Towards an understanding of global ‘private ordering’ in ICANN: text mining 23 years of Uniform Domain-Name Dispute-Resolution Policy (UDRP) Decisions

ABSTRACT

To assess the prospect of the ICANN Uniform Domain-Name Dispute Resolution Policy (UDRP) enabling global ‘private ordering’ for domain-name disputes, this study analyses textual data from 75,590 UDRP complaints involving 142,423 domain names. Using data provided by DNS Research Federation’s Data Analytics Platform (DAP.Live), we ask three major research questions: To what extent does the UDRP process differ between resolution bodies? What are the most prevalent themes as represented by keywords and topics? To what extent have these topics changed over time? Using descriptive statistics and a series of inductive text-mining techniques (term-frequency, term frequency-inverse document frequency, and topic modelling), we find substantial evidence for the ongoing stability of the UDRP. Case growth has continued since 2000. There is strong global support for two of the six DRSPs, WIPO and NAF. Average decision time varies substantially by DRSP with WIPO at 63 days and CAC at 36. Panelists heavily employ precedent when adjudicating complaints. Trademark holders continue to dominate the process, winning about 90% of complaints; however, successfully contested cases show strong UDRP jurisprudence supporting non-trademark holders. Topic models created capture both abstract (jurisprudential) and concrete (cybercrime) concepts and show spikes in cybercrime during COVID-19.

KEYWORDS:

• Text-mining
• private ordering
• ICANN
• UDRP
• internet governance
• DAP.Live

Introduction

Domain names are an easy-to-remember textual identifier at the heart of the internet’s Domain Name System (DNS). These domain names are used to describe the online address of a website that one visits. For example, ‘american.edu’ is the text used by our academic home, The American University in Washington, DC. That domain name is linked to the numeric internet protocol (IP) address, 147.9.2.4, which indicates the location of the computer serving up data for that specific domain name at the university. Each of these domain names must be unique in a top-level domain. Thus, in the above example, there can only be one domain name with the specific text: ‘american’ in the.edu top-level domain. The uniqueness of domain names makes them a scarce resource. As a result, global competition for them is fierce.

With 351.5 million domain name registrations in circulation and an annual increase of 10.4 million domain name registrations, the challenge of determining the appropriate allocation of rights to specific domain names persists (VeriSign Inc. 2022). The United States Government, the Internet Corporation for Assigned Names and Numbers (ICANN), and several scholars have called this challenge the ‘Trademark Dilemma’ (Freire and Kokjohn 2017; ICANN 2000; Kur 1998; Loutocký 2015; NTIA 1998). The two horns of this trademark dilemma refer to who should own this unique resource – should it be whoever registers the domain name first, or a ‘legitimate’ trademark holder?

Since 1999, the Uniform Domain-Name Dispute-Resolution Policy (UDRP) process, adopted by ICANN and put into active practice in 2000, has been at the core of managing this trademark dilemma. The UDRP has resolved more than 75,590 disputes involving more than 142,423 domain names. In this paper, we examine the role of the UDRP in managing this trademark dilemma from the perspective of ‘private ordering’ and what its continued existence means for the global governance of this critical aspect of the DNS.

We approach the concept of ‘private ordering’ by drawing upon our expertise studying international regime formation processes for high technology, including the internet (Cogburn 2016; 2017). We see the UDRP as central to maintaining a ‘strong’ global regime for maintaining the stability of the DNS. This regime follows Krasner (1982) and Cowhey (1990) in promulgating an interlocking set of norms, principles, values, and decision-making processes. The UDRP also has substantial confiscatory powers (meaning its ability to strip a domain holder of the rights to that domain if they lose in the dispute resolution process) which leads us to see the UDRP at the centre of this strong domain-name dispute regime.

The UDRP and managing the ‘trademark dilemma’

At a high level, the trademark dilemma refers to the challenge of ‘registering domain names to those that first claim them while respecting trademark law’ (Freire and Kokjohn 2017; Kur 1998; NTIA 1998). According to ICANN, concerns over the trademark dilemma first began with the realisation that a trademark being used as a domain name without the trademark owner’s consent could have far-reaching implications for consumers who might be misled about the source of the product or service offered on the internet (ICANN 2000). For trademark owners, a trademark used as a domain name without their consent would also erode their ability to protect their rights without incurring high litigation fees (ICANN 2000). These initial concerns sparked a shared consensus among early internet governance advocates and stakeholders that for an effective commercial market to have any chance of thriving reliably in cyberspace, business owners must have confidence that their trademarks would be sufficiently protected (ICANN 2000).

