Inter-organisational patent opposition network: How companies form adversarial relationships
II NTER - ORGANISATIONAL PATENT OPPOSITION NETWORK : H
OWCOMPANIES FORM ADVERSARIAL RELATIONSHIPS
Tomomi Kito
Faculty of Science and EngineeringWaseda University3-4-1 Okubo, Shinuku-ku, Tokyo 169-8555 Japan [email protected]
Nagi Moriya
Waseda Innivation LaboratoryWaseda University1-6-1 Nishiwaseda, Shinjuku-ku, Tokyo 169-8050 Japan
Junichi Yamanoi
Waseda Innivation LaboratoryWaseda University1-6-1 Nishiwaseda, Shinjuku-ku, Tokyo 169-8050 JapanSeptember 10, 2020 A BSTRACT
Much of the research on networks using patent data focuses on citations and the collaborationnetworks of inventors, hence regarding patents as a positive sign of invention. However, patenting is,most importantly, a strategic action used by companies to compete with each other. This study shedslight on inter-organisational adversarial relationships in patenting for the first time. We constructedand analysed the network of companies connected via patent opposition relationships that occurredbetween 1980 and 2018. A majority of the companies are directly or indirectly connected to each otherand hence form the largest connected component. We found that in the network, many companiesdisapprove patents in various industrial sectors as well as those owned by foreign companies. Thenetwork exhibits heavy-tailed, power-law-like degree distribution and assortative mixing, making itan unusual type of topology. We further investigated the dynamics of the formation of this networkby conducting a temporal network motif analysis, with patent co-ownership among the companiesconsidered. By regarding opposition as a negative relationship and patent co-ownership as a positiverelationship, we analysed where collaboration may occur in the opposition network and how suchpositive relationships would interact with negative relationships. The results identified the structurallyimbalanced triadic motifs and the temporal patterns of the occurrence of triads formed by a mixtureof positive and negative relationships. Our findings suggest that the mechanisms of the emergenceof the inter-organisational adversarial relationships may differ from those of other types of negativerelationships hence necessitating further research.
Innovation is one of the key issues in the recent economic paradigm and has consequently attracted a growing numberof academic studies over time. Enhanced by recent computational advances, patent statistics derived from large-scalepatent databases have been extensively used to help gain insight into innovation processes, drivers, and appropriatemeasures for innovation values [26, 47, 21].Network science has significantly contributed to patent data science. Considerable effort has been devoted to analysingnetworks of patents connected by citations, with the intent to understand technological change and impact [9, 46]. Theconsensus here is that the patent citation network structure reflects the evolutionary pattern of human technology. Themeasures for the technological importance of a patented invention are mostly based on how frequently the patent iscited by subsequent patents (e.g., [13, 14, 1]). Another stream of patent network research is on the social structure a r X i v : . [ phy s i c s . s o c - ph ] S e p PREPRINT - S
EPTEMBER
10, 2020of collaboration among patentees. Through the analyses of co-applicant patent networks, various factors that maycontribute to the production of high-quality inventions have been identified. These factors include the proximityof geographical locations and technological fields of collaborating organisations, their positioning in the inventors’networks, and so on [3, 12, 15].These studies consider patents as a positive sign of invention, which is a generally acknowledged view. How-ever,patenting is, most fundamentally, a strategic action used by applicants to gain legal right in order to exclude othersfrom making, using or selling an invention. While inter-organisational relationships can be intermittently friendly (e.g.,R&D collaboration and patent co-ownership), they are mostly adversarial/rivalry. This (rather essential) aspect of patent-ing has not received much attention in network science, and data science in general, to date. In management science,the strategic management of patents has been increasingly considered as the core for enhancing the competitivenessof companies. Considerable research has been conducted on the development of theoretical frameworks, case studieson the global patent wars, and analysis of the financial impact of strategic actions (for summary see Refs. [41, 17]).However, the existing research does not go beyond treating the dyadic relationships between companies. It furtherprovides no view on the network of companies connected by strategic actions.This study illuminates patent opposition as a company’s key strategic action against others. Patent opposition is a legalaction that a company (or individual) can take to challenge the validity of a patent within a certain period (usually 6–9months) after grant. If an opposition is ‘successful’, the opposed patent is revoked and cannot take effect in any ofthe signatories. Therefore, companies thus oppose patents owned by rival companies clearly intending to hinder theirinnovation activities [23, 43, 19]. In this study, we construct the patent opposition network where the nodes representscompanies, rather than patents, and the edges represent oppositions. We analyse the properties of this network, with theaim of gaining insights into how companies may form adversarial relationships between each other. In social networkanalysis, although the main stream of research addresses networks with positive relationships (e.g., friendship andco-authorship), the importance of negative relationships (e.g., interpersonal dislike, conflict, and social exclusion) hasattracted more interest [8, 28, 45]. Scholars have pointed out distinct features of negative relationship networks that arefundamentally different from positive relationship networks. Such features include the low connectivity among nodes,and a low level of transitivity – that is, while friends of friends are usually friends in positive relationship networks, theenemies of enemies are not necessarily enemies in negative relationship networks [10]. However, it is yet to be studiedwhether such insights derived from the analyses of social relationships are applicable to inter-organisational adversarialnetworks. We examine this by analysing the patent opposition network.Furthermore, we consider patent co-ownership among companies that have been involved in opposition. We investigatehow