight that in most cases, the CPC code for transportation has a lower value than those of the other technology codes (e.g. electricity, physics, mechanical engineering). These incremental technology domains appear in all the topic clusters.

The disruptive category (radical and multi-sector application) shows the highest average number of unique CPC codes (2.09), indicating at least two technology classifications are relevant for patents in this path. We calculated the average number of the emerging technology classification for all the topics which is a novelty category assigned to patents that are expected to have a significant economic social and impact. For this paper, Topic 5 (collision prevention assistance) had the highest number (0.27), which indicates that this topic contains patents with the highest potential for disruption. Finally, Topics 6, 10, 18, and 24 are distributed between two digital technology categories, providing more nuance about the characteristics of different technology categories. These topics have the characteristic of being primarily distributed between two distinct technology classifications and a low presence across different technology sectors (lower average number of unique CPC codes).

5. Discussion and conclusions

The present study aimed to respond to the need for conceptual developments in relation to the novelty and breadth dimensions of digital technologies and how they influence industry digitalisation (Dąbrowska et al. 2022; Lanzolla et al. 2020; Lebovitz, Lifshitz-Assaf, and Levina 2022). Our findings broaden the way we think about single emerging technologies and digital technologies in particular (Röder, Both, and Hinneburg 2015; Kapoor and Klueter 2021; Bailey et al. 2022; Cirillo et al. 2022) because it demonstrates that industry digitalisation is the outcome of introducing new creative processes to embed digital capabilities into physical objects as well as building new digital capabilities outside a firm’s main industry. This paper offers two distinct contributions: namely, we propose an integrating framework to clarify the novelty and breadth of digital technologies and their impact on industry digitalisation of the automotive industry and provide a depiction of digital technologies at the meso-level.

The first contribution of this study shows that digital technologies (referred to as topics and categories of clustered topics in this paper) are primarily incremental and emerge from technologies beyond the automotive industry, including telecommunications, consumer products, defence, and machine manufacturing. While this finding is not surprising in itself, it is an important insight that calls for further and more comprehensive analyses to fully capture the interindustry dynamics of digital transformation even in cases where the main focus is on particular applications or industries (e.g. Brynjolfsson and McAfee 2014; Iansiti and Lakhani 2020). The proposed integrating framework of digital technology categories helps explain the impact of such technologies on industry digitalisation based on a combined analysis of novelty and breadth. The analysis brings attention to the importance of incrementally improving applications with digital technologies, referred to as the augmenting category in the framework, and introduces two additional functions of digital technologies, referred to as the spanning and disrupting categories. Understanding these categories more fully, especially their combinations, fulfils the need for such cross-sector analyses and enriches scholarship on technological change in general (Dosi 1984) and digital transformation in particular (Bodrožić and Adler 2022; Coccia and Watts 2020; Dąbrowska et al. 2022; Lanzolla et al. 2020).

Our second contribution depicts the characteristics of four digital technology categories across the novelty and breadth dimensions. The augmenting category involves a range of digital technologies primarily concerned with incremental changes for the automotive industry, which may be improved through automation or augmentation strategies (Lebovitz, Lifshitz-Assaf, and Levina 2022; Raisch and Krakowski 2021). Indeed, this study highlights several digital technologies that have renewed the established technological path for the automotive industry. Incumbents like Scania and Volvo have become integrators of new digital technologies (i.e. vehicle handling, perception sensors) initiated by suppliers from the same sector (i.e. Autoliv, Veoneer) and new industry entrants (e.g. Ericsson, Husqvarna). Thus, this study also reveals that digital transformation within the automotive industry mainly arises via an incremental creative accumulation process (Bergek et al. 2013) enabled by the development of new digital capabilities and the extension of component-related knowledge.

Moreover, our analysis of the spanning category reveals that digital technology topics with incremental characteristics emerge from multiple sectors, such as the electricity, construction, and mechanical engineering sectors, confirming that most digital technology topics are based on combinations of inventions from different sectors (Brynjolfsson and McAfee 2014; Teece 2018). This observation adds to the literature on technology recombination, which suggests that digital technologies have the potential to renew multiple industries because of their generative characteristics (Nambisan, Wright, and Feldman 2019; Kallinikos, Aaltonen, and Marton 2013; Thomas and Tee 2022). While our findings confirm previous research suggesting that digital technologies have integrative capabilities that enable the transformation of multiple industries (Lee and Berente 2012), it diverges from previous research investigating digital technologies as purely complementary to products or services in other industries, such as drug discovery (Dougherty and Danielle 2012) and navigation systems (Svahn, Mathiassen, and Lindgren 2017). Our findings highlight that digital technologies in the spanning category lead to the emergence of new actors and industry segments and potentially have a significant economic impact on industry digitalisation.