Furthermore, determining the allocation of rights to unique domain names is important because these rights also play a critical role in enhancing cybersecurity measures against nefarious actors who may attempt cybersquatting or reverse domain name hijacking campaigns (Burton and Lain 2020; Crawford and Sherman 2018; Falco, Noriega, and Susskind 2019; Kulikova 2021; Mueller 2001; Springer 2001; Woodard 2009). Cybersquatting is a term coined to describe the process of registering trademarked domain names to gain a profit by selling the registered domain names back to existing trademarks and business competitors (Mueller 2001). However, benign cybersquatting can also occur when an individual or organisation unintentionally attempts to use a domain name that already exists. When left unchecked, cybersquatting, whether benign or malicious, threatens the health of online commerce because it allows people to ‘mislead consumers about the source of the product or service offered on the Internet … ’ (Woodard 2009).

In response to these emerging issues, on 1 July 1997 President Bill Clinton directed the Secretary of Commerce to privatise the DNS in a manner that ‘increases competition and facilitates international participation in its management’ (ICANN 2000; Kuerbis and Mueller 2017; Nakayama 2022). ICANN was also created in 1998 to oversee and ‘perform technical coordination of core internet resources, most notably, domain names’ (Klein 2002). ICANN is a ‘private, non-profit entity’ that makes ‘effective internet governance possible for the first time [by realising] the governance potential in the [DNS] and leveraging the internet to achieve global governance’ (Klein 2002; Mueller 1999).

One of the core tasks assigned to ICANN upon its establishment was developing a solution to the trademark dilemma (ICANN 2000). To meet this core objective, ICANN collaborated with the World Intellectual Property Organization (WIPO) to develop the UDRP process. The collaboration between ICANN and WIPO yielded a Final Report focused on the tension between domain names and trademarks and included the recommendation to create the UDRP and its rules and processes (WIPO 1999). Specifically, the report recommended the establishment of a ‘mandatory administrative procedure concerning abusive registrations’, which would allow for a ‘neutral venue in the context of disputes that are often international in nature’ (WIPO 1999). On 24 October 1999, the ICANN Board adopted the UDRP, and on 1 December 1999, it was launched (Freire and Kokjohn 2017).

The UDRP stipulates that ‘most types of trademark-based domain-name’ disputes must be resolved by agreement, court action, or arbitration before a registrar who will cancel, suspend, or transfer a domain name (ICANN 1999). In other words, any registrant who registers a domain name with ICANN is agreeing to the terms and conditions outlined in the UDRP process and must bring any disputes over domain names through the UDRP process first. In this way, the UDRP process ensures that domain name rights are appropriately allocated, thereby protecting the interests of consumers and trademark owners.

To invoke the UDRP, a complainant should ‘either (a) file a complaint in a court of proper jurisdiction against the domain-name holder (or where appropriate an in-rem action concerning the domain name) or (b) in cases of abusive registration submit a complaint to an approved dispute-resolution service provider’ (ICANN 1999). All disputes are decided using a three-prong test outlined in the UDRP (1999) which is as follows:

1. Identical or Confusingly Similar: a complainant has to prove that the disputed domain name is identical or confusingly similar to a trademark and service mark in which it has rights.

2. No Rights or Legitimate Interests: a complainant has to prove that the domain name registrant has no rights or legitimate interests.

3. Bad Faith: a complainant has to prove that a domain name has been registered and is being used in bad faith.

After a complaint has been filed, the ‘provider then forwards the complaint to the domain name registrant, who has twenty days to [respond]’ (Sorkin 2002).

In addition to the three-prong test, all UDRP disputes are decided by a ‘panel comprised of one or three members; unless one of the parties elects a three-member panel, a single panelist is appointed’ (Sorkin 2002). Every provider maintains a list of qualified panelists from which the sole or lead panelist is selected (Sorkin 2002). For disputes where a ‘three-member panel is to be appointed, each party may nominate prospective panelists drawn from any provider’s list, and the provider administering the proceeding will attempt to appoint a panelist nominated by each party’ (Sorkin 2002). Panels are often comprised of ‘experienced intellectual property attorneys, academics, or (in the case of the NAF) retired judges’ (Sorkin 2002).

