Discovering the proximity between technical fields have been proved is important for the global understanding of science and technology, facilitating, for example, knowledge technology transfer, research collaboration between institutions (Woerter, 2012), or the identification of technology opportunities (Jaffe, 1986). By discovering technology proximity, we mean to assess the similarity of technologies, namely the share of knowledge and techniques. In the case of characterizing technology using patent information, proximity could be evaluated by identifying the commonalities at classification level (Jaffe, 1986; Simon and Sick, 2016; Alstott et al., 2017). In other words, two technologies are closer to each other than a third one when the first two have more classification information in common in any way that the third one (Boyack et al., 2000; Schoen et al., 2012; Woerter, 2012; Okubo, 2017).

Revealing the technical fields where companies have industrial or research developments could play a key role in the decision-making of technology players in areas such as analysis of competition (Schoen et al., 2012), firm collaboration (Simon and Sick, 2016), or mergers and acquisitions (Simon and Sick, 2016).

Our objective with the present work is the visualization of the relative position of technology, that is placing in a graph technical fields and tech companies with respect to each other. The idea is to use a single data structure based on patent classification information for characterizing both, technical fields and organizations (company, research institution, or university lab), in order to visualize them as items of the very same body, namely technology. In this way, observations such as proximity or dynamics could be extended from technical fields and tech companies to each other.

We have characterized a technical field or an organization (company, research institution, or university group) by a single data structure, the TechSpectrum, formed by the aggregation into bins¹ of the IPC codes assigned to a set of patents and its prior art, mixed in a specific proportion, and ordered according to the IPC (Perez-Molina, 2018).

This binning and ordering is done at different IPC granularities, generating a particular TechSpectrum for each IPC level² (see for example in Figure 1 the TechSpectrum graph at IPC-SubClass level for propulsion of electric vehicles—IPC code B60L—and Toyota Motors).

A TechSpectrum is a histogram, each bin of which corresponds to an specific IPC code, and it could be considered as an ax of a multidimensional space, namely the IPC space. The bin values would be in consequence coordinate values. Under this perspective, the visualization of technology, that is a technical field or a tech companies represented by selected sets of patents and its prior art citations, would be merely be the representation of dots in the positions corresponding to the respective set of coordinates values, namely the bin values.

We have developed an analysis tool which generates two graphs, the Technology Map—TechMap—and the Focused Technology Map—F-TechMap—.

The first graph, the TechMap, visualizes a given set of items, thus technical fields and tech companies, in a 2D plot in such a way that those related items they will appear closer between them that the unrelated ones.

The fact that our system has a high dimensionality, (the dimension is 8, 132, 651, and 7,590 at IPC–Section, Class, subClass, and Group, respectively)³, makes difficult a straightforward representation difficult. An approach to visualize such high dimensional spaces is, first to reduce the dimensionality to two dimensions, and then generate a visualization in a 2D plane. For this task we have applied MultiDimensional Scaling—MDS—, the downside of such an appreciable reduction of dimensionality is the introduction of a geometric distortion (Colange et al., 2019).

Note that for our aim of exploring proximity between technologies, MDS has the interesting feature of reducing the dimensionality to two dimensions preserving as much as possible the relative distances between the items to be visualized. In other words, items that are close in the IPC space stay close to each other, and those far stay far. On the other hand, the absolute positioning of the items within this visualization is meaningless, namely the fact that a given item is located in a particular position in the 2D plane provides no information, only the closeness or farness to any other item is meaningful (Borg and Groenen, 2005).

MDS visualizes the items in a 2D plane based on the distance information using a nonlinear dimensionality reduction, so the distances between the elements are is the key information to gather. This information is technically provided as a distance of dissimilarity matrix.

The distance between items to visualize is computed with the soft cosine (Sidorov et al., 2014) algorithm, wherein the features are the IPC codes, and the similarity measure between features is computed by applying a hierarchical similarity measure for hierarchical classification schemes (Caspersen et al., 2017).