While the above categories for industry digitalisation appeared the most in the data upon which this study draws, unexpectedly, only one digital technology topic was identified as radical and implementable across multiple sectors – namely, the disrupting category. This category has the potential to transform both the automotive industry and several other industries, such as the telecommunications and manufacturing industries as well as have impact on city planning, transport regulation and policy and user acceptance. Until now, the established view from earlier studies shows that technological change is rare and emerges from radical innovations, such as digital imaging (Benner and Tripsas 2012) or semiconductors (Dosi 1984). Our finding extends the innovation literature and the assumption that digital transformation involves industrial disruption by definition (Kwon et al. 2019; Bodrožić and Adler 2022). We show that these kinds of digital technologies have the potential to generate a significant social and economic impact without disrupting an entire industry but instead transforming it. Moreover, we suggest that digital transformation may emerge from technologies spanning several sectors, enabling industry digitalisation across multiple technology domains.

The most notable conclusion from this study calls for slowing down our expectations on disruption emerging from radical digital technologies unless these are combined with corresponding but less radical technologies that enable full implementation across a range of applications. A concrete illustration of such a combinatorial application in the automotive industry is collision prevention assistance technology, which has the potential to change the entire automotive industry and also generate disruption in several related industries – basically everything related to mobility transformation. Realising such potential, however, requires a wide range of incremental improvements in related technologies, such as navigation, speed control, and road friction analysis. This insight illustrates the value of moving beyond studying digital transformation as the result of the successful implementation of revolutionary technologies to instead studying it as the result of interdependent clusters of both radical and incremental technological advancements (Bailey et al. 2022).

In addition to calibrating our expectations on radical innovations emerging from the infusion of digital technologies in mature industries, such as the automotive industry, this study also provides insights applicable to innovation policy. Increased awareness of the required combinations of technologies of varying novelty and breadth that feed into related and interdependent applications and sectors calls for broader innovation policy. Instead of focusing resources on directly promoting specific technologies or industries (Grillitsch et al. 2019), which has been a trend in innovation policy for several decades, it is worth considering more general policy measures, not directly targeting predefined sectors. The rationale for such a call is the unpredictable nature of digital innovation and link between technology progress and industry application, and thereby the highly limited ability for policy to predict what application will eventually generate innovative outcome. Such a more general approach would thus enable both identified technologies and applications and not yet identified technologies and applications to take advantage, thereby avoiding the risk of dominant regimes or actors capturing policy measures which would generate better value when targeting alternative actors.

Finally, we also want to emphasise the ample opportunities provided by digital social research methods for management scholars (George et al. 2016; Hannigan et al. 2019). For example, to further develop and validate our integrating digital transformation framework, scholars could extend the findings presented in this paper to perform multi-industry analyses that explain the types of knowledge combinations that lead to radical digital technologies (disruptive category) or the digital technologies that have a wide scope of application (spanning category). Furthermore, intuitively, a multi-industry analysis that includes other sources of information, such as industry reports and other forms of text (Grimmer, Roberts, and Stewart 2022), might help provide more granularity and variety across digital technology topics. This research design strategy is particularly important for understanding the characteristics of the transformative and disruptive categories. As suggested by Salganik (2018), only further application of computational research methods and information can help improve understanding of phenomena (i.e. digital transformation and industry digitalisation in our case).

Further research using patent data could also enable researchers to study what personal characteristics of inventors lead to digital technologies of different novelty and breadth. Such a study could untangle the breadth and depth of knowledge expertise needed to create, for example, multi-industry digital technologies. This type of analysis could help respond to questions about industry path dependency (Benner and Tripsas 2012) and knowledge search heuristics (Lopez-Vega, Tell, and Vanhaverbeke 2016). Another promising research area is studying how digital technologies and what innovation strategies trigger firms, universities, government institutions, and society at large to work towards tackling societal development goals and grand challenges and ultimately shape industry transformation in developed and emerging markets (Lopez-Vega and Lakemond 2022). Only by doing so will we be able to explain the benefits of emerging digital technologies and the opportunities provided by the digital age.