the companies’ strategies for “suppressing the inventions of other companies”—captured by oppositions—and“accelerating inventions”—captured by co-ownership—may interact. In social network analysis, theories have beendeveloped to discuss how positive and negative relationships might evolve together. The main theory is the structuralbalance theory [22] [5], which states that when a triad is a signed graph (i.e., a graph in which each edge has a positive ornegative sign), the structure is balanced if the multiplicative product of the signs of the edges is positive and imbalancedif this product is negative. In other words, the stability of the various triads conforms to the following simplified socialprinciples: (1) My friend’s friend is my friend, (2) my friend’s enemy is my enemy, (3) my enemy’s friend is my enemy,and (4) my enemy’s enemy is my friend. Here, positive and negative relationships are interpreted as being friendsand enemies, respectively. The balance theory suggests that balanced rather than imbalanced triads will become morefrequent over time. In other words, the number of instances of the triadic relational patterns (1)–(4) would increaseover time. However, how such balanced structure may be achieved remains insufficiently studied. Although somescholars have investigated the dynamic interplay between positive and negative relationships in online social medianetworks [31], there are enough reasons to assume that the properties of strategically formed relationships amongcompanies would be substantially different from the properties of human social interactions [29]. We apply temporaldyadic and triadic motif analysis to our data to gain insights into the dynamic formation of adversarial and collaborativerelationships in the inter-organisational network.The rest of this paper is organised as follows. Section 2 gives a detailed description of the rationale of focusing onpatent opposition and the legal background of opposition in more detail. It also explains the procedures throughwhich we collected and reformatted the data for our analysis. Sections 3 and 4 describe our main analysis of theinter-organisational patent opposition network. Section 3 investigates the global pattern of the network, while Section 4investigates how such a global pattern may have been formed by conducting the temporal motif analysis. Section 5summarises our findings and concludes the paper. 2
PREPRINT - S
EPTEMBER
10, 2020
Patent opposition is a legal means that allows third parties to challenge the validity of a patent within a given period, usually 6–9 months, after a grant. However, this period varies with jurisdiction. When an opposition is made, thevalidity of the opposed patent is decided by the patent office rather than courts. There are three possible outcomes: thepatent is maintained in its current form, amended, or revoked. Besides revocation, amendment can also be consideredas a victory for the opposing party, since it impels the narrowing of the scope of the challenged patent [23]. Making anopposition is clearly an adversarial action of a company against its competitors as it is considerably costly.While litigation is also a legal means for a company to render the patents of other companies invalid, it is substantiallydifferent from opposition. Litigation is a legal dispute that cannot not be settled by the parties involved out of courtand therefore requires adjudication [41]. The costs incurred during litigation are several orders of magnitude higherthan those during opposition. For example, in the case of the European Patent Office (EPO), oppositions relatively costaround US$45,000 [6], whereas litigation can cost above US$ 1 million [2]. Consequently, compared to opposition,litigation is an uncommon event [41], and a possible option for large companies only. Other types of business entities thatmay initiate litigation are those referred to as “trolls”. They do not engage in production or research and development(R&D) themselves, but rather acquire patents from failed companies (or independent innovators) and assert themagainst producing entities to win court judgments for profit [23]. The proliferation of trolls has recently become a majorconcern, as they cause non-negligible impact on invention [7]. However, even with legal expertise, it is very difficult todetect and block their activities.Since this study primarily aims to investigate how companies are connected via adversarial and collaborative relation-ships in patenting, we limit our focus only on patent opposition and joint ownership, but not litigation. By doing so,we can include small- and mid-sized companies that may play key roles, and exclude the influence of trolls that aresomewhat insulated from strategic interrelations among inventing companies.While legal experts and practitioners have long discussed the importance and effectiveness of the opposition pro-cesses [33, 11], economic literature has focussed more on the determinants of opposition. As the most significantdeterminant, empirical evidence converges in pointing to patent value. Simply put, the most valuable patents are morelikely to be opposed. Companies known for being innovative tend to own opposed patents far more often than onaverage [19]. It has been ascertained that the monetary value of the patent is positively correlated with the forwardcitation count, which is the number of subsequent patents that cite the patent [18, 30]. The forward citation is thusconsidered as the most effective indicator for patent value. Further, some scholars have also asserted that it significantlycontributes to increasing the opposition probability [20, 44]. These studies provide some insight into the characteristicsof opposed patents and their owners, rather than on how companies may be connected via the oppositions.In this study, we collected the data on opposed patents and owner companies, and constructed a network of companiesinterrelated through opposition. For this purpose, we obtained data from the “Orbis Intellectual Property Database(Orbis IP)” [39] and “Orbis Database (Orbis)” [38], both provided by Bureau van Dijk who is one of the most majorpublishers of business information. Orbis IP contains information on approximately 115 million patents worldwide,such as publication information, ownership, industry, history of transfer, and opposition. Orbis is a database thatcontains more than 360 million (mostly private) companies. Companies that appear in the Orbis IP database are linkedto those registered in the Orbis database via the same ID. Simultaneously using these two databases allowed us toidentify various companies and their opposition and collaboration relationships.