Within ‘fourteen days of its appointment’, the panel decides on the dispute based on the ‘statements and documents submitted, applying any rules of law that it deems applicable’ unless there are extenuating circumstances (Sorkin 2002). The panel may decide to cancel, suspend, or transfer the domain name (ICANN 1999). All panel decisions are ‘communicated to the parties and subsequently published on the provider’s website’ (Sorkin 2002). Domain name registrants have ‘ten business days to implement an order from the panel to transfer or cancel a domain name’ (Sorkin 2002). However, the domain name registrant may also

suspend the implementation of an order from the panel to transfer or cancel a domain name by providing documentation of how it has filed a lawsuit against the complainant in the mutual jurisdiction previously specified by the complainant within ten business days of the released decision. (Sorkin 2002)

By serving as the primary avenue for stakeholders to bring disputes over domain name registration rights, the UDRP process helps to balance conflicting interests, such as stability versus openness or consumer protection versus open innovation. Since its creation and adoption, more than 75,000 UDRP decisions have interpreted and applied the language stipulated in the UDRP process, producing an extensive body of precedence. The following sub-section discusses the UDRP process as an instance of private ordering.

Understanding the UDRP as ‘private ordering’

When disputes over domain names are brought to one of ICANN’s approved Dispute-Resolution-Service Providers (DRSPs), the panelists overseeing the dispute have the authority to cancel or transfer a domain name if a complainant proves that the domain name meets the elements of the three-prong test (Mueller 2001; Woodard 2009). However, because the UDRP process was established and is managed by ICANN, a non-governmental entity, the process has been criticised by some sceptics who express concerns over its authority to order the transfer or cancellation of domains as a private ordering body.

Private ordering is defined as a ‘system of rules that private actors develop, examine, and impose beyond the authority of the law’ (Contreras 2017). In other words, private ordering occurs when private actors share regulatory authority and influence (Schwarcz 2002). Private ordering is deeply rooted in historical and legal precedent (Eisenberg 1976; Katz 1996; Milhaupt and West 2000; Richman 2004; Schwarcz 2002). Throughout history governments have relied on private ordering to help govern commercial, financial, and business sectors. For example, the U.S. entrusted the power to promulgate public accounting standards to the Financial Accounting Standards Board (FASB), a privately organised, but independent entity (Schwarcz 2002). Similarly, and most relevant to this discussion, the United States (U.S.) delegated the DNS and the assignment of IP numbers to ICANN, a private nonprofit corporation (Schwarcz 2002).

While the UDRP is frequently cited in discourse and research about private ordering and domain name disputes, it is important to also acknowledge the existence and roles of alternate domain name dispute resolution (abbreviated as ‘ADDR’) policies. Most notably, the EU Alternate Dispute Resolution (euADR),.au Dispute Resolution Policy (.auDRP).IT (Modified UDRP),.NL (Modified UDRP),.U.K. (Nominet Dispute Resolution Service),.CN (CNDRP),.U.S. (usDRP), and.JP (JP-DRP). Each serves as ADDR bodies to the UDRP process (Publications Office of the European Union 2018).

One of the perceived benefits of filing a complaint with one of these ADDRs is the tailored fit and context to the country’s unique culture, economic and political landscape (Publications Office of the European Union 2018). Each of these ADDR entities can define their own terms, conditions and domain name dispute resolution policies that are more aligned with the law enforcement standards of their country (Publications Office of the European Union 2021). Although these ADDR bodies play a significant role in addressing the challenges of determining domain name rights, their localised scope means they are not as internationally recognised as the UDRP, thus may not be as relevant in multinational, multijurisdictional disputes (Publications Office of the European Union 2018).

Due to the transnational and borderless nature of cyberspace, private ordering was initially viewed by scholars and stakeholders as a promising solution to internet governance because it could encourage a process of ‘tacit coordination [that] can create and maintain technological standards, when the conditions – which we don’t fully understand and must study – are right’ (Radin and Wagner 1998). Proponents of private ordering often point to the benefits, such as efficiency, reducing the cost of regulation, and efficacy, as proven through its wide adoption throughout history (Benkler 2000; Katz 1996; Radin and Wagner 1998; Schwarcz 2002). In contrast, critics of private ordering express concerns over its ideological and legal ambiguities (Eisenberg 1976; Elkin-Koren 2005; Macey 1997; Milhaupt and West 2000; Williamson 2002). These concerns extend to the legitimacy of private ordering, its ability to enforce decision outcomes, lack of accountability and oversight, ambiguous transparency, and potential for biases or conflicts of interest (Klein 2004; Radin and Wagner 1998; Schwarcz 2002).