Acknowledgement

The authors would like to thank the special issue co-editor Prof. Riccardo Leoncini and the anonymous reviewers for their guidance and insightful comments. Also, the authors would like to express their gratitude to the survey respondents and research assistants of the www.organizationalrenewal.com project. Responsibility for the information and views expressed in the article lies entirely with the authors.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This study was supported by Marianne and Marcus Wallenberg Foundation. MMW 2016.0014

Notes

¹ Link to the database: https://www.ekh.lu.se/en/research/economic-history-data/swinno

² https://amueller.github.io/word_cloud/.

³ https://pyldavis.readthedocs.io/en/latest/.

⁴ https://pandas.pydata.org.

⁵ https://docs.python.org/3.7/library/re.html.

⁶ https://www.nltk.org/_modules/nltk/tag.html..

⁷ http://www.nltk.org/_modules/nltk/stem/wordnet.html..

⁸ https://radimrehurek.com/gensim/index.html.

Appendices Appendix 1.

Topic Modelling Procedure

Appendix 2.

Dataframe, Corpus, and Topic Modelling

The first part of the data analysis required the patents to be in text form (i.e. to be in English and transferred from a PDF to a text file [.txt]). That means that the starting point or prerequisite for creating our corpus of all 455 relevant patents was that they had to be in a.txt file format, which is computer readable and thus easy to import into Python. We created these.txt files and included all information from the patents from the abstracts to the claims. To work with the patent data, we imported each text file into Python by transferring the data and storing it in a pandas dataframe. Pandas⁴ is a common open-source Python library that offers easy-to-use high-performance data structures and data analysis tools for the Python programming language. Within the dataframe, which one can imagine as a matrix, we had 455 rows and two columns— one containing the patent IDs and one containing the respective text content of each patent. The text content was stored as a so-called string, which is the text representation within the Python environment. Once all of the patents were imported into Python and stored in a computer-readable format, pandas allowed us to easily work with the data.

To serve as a good input in the subsequent topic modelling and to enable the generation of expressive topics, the text data needed to be processed. This procedure started with converting the text of each patent into lowercase, ensuring that we treated all words the same and avoided subsequent specific factor errors (McKenny et al. 2018). Afterwards, we used regular expressions⁵ (RE) to specify a set of strings that matched the REs, allowing us to coarsely delete strings without any meaning for our purpose. In our case, the specification was that the strings should only contain alphabetic letters from A to Z and be at least two characters long, which meant that all unnecessary numbering, brackets, and fill words, which are numerous in patent text, were deleted.

Having pre-cleaned the patent text, we proceeded by tokenising the strings, which, until that moment, represented one patent each. Tokenising means that we broke down large strings containing multiple sentences into individual words separated by commas, which we then stored individually as strings. This step represented a crucial prerequisite for the actual topic modelling described below and allowed us to further clean the text. The next step was to eliminate all words that had no inherent meaning, also called stopwords, which was done by setting up a stopword list and then deleting all words in this list from the patent text. To find relevant stopwords, we started by using the pre-defined stopword list provided by nltk to cover general stopwords like ‘the’, ‘and’, or ‘of’. We then manually extended this list with other stopwords that appeared to have no direct meaning or were too general like ‘figure’, ‘claim’, or ‘illustrate’ as almost all patents contained the terms ‘patent claim’ and ‘figure’, but these terms did not convey any meaning related to autonomous vehicle technologies.

The final step in creating the corpus was to lemmatise the words. This means that the words were transformed to their most basic form, which in turn reduced the number of words with the same meaning. In other words, we tried to reduce the number of distinct words with the same meaning – for example, plurals (reduce ‘vehicles’ to ‘vehicle’). We began this procedure by tagging every word in the corpus according to its position in the text, which was computed using the nltk.pos_tag() function⁶ This function took in every word and added a part-of-speech tag (pos_tag) to it, which specified some property of the word or token in the text, such as noun, verb, adjective, etc. This extra information about the context of the words allowed for better lemmatisation, which we performed using another part of the nltk library, the WordNetLemmatizer⁷ This module finally executed the lemmatisation by taking into account the pos_tag of the words and thus performed the last step in creating the corpus. The resulting text data was still stored in the pandas dataframe, which was then saved on the computer’s hard drive. This step was done for two reasons: first, to optimise computing power and time by decoupling the time-intense process of generating the text corpus from the topic modelling and data analysis and, second, to have a version history of the dataframe, which in turn assisted in assessing what corpus was better suited for our purpose by allowing a comparison of two corpuses (one for training and one for validation).