This section describes the method we used to acquire and pre-process data for the analysis of the inter-organisationalopposition network. We extracted data from Orbis and Orbis IP and reformatted it as follows. Figure 1 describes thedata structure. In the opposition network, nodes and directed edges represent companies and opposition relationshipsrespectively. Collaboration relationships among companies are also captured as undirected edges. Figure 2 illustratesthe rules through which we created these two types of edges.1. We extracted the list of all the oppositions made between 1980 and 2018 from the Orbis IP database. Eachopposition case was tagged to a patent ID, opposition date, name and ID of the opponent company (i.e. thecompany that made the opposition). The company ID is blank if the company is not registered in the Orbiscompany database.2. We then extracted all the patents that appeared in the list of oppositions and identified 31,470 patents. We onlyused the items listed in Fig. 1. We identified all the co-patenting companies if the patent is jointly owned by3
PREPRINT - S
EPTEMBER
10, 2020several companies. The numbers of forward citations, transfers and oppositions were counted as of the end of2018.3. We also extracted the list of events of patent transfer for the patents that appeared in the list of patentoppositions. The list contains information on the patent ownership that was transferred from which company(i.e., vendor) to which company (i.e., acquirer), and the date on which the transfer occurred (i.e., the transferdate).4. We searched all the company IDs (marked with ∗ in Fig. 1) in the Orbis company database and consequentlyidentified company pairs (opposing and opposed companies) for each patent opposition. In the case where anopposed patent was jointly owned by several companies, we created a relationship between the opponent andeach of the owner companies. In the analysis, the companies whose IDs were not found in the Orbis databasewere neglected. Through this procedure, we identified 11,480 companies and 26,433 opposition relationshipsamong them.5. We also obtained collaboration relationships among companies that were involved in opposition. For simplicity,we defined collaboration as patent co-ownership. A collaborative relationship between a pair of companiesappears when the patent co-owned by the two companies was approved (i.e., the appearing date), anddisappeared when expired or transferred to another company (or companies). Patent ownership transfer canoccur when the patent right is acquired by another company or when the owner company sells the patent right.In such cases, the original collaboration relationship disappears, and a new collaboration relationship is formedwith the new owner (or, each of the new owners in a case with more than one new owner). We identified 1,554unique collaboration relationships.6. Based on the list of inter-organisational opposition relationships (see Fig. 1), we created the oppositionnetwork, a network of companies connected via opposition relationships. In the network, nodes and directededges represent companies and opposition relationships respectively. In other words, if company A opposeda patent owned by company B, an opposition edge was created from company A pointing to company B.Additionally, we added undirected edges representing collaboration relationships.Regarding each patent’s industrial section, the information on the IPC (International Patent Classification) codewas available. The IPC represents the whole body of technical knowledge that may be considered suitable to thefield of patents for innovation, and is divided into the following eight sections: (A) HUMAN NECESSITIES, (B)PERFORMING OPERATIONS, TRANSPORTING, (C) CHEMISTRY, METALLURGY, (D) TEXTILES, PAPER, (E)FIXED CONSTRUCTIONS, (F) MECHANICAL ENGINEERING, LIGHTING, HEATING, WEAPONS, BLASTING,(G) PHYSICS, (H) ELECTRICITY. The sections are the highest level of hierarchy of the classification. We used thissection information in this study. Each patent is tagged to one of these sections. Prior to the analysis, we checked the characteristics of the opposed patents in our data. Figure 3 shows the cumulativedistributions of the numbers of forward (Figure (a))and backward citations (Figure (b)) for opposed and non-opposedpatents. Here, non-opposed patents totalled 1 million patents with no record of oppositions, sampled uniformly atrandom from the Orbis IP database. The y-axis is shown on a logarithmic scale. Figure (a) indicates that the distributionsof the number of forward citations are clearly different between the opposed and non-opposed patents. The maximumnumber of forward citations of non-opposed and opposed patents was 56 and 2,362 respectively. Although not all theopposed patents attract citations, the claim made by the literature stating that the number of forward citations indicatesthe higher likelihood of opposition seems to hold true [20, 44]. As explained in Section 2.1, a high number of forwardcitations is regarded as a sign of a high patent value. Oppositions have been made against patents with many forwardcitations (high values). This fact may differentiate the adversarial relationship based on opposition from that based onmere negative feelings (e.g., dislike) in social networks.In contrast, Figure 3 (b) shows that there is no clear difference in the distribution of the number of backward citationsbetween opposed and non-opposed patents. Some scholars reported that they did not find any impact of backwardcitations on the likelihood of an opposition [43] through their analysis of patent data from the EPO. We found that thisholds true even when patents registered at other patent offices are taken into account.