Another way to think about private ordering is to consider the conceptual approach known as international regime theory, supported by hegemonic stability theory. Although it builds upon a longer theoretical trajectory, regime theory is largely credited to Stephen Krasner (1982) and was expanded to telecommunications and high technology regimes by Peter Cowhey (1990). Regime theory sees the global governance of transnational phenomena accomplished through dense networks created as actors interact on the world stage to create norms, rules, principles, values and decision-making procedures. The regular ICANN meetings, along with other broader cyber policy meetings such as the annual UN Internet Governance Forum (IGF), help to facilitate the DNS regime elements. The UDRP has its own norms, rules, principles, values and decision-making procedures. Following hegemonic stability theory an international regime is seen as ‘strong’ when it is supported by a powerful hegemon, or dominant world actor, and includes enforcement mechanisms. During the formation of ICANN and the UDRP up to the transition of the Internet Assigned Numbers Authority (IANA) stewardship functions to the international community in September 2016, the U.S. served as this dominant hegemon (Cogburn 2016; Gilpin 1981; ICANN, n.d.; Internet Society 2022; Keohane 1984; Keohane and Nye 1977; Krasner 1983; NTIA 2016). The central role and active engagement of the U.S. in the process of creating ICANN and the UDRP helped stabilise this emerging international regime for the DNS; however, the process still takes place in a non-governmental, ‘private’ structure.

Finally, the UDRP has very aggressive enforcement of its decisions on disputes, as the loser of a dispute must forfeit ownership of the disputed domain name. The combination of these elements helps us to conceptualise the UDRP as a strong international regime for domain-name dispute resolution. This dispute resolution process happens through DRSPs, which will be detailed in the following section.

The key role of DRSPs

The UDRP process is often lauded by internet governance stakeholders for the strides it has made in showing that it is possible for an international, non-profit and privatised entity, such as ICANN, to facilitate a dispute process over domain names that encourages participation from countries around the world in a manner oriented towards multi-stakeholderism (Hofmann 2016; Lazanski 2019; Moynihan 2021; Norton 2012; Sahel 2016; Strickling and Hill 2017). Central to this process are the six active ICANN-approved DRSPs (ICANN 2012), which at the time of writing are as follows:

1. Czech Arbitration Court Arbitration Center for Internet Disputes (CAC);

2. World Intellectual Property Organization (WIPO);

3. Canadian International Internet Dispute Resolution Centre (CIIDRC);

4. National Arbitration Forum (NAF);

5. Asian Domain Name Dispute Resolution Centre (ANDRC); and

6. Arab Center for Domain Name Dispute Resolution (ACDR).

Two other DRSPs, eResolutions and CPR: International Institute for Conflict Prevention and Resolution Dispute Resolutions, were closed down (ICANN 2012). When deciding to file a complaint, complainants have complete freedom to choose between these DRSPs showing the UDRP was designed with accessibility in mind (ICANN 2012; Norton 2012).

Due to the privatised nature of ICANN, the UDRP process is also praised for increasing both the legitimacy and transparency associated with the dispute process (Mueller 2001; Norton 2012). Two other strengths of the UDRP process are its relatively inexpensive and swift nature in resolving disputes (Mueller 2001). Although the costs of proceedings can vary between each of the six ICANN-approved venues, a typical dispute over ‘one and five domains costs around $1,500 for adjudication before a panel of one, and $4,000 before a panel of three’ (Norton 2012). In contrast, a dispute ‘involving six to ten domains costs $2,000 before a panel of one, and $5,000 before a panel of three’ (Norton 2012). On average, panelists decide within ‘sixty days of the complaint’s filing date’ (Norton 2012). In this way, the UDRP is regarded by stakeholders as a relatively successful (though imperfect) attempt at providing an early solution to the unprecedented global nature of the domain name rights issue in internet governance.

Critiques of the UDRP process: bias towards trademark holders?

Despite the UDRP’s strengths, critics argue it suffers from three key flaws: (1) biases in favour of trademark owners; (2) lack of an internal appellate body; and (3) affords panelists too much freedom (Woodard 2009). Another prominent debate over the strength and validity of the UDRP is the extent to which venue-shopping occurs and affects the perceived legitimacy of decisions (Beyers and Kerremans 2012; Norton 2012). Other studies contend that the UDRP’s current scope is too narrow and should be expanded to include disputes about trade names, geographical indications, and country codes (Samuels and Samuels 2003; Sharrock 2001). In this section, we address some of these criticisms, beginning with the concern that it is biased in favour of trademark owners. A particularly stringent criticism along these lines is that trademark interests set up a system benefiting them and continue to reap those benefits.

It is true that to help create a mechanism to support their global domain name and trademark interests, the international business community engaged actively in the formation of both ICANN and the UDRP.

In ICANN’s formation, the International Chamber of Commerce (ICC) and nine other business-oriented organisations participated in the formation of the Domain Name Supporting Organization (DNSO), these were: Electronic Commerce Europe (ECE), European ISP Association (EuroISPA), Information Technology Association of America (ITAA), International Chamber of Commerce (ICC), International Council of Registrars (CORE), International Trademark Association (INTA), Internet Society (ISOC), Policy Oversight Committee (POC), World Information Technology and Services Alliance (WITSA) and American Intellectual Property Law Association (AIPLA) (Electronic Commerce Europe et al. 1999).