Our second graph, the Focused TechMap—F-TechMap—is inspired on by the work of Urpa and Anders (2019), and takes the shape of a circular—radial graph with the focus at the center of the circle and a set of radii, each corresponding to a single IPC code at a certain level of resolution. Once an item between the technical fields or companies present in the TechMap is chosen as to be the focus, then every other item to be visualized is located at the computed distance on a radius corresponding to a given IPC-code.

The F-TechMap also uses the distances computed with the soft cosine algorithm, and it is a complementary graph to the TechMaps because it immediately uses computed distances avoiding the distance distortions produced by MDS in its projection on the 2D plane. The dynamic version of these two graphs will provide rich information about the evolution of companies' technical developments and the evolution of technical fields.

Two study cases are presented with some technical fields and tech companies. The data collection of the study cases is done using PatStat–online⁴, an EPO's database in the field of patent intelligence and statistics⁵. A set of queries is executed in PatStat to collect the patents and citations of the technological fields or companies under study. In all our study cases, patents are limited to the 1980–2015 range of years in order to have a large interval of time⁶. The patent documents collected are also constrained to be published by the USPTO⁷ in order to avoid duplication.

The research question of the present work is what can we learn about tech companies' development activities, and about technical fields' evolution by visualizing patent classification information-based maps.

The potential contribution of our work in the field of innovation management lies in the creation of a new graphical tool for the relative positioning of technologies (technical fields or tech companies) based on patent classification information, improving the perception and analysis of research and industrial activities of technology players. In the field of history of technology, our procedure contributes by facilitating the study of the evolution of specific technical fields and tech companies.

The rest of the paper is organized as follows: Section 2 discloses related works. Section 3 discloses the procedure to generate our graphical tool. Section 4 presents some case studies. Section 5 presents a discussion and the applications of our tool. Section 6 outlines future research and conclusions.

In our work a technology is defined as a set of patent documents assigned to a technical field or belonging to a tech company. The set of patents owned to by an inventor, or a group of inventors could also be considered, but this work is limited to tech fields and companies.

The patents are collected from PatStat, an EPO's database in the field of patent intelligence and statistics. A set of SQL–queries is executed in Patstat to collect the selected patents—and its prior art citations—of the technological fields or companies under study. The patent documents collected are constrained to first, to the range of years, from 1980 to 2015, in order to have a large interval of time, and second, to be published by the USPTO to avoid duplication⁹.

The TechMap generation procedure is outlined in the following five steps:

Step 1: Data gathering.—First, we collect the set of patents published for the specific technical fields or tech companies to be visualized, then its cited prior art is also collected and mixed, forming a final collection. In this final collection, the prior art is weighted to a certain percentage which experimentally we have set to 10%;, the idea of this reduction is to keep the basic composition of the initial set of patents but enriching it with the prior art citations.

Step 2: TechSpectrum generation.—For each final collection, the classification IPC codes are identified, and ordered in bins according to the IPC at the first four levels of classification resolution¹⁰.

Step 3: Distance computation.—A matrix of distances between every and each and every item (technical fields or tech companies) to be visualized in the Tech Maps is computed using the soft cosine distance (Sidorov et al., 2014), which is uses the following formula:

A and B represents vectors formed by bins of the final collections of two specific technologies (tech fields or companies). A_i and B_j—are the bin i and j of each vector, namely the IPC codes ordered according to the IPC at position i, and j of the two specific technologies.

For computing the similarity factor among each possible couple of IPC codes—S_ij—we have adapted the method proposed by Wu and Palmer (1994) to structural information of the hierarchical IPC scheme, namely the number of IPC levels between the path of each couple of IPC-codes C_i and C_j resulting in the following formula:

N_i, N_j, and N_IJ are the number of nodes on the path from C_i to C_IJ, from C_j to C_ij, and from C_ij to the Root node, where C_IJ is the least common IPC code of C_i and C_j.

In this manner, four distance matrices are formed with the soft cosine distances between every possible couple of items corresponding to the first four IPC levels. Then, we mix these matrices to form a global matrix¹¹.

K_S, K_C, K_SC, and K_G, and d_C, d_S, d_SC, and d_G are the mix factors and the soft cosine distances at IPC Section, Class, SubClass, and Group, respectively. We have experimentally set the parameters to K_S=1, K_C=2, K_SC=3, and K_G=5.