In this section, we analyse the inter-organisational patent opposition network, created via the sixth step of the proceduresdescribed in Section 2.2. In this network, the nodes and edges represent companies and opposition relationships thatexisted at some point in time, respectively. The network comprises of 11,480 nodes (companies) and 26,433 directed4
PREPRINT - S
EPTEMBER
10, 2020
Patent information : List of patent oppositions : List of patent transfers :
Patent IDOpposition dateOpponent nameOpponent ID * Patent ID List of owner companies’ ID *Application dateApproval dateExpiration dateIndustrial section
List of inter-organisationalopposition relationships:
Opposing company ID *Opposed company ID *Opposed patent ID
List of collaboration relationships:
Company ID *Company ID *Appearing dateDisappearing date
Figure 1: Data structure and data items used in the studyedges (opposition relationships). The network is a directed multigraph with no loops. In other words, two nodes maybe connected by more than one directed edge in the same direction. This happens when a company opposes multiplepatents that are owned by the same company. The number of unique pairs of nodes that were connected by one or moreedge(s) (i.e., the number of edges when disallowing multiple edges) was 14,320.
Figure 4 visualises the opposition network. There is one large connected component (displayed in the middle in thefigure) and many small connected components (composed of up to 17 nodes, displayed in the periphery of the figure).A total of 7,489 companies are found in the largest component, which contains approximately 88% of oppositionedges. In the figure, each node’s colour represents the geographical location of the company identified by the standardcountry code provided by the Orbis database. The existence of one large connected component suggests that, contraryto the expectations, opposition relationships are not bounded by the countries (or jurisdictions) in which companiesare located, or industrial sections to which the patents belong. Regarding the country, we found only 26.6% of theopposition edges that connect two companies located in the same country. The remaining 73.4% of the edges connectcompanies located in different countries. Short of the between-country edges, the largest connected component breaksdown into parts, and the largest connected component size becomes 1,617. Meanwhile, if we remove within-countryedges and maintain between-country edges, 5,410 nodes are still connected. In other words, a majority of companiesin the largest connected component have opposition relationships with those in different countries. Regarding theindustrial section, Figure 5(a) shows the scatter plot of the in-degree of companies (i.e., the number of incoming edgesthe company has) and the number of industrial sections (up to 8, as described in Section 2.2) to which its owning patentsbelong. Likewise, Fig. 5(b) plots the out-degree of companies (i.e., the number of outgoing edges of the company) andthe number of industrial sections to which the opposing patents belong. As shown in the figure, many companies ownand oppose patents in various industrial sections. 5
PREPRINT - S
EPTEMBER
10, 2020
Jointownership
CB A
OppositionPatent D Illustration of inter-organizational relationships
Opposition Relationship(directed edge)Collaborationrelationship(undirected edge)
Representation in the opposition network
AB C
Opposition Relationship(directed edge)Joint ownership CB Patent CB PatentPatent ownership transfer
Time t Time t ( t > t ) B C
From t until t B CD
Collaborationrelationship(undirected edge) (a)(b)
Illustration of inter-organizational relationships Representation in the opposition network
Collaborationrelationship(undirected edge)
After t to the expiry date of the patent Figure 2: Rules of edge formation for the construction of the opposition network
Notes: (a) The left panel explains the situation in which Company A opposes a patent jointly owned by companies Band C. This situation is represented in the opposition network as directed opposition edges formed from A to B and fromA to C, and a collaboration edge formed between B and C (illustrated in the right panel). (b) The left panel describesthe situation in which companies B and C start owning a joint patent at t , and at some time later t B’s patent rightwas transferred to another company D. This situation is represented in the network as a collaboration edge between Band C from t to t , and another collaboration edge between D and C after t until the patent’s expiry date. PREPRINT - S
EPTEMBER
10, 2020 (a)(b) N u m be r o f pa t en t s Number of forward citations
Opposed patentsNon-opposed patents N u m be r o f pa t en t s Number of backward citations
Opposed patentsNon-opposed patents
Figure 3: Cumulative distribution of the numbers of (a) forward citations and (b) backward citations for opposed andnon-opposed patents 7
PREPRINT - S
EPTEMBER
10, 2020
DE (36.9%)US (14.1%)IT (7.0%) FR (5.9%)CH (5.6%)GB (4.3%) JP (3.2%)NL (4.1%)Others (19.9%)
Figure 4: Inter-organisational opposition network.
Notes: Nodes represent companies, and edges represent opposition relationships. Regarding the visualisation, aforce-directed layout algorithm [25] was applied, making nodes that are tightly connected move close to each other.Node colour represents the country in which the company is located, identified by the country code provided by theOrbis database. The percentage (shown in brackets) is the ratio of the number of the companies in the region.
Social networks with negative relationships are usually highly disconnected and do not have any clustering [10].The cross-country and cross-industry connections among companies make the opposition network distinct from suchnetworks.