And in the process that developed the WIPO Final Report which recommended the structure, rules and processes of the UDRP, WIPO was served by an expert advisory panel that included members from the business community along with other governmental and non-governmental constituencies (WIPO, n.d.(a)). The WIPO process also received inputs directly from companies (n = 181), which were the largest constituency involved in consultation with WIPO (compared to states (n = 28)), inter-governmental organisations (n = 5), non-governmental organisations (n = 74), with the only exception being individuals (n = 183) (WIPO, n.d.(b)). The companies involved were well-known U.S.-based (e.g. America Online, AT&T, Eli Lilly and Company, Ford Motor Company, Microsoft Corporation and Time Warner) and global (e.g. British Telecommunications, Deutsche Telekom, Singapore Telecom and Telstra). In addition, numerous trademark associations participated including: American Society of Composers, Authors and Publishers (ASCAP), Association of European Brand Owners (MARQUES), Brazilian Intellectual Property Association (ABPI), International Intellectual Property Alliance (IIPA), International Trademark Association (INTA), Japan Intellectual Property Association (JIPA) and Motion Picture Association of America (MPAA). It is reasonable to assume that this substantial participation by trademark interests helped to influence the structure of the UDRP to meet their needs. However, this corporate participation was counterbalanced with substantial participation by key broad-based non-governmental organisations, including: Association for Computing Machinery (ACM), Center for Democracy and Technology (CDT), Computer Professionals for Social Responsibility (CPSR), Electronic Frontier Foundation (EFF), International Congress of Independent Internet Users (ICIIU) and several universities, such as Michigan, Stanford, Fordham and Tamkang.

Another notable critique of the UDRP is that it is too ambiguous and should not be considered an arbitration process, but rather a new procedure (Komaitis 2005; Norton 2012; Woodard 2009). The UDRP does depart from the accepted characteristics of arbitration in a few ways. First, participation in UDRP proceedings is ‘mandatory for domain name registrants but optional for trademark owners (as they are not bound by contracts with domain name registrants), who may choose to take their trademark or related claims directly to court’ (Sorkin 2002). In these cases, some scholars argue that UDRP proceedings are biased because trademark owners can ‘select the timing and [DRSP] and have a much greater influence than the domain name registrant over the selection of the panel’ (Sorkin 2002). While losing trademark owners may ‘seek legal remedy at leisure as an adverse award merely maintains the status quo, losing domain name registrants have only 10 days after the decision (and quite possibly less than two months after receiving the initial notice of the dispute) to prepare and file a lawsuit, possibly in a remote jurisdiction, to prevent the immediate loss of its domain name’ (Sorkin 2002).

Moreover, because a losing domain name registrant can ‘block implementation of a cancellation or transfer order by filing a lawsuit following the decision or (depending on the panel) merely file a second UDRP complaint’, UDRP decisions are non-binding (Sorkin 2002). Finally, UDRP proceedings are ‘conducted almost exclusively via documents, with live hearings being rare or nonexistent’ (Sorkin 2002). In this way, the non-binding and ambiguous nature of the UDRP are traits that disqualify it from being considered as a true example of arbitration because it leads complainants and respondents to uncertainty, lack of enforcement, and inconsistency even after a complaint has been decided (Emerson 2004; Kelley 2002; Komaitis 2005; Lovells 2017; Sharrock 2001; Sorkin 2002; Weinberg 2000; Zalnieriute 2020). For these reasons, even if the UDRP process were to brand itself as a new procedure unique from arbitration, some scholars maintain that it would still fail to promote a fair discovery process, a rigorous ‘hearing’ process and unprejudiced procedural rules (Komaitis 2005; Norton 2012; Woodard 2009).

To address these critiques, previous studies have proposed distinguishing the UDRP process as a new, independent process that is not equal to arbitration (Komaitis 2005). Critics also propose that the current vulnerabilities in the UDRP process could be addressed and remedied if ICANN chose the DRSP and created a UDRP oversight body (Woodard 2009; Zalnieriute 2020). Furthermore, dissenters argue that the UDRP should provide clearer legal guidelines and consider limiting the publication of UDRP decisions by each forum to prevent the issue of venue shopping by complainants (Woodard 2009; Zalnieriute 2020).