Step 4: MDS computation.—The dimensionality reduction is made at the global level by MDS using the SKlearn python library¹². The MDS results in a collection of 2D coordinates corresponding to each item to be visualized.

Step 5: TechMap visualization.—The MDS results are visualized using the MatPlotLib python library (Devert, 2014). In the graphs, in order to facilitate the understanding, the tech companies are visualized as blue dots and the technical fields as red dots. An example of a TechMap with an IPC code¹³ and some tech companies is shown in Figure 2, left graph.

The Focused TechMap generation procedure is outlined in the following three steps:

Step 1: TechMap generation.—The global TechMap is generated and visualized for a given set of items (technical fields or tech companies) executing the TechMap generation procedure explained above.

Step 2: Focus selection.—Once the TechMap is generated, we select one of the visualized items to be the focus of the graph.

Step 3: Focused TechMap visualization.—The focus is drawn at to the center of a circle wherein a set of radii is in turn plotted; this set contains a radius corresponding to each IPC code. Then, each item is drawn at its real computed distance to the focused item, on the radius corresponding to its main IPC code. The main IPC code for technical fields and tech companies is the IPC code with the highest figures. An example of a focused TechMap with an IPC code and some tech companies is shown in Figure 2, right graph.

As mentioned above, the data collection of the study cases is done using SQL queries in PatStat to gather the selected patents, and its their prior art citations, and the corresponding assigned IPC codes. The dimensionality reduction, and visualization is implemented with MDS, and the distance or dissimilarity matrix in all the study cases is computed using the SoftCosine distance.

For the study cases, we have selected two technical fields, namely: medical science—A61—and vehicle technology—B60—, and the following companies¹⁴: Apple (APL), Boston Scientific (BSc), Cisco (CSC), Ericsson (ERS), Ford Motor (FRD), GE-medical (GEm), General Motors (GMt), Hyundai Motors (Hyu), IBM (IBM), Medtronic (MDT), Nissan (Nss), Microsoft (MSFT), Nihon Kohden (NKH), Olympus-medical (OLYM), Omron-medical (OMR), Philips (PHLP), Shimadzu (SHI), Siemens Healthcare (SIEM), Toyota Motors (TYT), and Volkswagen (VW).

We have chosen the medical field for personal and academic interest since the origin of this work is in a research project in the biomedical engineering field. The second study case, vehicle technology, was chosen for its general interest as a technology in undergoing important transformation at present and because it encompasses multidisciplinary properties. On the other hand, the set of tech firms was selected to have a heterogeneous group with big- and mid-size corporations from different sectors and geographic origins.

The visualization of proximity between technical fields and tech companies with respect to each other furnishes an insight into industrial and research developments. Such visualization provides information about firms' know-how, which facilitates potential collaborations and, technology opportunities and weakness. From this perspective, our work presents two complementary graphical tools for studying the relationship of technical fields and tech companies, and its evolution.

The information provided by our tool can help technology decision makers to understand competitors. For example, the evolution of motor car companies toward electricity technology can be explored and better understood by generating the TechMaps and then focusing on specific IPC codes or companies. Such graphs could also help to perceive the kind of knowledge that a specific technology gathers at a certain moment. Furthermore, the identification of these trends can be of great interest to researchers in the area of the history of technology.

The relative positioning of tech companies can be used to discover know-how or to explore complementarities between companies for potential mergers and acquisitions, or research collaborations. See, for example, the position of Apple in the TechMap and the focused TechMap of Figure 2 as a telephone set developer together with Ericsson in relation to IBM and Microsoft, two computing companies in principle similar to Apple.

It is important to highlight that our approach is to consider technical fields and companies (both defined by sets of IPC codes) as similar bodies, and they are represented by the same data structure, the TechSpectrum, namely the histogram of the IPC codes assigned to a set of patents and its citations mixed in a specific proportion, and ordered according to the IPC. This is a major difference with respect to the previous studies on visualization of technical fields and companies.

This conception of technical fields and tech companies as similar bodies of technology bring important implications in to the processing and visualization of technology. First, the processing is the same for technical fields and companies, so changes in time in both are tacking taken equally into account providing an overview on technical change as a whole. Second, observations for technical fields can be extended to companies.