Fig. 6 shows the complementary cumulative distributions of the opposition network’s in- and out-degrees, P ( x ) , evalu-ated from the raw data. The vertical axes show the proportion of companies that have a number of incoming/outgoingedges equal to or greater than the value given on the horizontal axis. Here, multiple edges were considered. In otherwords, the distributions are equivalent to the weighted degree distribution of the network in which multiple edges aredisallowed and multiple occurrences of oppositions between a given pair of nodes are reflected as the weight of theedge. The distributions of both in- and out-degrees exhibit very heavy tails, meaning that a small number of nodes inthe network hold the majority of the edges. Figure 7 shows the distribution of multiple edges. This distribution is alsoheavy-tailed, indicating that there are node pairs between which a considerable number of oppositions occurred.8 PREPRINT - S
EPTEMBER
10, 2020 (a)(b)
Figure 5: Patents’ industrial sections and company’s (a) in- and (b) out-degree distributionsThese results may imply that “the rich get richer” (the preferential attachment mechanism) works even for the formationof adversarial relationships in the inter-organisational network, which sets the opposition network aside from socialnegative relationship networks. Existing network formation models, including negative social relationships, to the bestof our knowledge, are always based on mechanisms to balance triangles (see Ref. [32] and references within).We also calculated the degree correlation coefficient by ignoring the direction of edges [35, 36]. The degree correlationcoefficient is the Pearson correlation coefficient between the degrees found at the two ends of the same edge. This value r varies between − ≤ r ≤ : the network is assortative if r > , and is disassortative if r < . Being assortative(or assortative mixing) means that large-degree nodes tend to form edges to each other and avoid connecting withsmall-degree nodes. The examination of various real-world networks suggested that social networks (such as co-authorship networks) exhibit assortative patterns, whereas technological and biological networks exhibit disassortativepatterns [35, 4]. Further, [37] derives that assortativity is high in social networks based on the assumption that they canbe described as projections of affiliation networks. This assumption is rather strong, which explains that many empiricalsocial networks are actually disassortative. Regarding the opposition network, we obtained the degree correlationcoefficient r = 0 . , indicating its assortative pattern. In other words, the companies involved in many oppositions tendto have adversarial relationships with companies that similarly have many oppositions. Evidently and mathematically,most finite-sized heavy-tailed distribution networks (including scale-free networks) are disassortative [24], with theexception of projected affiliation networks only (e.g., co-authorship networks) [37]. Therefore, we can assert that theopposition network exhibits a rather unusual type of network topology.9 PREPRINT - S
EPTEMBER
10, 2020 -1 -2 -3 -4 Figure 6: Complementary cumulative distributions of in-degrees and out-degrees of the opposition network
Finally, we investigate the local relational patterns formed in the opposition network via the network motif analysis.Network motifs are small subgraphs that occur in a given network with significantly higher frequency than expected inthe equivalent random networks (in terms of the numbers of nodes and edges, node degree distribution, etc.). Networkmotifs are small building blocks of a network, and are crucial to understanding the structure and functions of thenetwork. Triads – subgraphs of three nodes and the potential relationships among them, particularly, are considered tobe structural foundations of social networks. Here, we mainly focus on the triadic relational motifs in the oppositionnetwork.
First, we identified triadic motifs in the cumulative opposition network – i.e., network in which directed edges representopposition relationships that existed at some point during the period of study. The presence of multiple edges wereneglected. In other words, if a company has opposed multiple patents owned by the same company, we simply concludedthe presence of a directed opposition edge between them. The network was therefore directed and unweighted. Weemployed the motif counting algorithm method implemented in the graph-tool [16], which is a widely used Pythonmodule. The algorithm (like many other motif analysis algorithms) measures the statistical significance of the occurrenceof each of the possible triadic relational patterns in a given network. The significance measure, Z-score, is defined as thedifference of the frequency of the motif in the target network and its mean frequency in a set of randomised networks,divided by the standard deviation of the frequency values for the randomised networks [34]. We used 100 randomisednetworks as null-models generated by reshuffling edges while retaining the degree sequence of the network.Figure 8 shows the results. In the figure, the identified triadic motifs (Patterns 1 to 7) are illustrated with their Z-scores.Pattern 1 (i.e., a significantly large number of companies oppose more than one company) and Pattern 2 (i.e., asignificantly large number of companies have patents opposed by more than one company) are apparent from the in- andout-degree distributions (see Fig. 6). Pattern 3 suggests that a chain of opposition relationships also occurs frequently.Companies involved in an opposition chain then tend to oppose each other too (Patterns 5 and 6).The frequent appearance of negative relationship triangles such as Pattern 4 indicates the high tendency of the occurrenceof “the enemy of my enemy is also my enemy” situation. This contradicts the low level of transitivity in social negativerelationship networks (i.e., “enemies of enemies are not necessarily enemies”) [10]. We further investigate how suchnegative triads may emerge over time. 10
PREPRINT - S
EPTEMBER
10, 2020 N u m be r o f node pa i r s Number of edges existing between a given pair of nodes
Figure 7: Destribution of the number of multiple edges
Motifwith 3 edgesZ-score 34.90 26.25 25.39 17.81Motifwith 2 edgesZ-score 91.90 82.36 45.342 31
Pattern 1
Pattern 2
Pattern 3
Pattern 4
Pattern 5
Pattern 6
Pattern 7
Figure 8: Triadic motifs identified in the opposition network
Pattern 4 can occur by adding another opposition edge to Patterns 1, 2 and 3. In other words, one can make the followingthree possible logical interpretations for the emergence of Pattern 4 (see Fig. 9): • Two companies that had been opposed by the same company formed an adversarial relationship (Pattern 4A). • Two companies that had opposed the same company formed an adversarial relationship (Pattern 4B). • A company got opposed by its opposer’s opposer (two steps away) (Pattern 4C).11
PREPRINT - S
EPTEMBER
10, 2020
Pattern 4A
Pattern 4B
Pattern 4C
Figure 9: Temporal patterns of the occurrence of Pattern 4 in Fig. 8
Note: The three open triads shown above the arrows can turn to Pattern 4 (shown below the arrows) by forming onemore edge over time.