Despite these critiques, the UDRP still holds the authority to order the confiscation or transfer of a domain name from one party to another following a decision (ICANN 2015; Isenberg 2022). While the UDRP does not define how domain names should be transferred or confiscated, it does allot a 10-day waiting period after a decision to allow the losing domain name registrant to appeal (Isenberg 2022). Through ICANN’s registrar accreditation agreement (RAA) and its established relationship with the UDRP, ICANN can further pressure registrars to fulfil their contractual obligations to abide by the UDRP’s decisions if a registrant is noncompliant (Isenberg 2022). In this way, the UDRP can also be viewed as having aggressive enforcement of its decisions on disputes because there is a clearly defined outcome and expectation issued through the UDRP’s decisions on the disputes (i.e. transfer, confiscate, or maintain the status quo of the domain name ownership) and a process that allows for following up with registrars who have not met that expectation.

Consideration must also be given to the evolving nature of the first element in the UDRP’s three-prong test – identical or confusingly similar. In more recent years, WIPO has evolved the scope of assessing whether a disputed domain name is identical or confusingly similar to a trademark and service mark by noting that

while each case is judged on its own merits, in cases where a domain name incorporates the entirety of a trademark, or where at least a dominant feature of the relevant mark is recognizable in the domain name, the domain name will normally be considered confusingly similar to that mark for purposes of UDRP standing. (ICANN 2022; WIPO 2017)

Thus, the evolving nature of the first element brings into question the extent to which exceptions, such as the creation of new generic meanings, acronyms, context and phonetic differences, will be fairly considered and consistently applied in decisions (World Trademark Review 2021).

Venue shopping across DRSPs

The principle of ‘venue-shopping’ is found across the disciplines of law, urban affairs, international affairs, and politics (Saptiochne and Smith 2012). Venue-shopping is best described as the process whereby individuals ‘[find] a decision setting that offers the best prospects for reaching one’s policy goals’ (Pralle 2003). Since complainants can choose their DRSP, early UDRP critics expressed concern over the potential for biases if venue-shopping occurred (Mueller 2001). For example, complainants could decide to send their disputes only to NAF because they believed that NAF would make decisions favouring them. As a result of this ‘forum-shopping’, other DRSPs may also receive fewer disputes, further skewing the fairness and efficacy of the UDRP process.

To assess the extent to which the strengths and limitations can be observed consistently throughout the UDRP process, this study conducts an exploration of the available disputes using computational text-mining techniques. While many studies have analysed the efficacy of the UDRP process, few have focused on leveraging computational text-mining to examine the most prevalent themes and trends in big data samples of the UDRP process. This article addresses this critical research gap by building on Mueller’s (2001) work through significantly expanding the sample, scope, and analytical techniques used to examine the characteristics and trends of the UDRP process. Specifically, this article extends Mueller’s sample size from 2166 UDRP proceedings to include the 75,590 UDRP proceedings made available through the DNS Research Federation’s (DNSRF’s) Data Analytics Platform (DAP.Live). We ask three overarching research questions: RQ1. To what extent does the UDRP dispute process differ between DRSPs?; RQ2. What are the most prevalent themes as represented by keywords and topics?; RQ3. To what extent have these topics changed over time? The following sections present our methodology and findings.

Materials and methodology

Techniques

To answer this study’s research questions, we leveraged descriptive statistics for non-textual analysis of the cases, and inductive text-mining techniques for textual analysis. Inductive text-mining techniques seek to explore and identify trends in the data – our use of them is detailed next.

Our analysis began with term frequency to get an initial overview of the dataset. This ‘bag of words’-based approach – where word order is not taken into consideration – assumes that words appearing frequently in a textual corpus are ‘important’ for that corpus.

Next, we used a slightly more sophisticated technique called ‘term frequency’ by inverse document frequency (TF-IDF). TF-IDF is a common heuristic used to identify important words by weighing term frequency against term frequency across multiple documents in the dataset. It is calculated by multiplying two measures, term frequency (TF) and an inverse measure of document frequency (IDF), so that a word that occurs in every document would be reduced in importance.

We then used an unsupervised machine-learning technique called ‘topic modeling’. We chose to use the Structural Topic Modelling (STM) algorithmic approach to topic modelling. Topic modelling assumes that every topic is represented by a collection of words associated with that topic, and that every document has multiple topics contained within. The goal of topic modelling is to identify coherent ‘topics’ within a dataset, and the documents that contain those topics.

Data processing

Data acquisition

The DNSRF web scraped UDRP cases from these four DRSPs: ADNDRC; CAC; NAF; and WIPO. These four DRSPs are the largest (see Table 2). The CIIDRC had a total of 143 UDRP cases, and the ACDR a total of 11 cases, available on their respective websites (ACDR, n.d.; CIIDRC, n.d.) which the DNSRF did not scrape. The scraped cases were either in PDF or HTML format therefore the DNSRF parsed them to extract relevant case metadata and transform them into a tabular format. Those processed cases were made available on DAP.Live as a dataset titled ‘UDRP Case Complainants’.