Our TechMap has some distinctive features distinctive from traditional technology maps or landscapes. In TechMap the items, technical fields or companies, are located relative to each other, there is not absolute positioning. In previous maps or landscapes, the location of technology items is was computed and coordinate data are were assigned to it, so that if you added or removed an item, it is was added or removed to the map but the rest of it remained unchanged; whereas in TechMaps, removing or adding a technology item results in a computation of the new position of each item. Another distinctive feature is the capability to visualize tech companies in relation to technical fields at different levels of conceptual resolution by going deeper in the IPC classification levels. This operation is a sort of conceptual zoom rather than a zoom-in graphical operation as is usually the case in -known technology maps or landscapes such us the multiresolution zooming of Boyack et al. (2000).

The presence of clusters of technical fields within our TechMaps can show up the existence of some commonalities which can be useful for researchers and innovation managers to better understand these technical fields. See for example in Figure 4 how the IPC subclass A61P, A61k, and A61Q appear grouped. Or the, not obvious proximity between measuring systems for diagnosis—A61B5—and instruments for auscultation—A61B7—shown in Figure 5. It is interesting to highlight the exploration possibilities opened by visualizing with our two graphical tools tech companies in relation to technical fields, such as in Figure 2 where IBM and Microsoft appear grouped when they are positioned within a set of companies and the technical field of telephone sets—H04M1—.

In this work we introduced a new graphical tool based on patent classification information to study technical fields represented by an IPC code (or a combination of codes) and tech companies. Each item, technical field, or company is characterized by a weighted mix of the classified (or assigned) patents and its cited prior art.

The first graph, the technology map—TechMap—, visualizes a set of given technologies, technical fields, and tech companies, positioning them in relation to each other. Our second graph, the focused TechMap, generates a visualization of the given technologies in relation to a selected one, the focus, which is located at the center of a circle with drawn radii corresponding to every IPC code. In this graph, the items are positioned at the computed distances from the focus, and on the radius representing its IPC code with the highest figures.

Although we have illustrated our tool with two case studies, namely medical and automotive technology, our tool has also been tested in a variety of fields such as heartbeat monitoring, 3D printing, hybrid vehicles, electronic devices, computer graphics, and or telecommunications to test for the ecological validity of our tool. So far, the authors found that this is a representative view in terms of ecological validity, except for the chemistry field which needs further dedicated testing due to its construct nature of how patents are triaged and presented. Concerning tech companies, we have tested the tool with numerous firms, and we have found that their use in our tool is constrained to firms active in patenting.

In the field of history of technology, our graphical tool contributes by facilitating the study of the evolution of specific technical fields, and to trace the divergence or convergence of tech companies. The contribution of the present paper lies in the creation of two new and complementary graphs based on patent classification information. The relative positioning of technologies (technical fields or tech companies) helps to better identify those that have some techniques in common because they appear close in the graphs, as well as to improve the understanding of the technical developments or trends in firms as was illustrated for motor car companies moving from pure mechanics to electrical technologies.

At present we are developing some algorithms for automatic clustering of items within TechMaps. Moreover, we foresee doing predictive modeling of the dynamics of the technologies to anticipate its their evolution, especially for tech companies.

Further research will be oriented to compute TechMaps with smaller time intervals in order to have more time resolution in the evolution perception to present animated versions of the visualizations and to investigate the modeling of trends.

In our study cases, we have highlighted some real data and events to give an indication of the consistency of our visualization tools. Further research will be carry carried out to systematically evaluate whether the company's perception of their technological developments corroborates with our tools.

We will explore the improvement of the similarity matrix using text similarity and citation network analysis of the patents classified in the respective IPC codes.

The datasets presented in this study can be found online via the following link: https://data.epo.org/expert-services/index.html.

EP-M conceived the original idea and the presented method and visualizations. FL supervised the findings of this work, discussed the results, and worked on the manuscript. Processed the experimental data, performed the analysis, and wrote the paper with input from both authors. Both authors contributed to the article and approved the submitted version.