Occurrence order of edgesNumber of instances
559 386 648 917 427 401
Pattern
4A 4B 4C1 2 31 23 31 2 31 2 1 2 3 1 23
Figure 10: Number of instances of Patterns 4A, 4B and 4C in the opposition network
Notes: The edge labels of the temporal triadic motifs correspond to the ordering of the edges. Patterns 4A to 4C (shownin Fig. 9) do not consider the ordering of the first 2 edges. For example, both the extreme left and second extreme leftmotifs impel the formation of Pattern 4A.
Such temporal patterns of the occurrence of edges can be identified by applying the temporal motif counting method.Various such methods have been proposed (e.g., [27]), with different algorithms and definitions of temporal motifs. Inour analysis, the matter of concern is the triadic closure process. In other words, we examine which patterns amongPatterns 4A, 4B, and 4C may be a more plausible logic than others. In the process of the formation of Pattern 4A, nodesmay be involved in other oppositions. For example, node 1 in Pattern 4A, shown in Fig. 9, may have opposed anothernode outside this triangle before the formation of the final edge between nodes 2 and 3. Otherwise, node 1 may haveopposed node 2 repeatedly. We considered all of such cases contributing to the emergence of Pattern 4A. In other words,our aim was to ensure that every occasion that edges form a particular pattern within the observation period is counted.Therefore, we employed the method provided by the Stanford Network Analysis Platform (SNAP) [40, 42], whichserves this purpose. Figure 10 shows the number of instances (i.e., the actual counts) of the occurrence of each temporaltriadic pattern, counted using this method. Each of Patterns 4A, 4B, and 4C can occur in two ways if the ordering of thefirst 2 edges is taken into account. Thus, the number of instances of Pattern 4A, for example, is the sum of 559 and 386(= 945).We found that, compared to Pattern 4A (945 times) and Pattern 4C (827 times), Pattern 4B occurred considerably morefrequently. In other words, two companies that have opposed the same company tend to form an adversarial relationship.This may imply that companies that have a common enemy share technological interests and thus are prone to becomingrivals. Indeed, the assertion that enemy of my enemy may also be my enemy holds trues.
In the opposition network, we concluded that the enemies of my enemy tend to be my enemies. Nevertheless, companiesalso have collaborative relationships. In this section, we investigate where in the opposition network collaborationrelationships occur and how the collaboration relationships would interact with opposition relationships.12
PREPRINT - S
EPTEMBER
10, 2020First, we focus on dyads. We consider a relationship between two companies adversarial when one company opposesthe other. They may form a reverse opposition edge (as shown in Fig. 8 for the case of triads). Otherwise, theirrelationship remains adversarial (or may become more neutral over time). In general, the relationship between a pair ofcompanies does not alternate between collaboration and rivalry. However, we found some cases in which an adversarialrelationship turned to a collaborative relationship. We counted the number of instances in which an opposition edgeappeared at time t and a collaboration edge appeared between the same pair of nodes at t ( t < t ). There are 14,320unique pairs of nodes connected via one or more opposition edges. Among those, 374 pairs formed a reverse oppositionedge, while 9 pairs formed a collaborative relationship later. Although very uncommon, it is possible for companies totransition from an adversarial to a collaborative relationship. Meanwhile, we found 14 out of the 1,554 unique pairs ofnodes that were in a collaborative relationship having formed an opposition relationship later in time. Companies veryrarely change their relationships with others, from rivalry to collaborative or vice versa. Further, if nothing else, there isa higher tendency for collaboration to turn to rivalry.Second, We moved onto the triads. As already discussed, Pattern 4 is formed by the triadic closure of Patterns 1–3 (i.e.,by adding an opposition edge between the unconnected pair of nodes), all of which evidently occur more frequentlythan expected (see Fig. 8). Pattern 4 is structurally imbalanced. The triangle becomes structurally balanced if thethird edge added to Patterns 1–3 is a collaboration rather than an opposition edge. We further counted the temporaloccurrence of such balanced triads while focussing on whether a collaboration edge was formed before or after twoopposition edges. Figure 11 summarises the results. Pattern 1 with a collaboration edge (hereafter, Pattern 1’) canoccur in the following two ways: two opposition edges occur before a collaboration edge, or vice versa, as shown in (a)and (b). Likewise, (c) and (d) are for Pattern 2 with a collaboration edge (hereafter, Pattern 2’), and (e) and (f) are forPattern 3 with a collaboration edge (hereafter, Pattern 3’), respectively. The number of each pattern’s occurrences isshown in the bottom row of the figure. For example, when a company had opposed the two companies, the case wherethese two companies started collaborating at any point in time after that occurred 331 times (see the panel (a)).As the figure suggests, Pattern 1’ occurred considerably more frequently (331 + 605 = 936 times) than Patterns 2’and Pattern 3’. In other words, the situation in which two rivals –or “enemies”– of a company collaborate together(= Pattern 