We downloaded the ‘UDRP Case Complainants’ dataset using DAP.Live. First, we created a stored query titled ‘UDRP Summary V2’ which contained 24 metadata variables related to each case (see Table 1). Next, we created an API titled ‘udrp-summary-v2’ with the underlying dataset being the stored query. Lastly, we used the httpaddress of ‘udrp-summary-v2’ to download the dataset as separate batches of CSV files each with a limit of 25,000 cases. The final download consisted of four batches and a total of 75,590 cases.

Table 1. Metadata parsed from UDRP cases.

Variable	Description
provider	DRSP (includes derivative DRSPs)
caseFiledDate	Date case was filed
caseNumber	Case ID
casePublishedDate	Date case was published
complainants	Case complainant(s)
decision	Case outcome
decisionDate	Date of case outcome
decisionText	Full text of case
decisionUrl	Case URL
domains	Doman(s) disputed in case
importedDate	Date case was added to DAP.Live
panelists	Case adjudactor(s)
panelSize	Number of adjudactor(s)
precedentCases	Precedent cases cited
registrar	Domain registrar(s) of domains disputed in case
respondents	Case respondent(s)
sectionA	Text of Section A
sectionB	Text of Section B
sectionC	Text of Section C
sectionCompContentions	Text of complainant(s) contentions
sectionFacts	Text of case facts
sectionFindings	Text of case findings
sectionHistory	Text of case history
sectionRespContentions	Text of respondent(s) contentions

Data cleaning

The data cleaning utilised R, via RStudio, and consisted of four parts:

1. Imported the four CSV files into RStudio and merged them into a single file.

2. Created a variable called ‘baseProvider’ which only contains non-derivative DRSPs. This was mainly to address ADNDRC which in the dataset has four derivatives: ADNDRC-bj (Beijing office), ADNDRC-hk (Hong Kong office), ADNDRC-kl (Kuala Lumpur office) and ADNDRC-so (Seoul office). This dataset was exported as ‘URDPfiles_Full’.

3. Removed cases that either had no ‘Decision Text’ or no ‘Decision’ resulting in 1572 removals.

4. Created a variable called ‘cld3_lang’ to store a case’s language then filtered out non-English cases which resulted in 5800 removals. To identify a case’s language, we utilised Google’s Compact Language Detector v3 (CLD3) which is an open-source ‘neural network model for language identification’ (Google, n.d.). We used the R package ‘cld3’ (Ooms 2023) to apply the CLD3 model on each case’s decision text.

A total of 7372 (9.75% of the 75,590) cases were removed (see Table 2). The result was a dataset called ‘UDRPfiles_English’ consisting of 68,218 cases and 25 variables.

Table 2. Details of data cleaning cases removed by DRSP.

	UDRPfiles_Full (n = 75,590)		UDRPfiles_English (n = 68,218)		Difference (n = −7372)
Provider	n	Proportion	n	Proportion	n	Percent change
ADNDRC	2411	3.19%	757	1.11%	−1654	−68.60%
Beijing	602	0.80%	132	0.19%	−470	−78.07%
Hong Kong	1574	2.08%	584	0.86%	−990	−62.90%
Kuala Lumpur	27	0.04%	12	0.02%	−15	−55.56%
Seoul	208	0.28%	29	0.04%	−179	−86.06%
CAC	647	0.86%	269	0.39%	−378	−58.42%
NAF	30,754	40.69%	30,617	44.88%	−137	−0.45%
WIPO	41,778	55.27%	36,575	53.61%	−5203	−12.45%

‘URDPfiles_Full’ was used for non-textual analysis of the cases; ‘UDRFfiles_English’ was used for textual analysis.

Results

Descriptive statistics

Number of UDRP cases

To show the growth in UDRP cases, we compared descriptive statistics from our dataset to those presented by Mueller (2001). Figure 1 shows that the number of cases peaked in 2022 at 7100 which was a 27% year-over-year increase on the second peak of 5580 cases in 2021. The average number of cases between 2000 and 2022 (fully reported years) was 3055.

Figure 1. UDRP cases decided per year.

Domain names

Our dataset also ‘indicates that systematic, large-scale cyber-squatters constitute a very small part of the total number of cases’ (Mueller 2001). Eighty-three per cent of UDRP cases involved only one domain name (see Table 3), and 93.98% of cases involved three or fewer. However, using Mueller’s (2001) definition of large – arbitrarily defined as six or more domain names – the number of large cases increased by 1.4 percentage points to 3.41% and, on an absolute basis, increased by over one-hundred times to 2577. In addition, our dataset introduces extra large cases (we arbitrarily define ‘extra large’ as 245 or more domain names) of which there were 12. The largest case was ‘Inter-Continental Hotels Corporation, Six Continents Hotels, Inc. v. Daniel Kirchhof’ (WIPO Arbitration and Mediation Center 2010) which involved 1529 domain names.