The second part of the topic modelling involved generating the topics using gensim⁸, an open- source library that offers text-analysis tools. We employed LDA, which is the most commonly used topic modelling method and represents a generative probabilistic model for unsupervised text-corpus analysis (Blei, Ng, and Jordan 2003). This method has also been employed by different innovation management scholars (von Krogh 2018; Hannigan et al. 2019). First, we loaded the saved dataframe into Python and prepared it for the topic-modelling algorithm. This step was necessary as the algorithm had certain input parameters that needed to be matched exactly to be functionable. First, we created a dictionary and a bag of words (BoW) out of the loaded dataframe using gensim. The dictionary was created using the gensim.corpora.Dictionary object, which, in essence, provided a representation of the distinct words that made up the whole corpus and further contained the word frequencies, or the number of occurrences of a certain word within the corpus. The raw dictionary contained 16,741 distinct words, which were represented through a number (the token_id) and ultimately made up the corpus of 1,579,447 words. These distinct words comprised words that occurred in only one document, such as ‘nutrient’, as well as words that occurred in almost all documents, such as ‘vehicle’ or ‘unit’. As the words that occurred in only very few documents either contributed nothing to cross-document topics or were corrupted, we filtered the dictionary, excluding all words that occurred in fewer than five documents, eventually leaving the dictionary with 4,651 distinct entries. We manually checked that the excluded words did not convey any relevant meaning about autonomous vehicle technologies. The BoW was created using gensim’s doc2bow function, which contained the information about the words in each document with their individual word frequencies within this document.

To derive to our final topic model, we iteratively generated the corpus and the topic model and assessed the resulting topics, which, in each iteration, resulted in either new stopwords that then had to be deleted from the corpus or adapted parameters for the topic-modelling algorithm. This procedure, especially the assessment of the resulting topics, was supplemented by two intrinsic performance indicators: the perplexity and coherence score of the model. The perplexity, measured as the normalised log probability of a retained test set, describes how surprised a model is by new data it has never seen before or, in other words, how well the model can represent or reproduce the statistics of the retained data. The coherence score measures the semantic similarity of the most probable words within one topic and thus represents a measure for human interpretability of the topics (Röder, Both, and Hinneburg 2015). In our case, the perplexity is -7.08, and the coherence score is 0.43, which represent acceptable values that are on the lower end of what is assumed to be ‘good’ according to (Stevens et al. 2012).

The dictionary and BoW were then the prerequisites and basic inputs for the topic-modelling algorithm. The algorithm had a multitude of parameters to specify exactly how it performed the modelling, which were largely kept at their default values, with changes only made to the following: num_topics = 30, random_state = 100, chunksize = 60, passes = 10, iterations = 500, per_word_topics = True. The most important and influential change was the number of topics (num_topics) that the algorithm should generate, which in our case was 30. The practitioners noticed that Topics 27 to 30 did not provide any meaningful insights, so these topics were excluded from our analysis.

Appendix 3:

Wordclouds for Topic 1 to Topic 26

Appendix 4:

Topic and cluster development over time

Appendix 5:

Technology Classification of Topics

Table

					Average no. of Cooperative Patent Classification (CPC) codes
Topic No.	Topic Name	Topic Cluster	Topic Characteristics	Average no. of Backward Citations	For Each Topic	Unique CPC Codes (Total 7 Categories)	Category A (Human Necessities)	Category B (Transpora-tion and Perf. Operations)	Category E (Fixed Construc-tion)	Category F (Mechanical Eng., Lighting, Heating)	Category G (Physics)	Category H (Electricity)	Category Y (General Tagging of Cross-Sect. Tech.)	Emerging Technology Classification
1	Autonomous speed control assistance	Driving assistance systems	Enhancing technologies for autonomous driving	4.44	9.30	1.91	0.00	3.63	0.05	0.00	4.67	0.93	0.00	0.00
2	Environmental sensor technology	Routing, positioning, & environmental monitoring	New sensor technologies	26.53	9.74	2.00	0.58	2.67	0.02	0.05	4.19	1.21	1.02	0.16
3	Smart navigation systems	Routing, positioning, & environmental monitoring	New navigation systems	6.17	9.83	1.83	0.00	5.83	0.17	0.00	3.67	0.17	0.00	0.00
4	Sensor technology for observing road conditions & obstacles​	Driving assistance systems	New sensor technologies	6.00	5.00	1.43	0.14	2.71	0.00	0.00	2.00	0.14	0.00	0.00
5	Collision prevention assistance	Driving assistance systems	Prevention technologies for autonomous driving	10.82	9.91	2.09	0.55	6.45	0.00	0.00	2.18	0.09	0.64	0.27
6	Connected system to transfer data about fleet mobility	Routing, positioning, & environmental monitoring	Connected data transfer systems for fleet mobility	13.00	5.19	1.50	0.06	1.31	0.00	0.00	3.31	0.13	0.38	0.06
7	Lane keeping/ overtake/traffic assistant​	Driving assistance systems	Enhancing technologies for autonomous driving	13.17	8.50	1.61	0.11	5.39	0.00	0.00	3.00	0.00	0.00	0.00
8	Wireless entertainment & communication system	Data, connectivity, & in-car technology	Enhancing entertainment & communication systems	9.14	8.79	1.93	1.14	3.43	0.00	0.00	4.00	0.07	0.14	0.14
9	Distance control assistance	Driving assistance systems	Enhancing control assistance technologies	0.00	7.33	1.67	0.00	5.33	0.00	0.00	2.00	0.00	0.00	0.00
10	Steering & braking control system​	Driving assistance systems	Enhancing control assistance technologies	3.08	7.62	1.77	0.23	5.77	0.08	0.00	1.38	0.15	0.00	0.00
11	Communication module for platooning (of aircrafts)	Routing, positioning, & environmental monitoring	Enhancing platooning support technologies	13.50	7.88	1.38	0.00	5.75	0.00	0.00	2.13	0.00	0.00	0.00
12	Sensor to identify & detect objects in the air	Routing, positioning, & environmental monitoring	Enhancing sensors	10.09	8.73	1.73	0.00	3.45	0.18	0.00	4.27	0.73	0.09	0.09
13	Range prediction route assistance	Routing, positioning, & environmental monitoring	Enhancing route assistance systems	12.07	10.13	2.07	0.47	4.20	0.00	0.67	3.80	0.33	0.67	0.20
14	Electronic control system for parking​	Driving assistance systems	Enhancing parking assistance technologies	11.48	8.81	1.74	0.13	4.61	0.06	0.00	3.55	0.00	0.45	0.10
15	Jaw & pitch estimation sensor technology	Routing, positioning, & environmental monitoring	Enhancing route assistance systems	6.33	9.67	2.00	0.00	4.50	0.00	3.17	1.83	0.00	0.17	0.17
16	Positioning system for navigation, incl. direction	Routing, positioning, & environmental monitoring	Enhancing route assistance systems	11.75	6.00	1.33	0.00	2.50	0.00	0.00	2.00	1.00	0.50	0.08
17	Failure detection system	Routing, positioning, & environmental monitoring	Enhancing failure detection systems	5.50	4.00	1.50	0.50	2.50	0.00	0.00	1.00	0.00	0.00	0.00
18	Device-connecting function for communication​	Data, connectivity, & in-car technology	Connection systems for communication	7.00	13.00	1.50	0.00	10.00	0.00	0.00	3.00	0.00	0.00	0.00
19	Camera sensor assistance for classifying the environment	Routing, positioning, & environmental monitoring	Enhancing camera sensor technologies	10.40	6.30	1.50	0.20	4.70	0.00	0.00	1.00	0.30	0.10	0.10
20	Drive mode assistance	Routing, positioning, & environmental monitoring	Enhancing drive mode assistance technologies	6.71	6.93	1.61	0.02	2.02	0.05	0.00	4.00	0.80	0.02	0.02
21	Fuel/energy consumption monitoring system	Routing, positioning, & environmental monitoring	Enhancing consumption monitoring systems	6.33	8.33	1.50	0.50	4.50	0.00	0.00	2.67	0.67	0.00	0.00
22	Optical track detection device​	Data, connectivity, & in-car technology	Enhancing track detection systems	3.35	7.70	1.30	0.00	6.00	0.04	0.00	1.61	0.00	0.04	0.04
23	Road friction analysis technology ​	Driving assistance systems	Enhancing road assistance technologies	18.07	7.67	1.33	0.00	0.73	0.00	0.00	1.47	4.73	0.73	0.20
24	Device to recognise light sources​	Data, connectivity, & in-car technology	New light source recognition technologies	2.50	4.25	1.50	0.00	0.75	0.00	0.00	3.00	0.50	0.00	0.00
25	Display & control system for seat positioning ​	Data, connectivity, & in-car technology	Enhancing seat positioning systems	8.48	8.00	1.30	0.00	2.48	0.00	0.00	5.22	0.22	0.09	0.04
26	Curve profile analysis technology	Routing, positioning, & environmental monitoring	Enhancing road assistance technologies	8.77	10.54	1.77	0.00	7.85	0.31	0.00	1.69	0.08	0.62	0.23

Appendix 6:

Patent Distribution according to CPC codes