1’) is more likely to occur compared to the situation in which collaborating companies share a commonrival (= Pattern 2’) or a rival of a rival of a company is its collaborator (= Pattern 3’). When we looked further into theemergence of Pattern 1’, the pattern (a) occurred considerably less frequently than the pattern (b). This suggests thatbeing opposed by the same company (i.e., node 1 in (a)) does not necessarily make two companies (i.e., nodes 2 and3) sharing technological knowledge or a common interest, which is reasonable given that many companies engage inpatenting in multiple industrial sections.In sum, we did not observe any sign of interactions between collaboration and opposition relationships. The mechanismfor the emergence of opposition relationships is clearly different from that of collaboration relationships, underliningthe necessity of further studies on the opposition network. We constructed the opposition network in which nodes represent companies and directed edges represent oppositionrelationships. This study is the first to map the topology of the inter-organisational patent opposition network. Thisstudy is novel not only as a study of patents but also as a study of networks with negative relationships.The results of our analysis elucidated the characteristics of this network that are distinct from other negative relationshipnetworks discussed in the social network analysis. A large number of companies are interconnected, suggesting thatthe formation of opposition relationships is not bounded by the country or industrial section. The coexistence of aheavy-tailed degree distribution and the assortative mixing by degree makes the opposition network topology a raretype, highlighting the need to be careful when applying knowledge derived from analyses of social negative relationshipnetworks. This need was further enhanced by the identification of structurally imbalanced triads overrepresented inthe network. The disagreement between the significant local structural patterns in the opposition network and thosein social networks suggests that the mechanisms of the emergence of this network would differ from social negativerelationship networks. Fundamentally, inter-organisational relationships are formed strategically. One feature thatcharacterises opposition relationships is that a high probability of opposition indicates a high value of a patent. Aninter-organisational adversarial relationship captured by patent opposition may, therefore, be considerably differentfrom social negative relationships, such as inter-personal dislike.The future direction of this research is to further investigate how opposition networks are formed and relate the identifiednetwork characteristics to the knowledge on patenting strategies. A more detailed analysis of patent attributes (e.g.,industrial sections, registered patent office, patent family, closely related inventions, etc.) is also necessary in order toenrich the understanding of the complex network of the patenting strategies of companies.13
PREPRINT - S
EPTEMBER
10, 2020
Temporal patternNumber of occurrence
331 605 26 46 42 9 (a) (b) (c) (d) (e) (f)
Figure 11: Temporal patterns of the occurrence of triads with two opposition relationships and one collaborationrelationship
Notes: Each panel corresponds to a temporal triadic pattern formed by two opposition edges and one collaborationedge that occurs in a specific order. The occurrence order of the two opposition edges was disregarded.
Funding
This work was supported in part by the JSPS KAKENHI (19K04893).
References [1] Acemoglu D., Akcigit U., Kerr W.R. (2016). Innovation network.
PNAS , 113: 41, 11483–11488.[2] Vermont, S. (2017). Survey of costs of patent litigation and inter partes review.
Pat. Att’y
Res. Policy , 33 (1), 127–145.[4] Barabási, A.-L. (2016). Network Science. Chapter 7, Cambridge University Press.[5] Cartwright, D., Harary, F. (1956). Structural balance: A generalization of Heider’s theory.
Psychol. Rev. , 63:277–292.[6] Chien, C., Helmers, C., Spigarelli, A. (2019). Inter partes review and the design of post-grant reviews.
BerkeleyTech. L. J. , 33 (3), 817.[7] Cohen L., Gurun, U., Kominers, S.D. (2017). Empirical Evidence on the Behavior and Impact of Patent Trolls: ASurvey, in
Daniel Sokol (ed.), Patent Assertion Entities and Competition Policy , Cambridge, Cambridge UniversityPress, pp. 27–59.[8] Duffy, M., Ganster, D., Pagon, M. (2002). Social undermining in the workplace.
Acad. Manage. J. , 45 (2), 331–351.[9] Érdi, P., Makovi, K., Somogyvári, Z. et al. (2013). Prediction of emerging technologies based on analysis of the USpatent citation network.
Scientometrics , 95, 225–242.[10] Everett, M.G., Borgatti, S.P. (2014). Networks containing negative ties.
Soc. Networks , 38, 111–120.[11] Farrell, J., Merges, P. (2004). Incentives to challenge and defend patents: Why litigation won’t really fix patentoffice errors and why administrative patent review might help.
Berkeley Tech. L. J. , 19(1), 1—28.[12] Fleming, L., Mingo, S., Chen,D. (2007). Collaborative brokerage, generative creativity, and creative success.
Adm.Sci. Q , 52: 443–475. 14
PREPRINT - S
EPTEMBER
10, 2020[13] Funk, R. J., Owen-Smith, J. (2017). A dynamic network measure of technological change,
Manag. Sci. , 63 (3),791–817.[14] Gao, Y., Zhu, Z., Kali, R. et al. (2018). Community evolution in patent networks: technological change andnetwork dynamics.
Appl. Netw. Sci. , 3: 26.[15] Graf, H. (2011). Gatekeepers in regional networks of innovators.
Camb. J. Econ.
35 (1), 173–198.[16] graph-tool: https://graph-tool.skewed.de/ (accessed May 21, 2020).[17] Grzegorczyk, T., Głowi´nski. (2020). Patent management strategies: A review.
J. Econ. Manag. , 40 (2) 36–51.[18] Harhoff, D., Narin, F., Scherer, F.M., Vopel, K. (1999). Citation frequency and the value of patented innovation.
Rev. Econ. Stat . 81 (3), 511–515.[19] Harhoff, D., Scherer, F.M., Vopel K. (2002). Citations, family size, opposition and the value of patent rights.
Res.Policy , 32, 1343–1363.[20] Harhoff, D., Reitzig, M. (2004). Determinants of oppositions against EPO patent grants: The case of biotechnologyand pharmaceuticals.
Int. J. Ind. Organ. , 22, 443–480.[21] Hasan, I., Tucci, C.L. (2010). The innovation-economic growth nexus: Global evidence.
Res. Policy , 39 (10),1264–1276.[22] Heider, F. (1946) Attitudes and cognitive organization.
J. Psychol. , 21: 107–122.[23] Helmers, C. (2018). The economic analysis of patent litigation data.
Economic Research Wroking Paper, WIPO(World Intellectural Property Organization) , 48.[24] Holme, P., Zhao, J. (2007). Exploring the assortativity-clustering space of a network’s degree sequence.
Phys. Rev.E , 75, 046111.[25] Jacomy, M., Venturini, T., Heymann, S., Bastian, M. (2014). ForceAtlas2, a continuous graph layout algorithm forhandy network visualization designed for the Gephi software.
PLOS ONE , 9(6): e98679.[26] Kogan, L., Papanikolaou, D., Seru, A., Stoffman, N. (2017). Technological innovation, resource allocation, andgrowth.
Q. J. Econ. , 132 (2), 665–712.[27] Kovanen, L., Karsai, M., Kaski, K., Kertész, J., Saramäki, J. (2011). Temporal motifs in time-dependent networks
J.Stat. Mech. , P11005.[28] Labianca, G., Brass, D. (2006). Exploring the social ledger: negative relationships and negative asymmetry insocial networks in organizations.
Acad. Manage. Rev. , 31 (3), 596–614.[29] Labianca, G.J. (2014). Negative ties in organizational networks.
Contemporary Perspectives on OrganizationalSocial Networks (Research in the Sociology of Organizations, Vol. 40) , Emerald Group Publishing Ltd., 239–259.[30] Lanjouw, J.O., Schankerman, M. (2001). Characteristics of patent litigation: a window on competition.
RAND J.Econ. , 32 (1), 129–151.[31] Leskovec, J., Huttenlocher, D., Kleinberg, J. (2010). Signed networks in social media.
Proc. ACM SIGCHI Conf.on Human Factors in Computing Systems (CHI) , 1361–1370. Atlanta, GA.[32] Marvel, S. A., Kleinberg, J., Kleinberg, R.D., Strogatz, S.H. (2011). Continuous-time model of structural balance,
Proc. Natl. Acad. Sci. USA , 108 (5) 1771–1776.[33] Merges, R.P., Nelson, R.R. (1990). On the complex economics of patent scope.
Colum. L. Rev. , 90 (4), 839–916.[34] Milo, R., Shen-Orr, S., Itzkovitz, S., Kashtan, N., Chklovskii, D., Alon, U. (2002). Network motifs: simplebuilding blocks of complex networks.
Science
Phys. Rev. Lett. , 89: 208701.[36] Newman, M. E. J. (2003). Mixing patterns in networks.
Phys. Rev. E , 67: 026126.[37] Newman, M.E.J., Park, J. (2003). Why social networks are different from other types of networks.
Phys. Rev. E
WSDM ’17: The Proceedings of theTenth ACM International Conference on Web Search and Data Mining . February 2017, 601–610.15
PREPRINT - S
EPTEMBER
10, 2020[41] Somaya, D. (2012). Patent strategy and Management: An Integrative Review and Research Agenda.
J. Manag. , 38(4) 1084–1114.[42] Stanford Network Analysis Platform (SNAP):https://snap.stanford.edu/index.html (accessed April 8, 2020).[43] Sterlacchini, A.(2016). Patent oppositions and opposition outcomes: evidence from domestic appliance companies.
Eur. J. Law. Econ. , 41, 183–203.[44] van Zeebroeck, N., van Pottelsberghe de la Potterie, B. (2011). Filing strategies and patent value.
Econ. Innov.New Technol. , 20(6), 539–561.[45] Venkataramani, V., Labianca, G.J., Grosser, T. (2013). Positive and negative workplace relationships, socialsatisfaction, and organizational attachment.
J. Appl. Psychol. , 98 (6), 1028.[46] Verspagen, B. (2007). Mapping technological trajectories as patent citation networks: A study on the history offuel cell research.
Adv. Complex Systems , 10 (1), 93—115.[47] Youn, H., Strumsky, D., Bettencourt, L.M.A., Lobo, J. (2015). Invention as a combinatorial process: evidencefrom US patents.