Table 3. Domain names contested per case.

Domain names contested per case	UDRP cases	Share of total (%)
1	62,758	83.08
2	6165	8.16
3	2069	2.74
4	1226	1.62
5	734	0.97
6–244	2577	3.41
245 or more	12	0.02
Total UDRP cases	75,541

Market share

Like Mueller’s (2001) findings, the share of disputes handled by the four DRSPs is unequally distributed. WIPO was still dominant with a market share of 55%. Figure 2 shows the cumulative share of all cases in our dataset from 2000 to 3 June 2023. Figure 3 shows a yearly breakdown of the market share of the four DRSPs. Although there are some yearly fluctuations with NAF having a market share greater than 51% in five of the 22 fully reported years (2000–2022), overall WIPO is still the dominant DRSP, having a market share greater than 51% in 18 of the 22 fully reported years.

Figure 2. Cumulative market share of DRSPs.

Figure 3. Market share of DRSPs by year.

Precedent cases

Precedent cases were extracted using a rules-based method called regular expressions (abbreviated as ‘regex’). The scraped and parsed data had some issues related to precedent cases thus to ensure high data-quality, very strict regex rules were used to extract the precedent cases analysed below. The limitation of this approach was that some precedent cases were missed.

Thousands of UDRP cases are decided each year which serves as a growing pool of precedent that DRSPs can cite, so we chose to examine this. Cases that begin with: ‘D’ are WIPO cases; ‘FA’ are NAF. Across our dataset, there were a total of 616,379 citations of 34,818 precedent cases. On a per case basis, 35% of cases cite 11–20 precedent cases (see Table 4), and most cases (56.5%) cite 6–20. The 2006 case ‘Gaylord Entertainment Company v. Nevis Domains LLC’ (WIPO Arbitration and Mediation Center 2006) used the most precedent citing 112 cases.

Table 4. Precedent cases cited per case.

Precedent cases cited per case	UDRP cases	Share of total (%)
1	5637	9.3
2	4915	8.1
3	4215	6.9
4	3428	5.6
5	3116	5.1
6–10	13,051	21.5
11–20	21,282	35.0
21 or more	5155	8.5
Total UDRP cases	60,799

The first two most cited cases, D2000-0009 (‘Talk City, Inc. v. Robertson’) and FA95095 (‘Vertical Solutions Mgmt., Inc. v. webnet-marketing, inc’), respectively, are from 2000 and accounted for 5.2% of total citations (see Figure 4). They are often cited together as a basis for a respondent’s failure to respond, allowing the panel to accept as true all allegations of the complaint (National Arbitration Forum 2022a). The fifth most cited case, D2000-0003 (‘Telstra Corp Ltd v. Nuclear Marshmallows’), is often cited as a basis for bad faith by a respondent having an inactive domain (Isenberg 2020; WIPO 1999).

Figure 4. Top cited precedent cases.

Figure 5 shows the top five cases cited by outcome. Each outcome cited some of the same cases; however, there were interesting insights when focusing on the unique cases cited. In complainant wins the most cited unique case was the 2006 case FA741828 (‘Hanna-Barbera Prods., Inc. v. Entm’t Commentaries’) which is cited as basis for part two of the three-prong test (rights and legitimate interests); specifically, a complainant first making a prima facie case that the respondent lacks rights and legitimate interests in the disputed domain name(s) before the burden shifts to the respondent to show it does have rights or legitimate interests (National Arbitration Forum 2016).

Figure 5. Top cited precedent cases by outcome.

In respondent wins the most cited unique case was the 2002 case FA116765 (‘Creative Curb v. Edgetec Int’l Pty. Ltd.’) which is cited as a basis for a complainant failing to prove all elements of the three-prong test (National Arbitration Forum 2022b).

In cases that are terminated, cancelled, or settled, the most cited unique case was the 2001 case D2001-0903 (‘Oki Data Americas, Inc. v. ASD, Inc.’) which is cited as a basis for part one of the three-prong test (identical or confusingly similar); specifically, that a domain name incorporates a trademark (National Arbitration Forum 2022c) although the UDRP ‘is not intended to prohibit, without exception, the incorporation of a third party’s trademark within a domain name in all cases’ (National Arbitration Forum 2022d).

In cases that result in a split decision, the only cited unique case was the 2003 case FA157287 (‘eGalaxy Multimedia Inc. v. ON HOLD By Owner Ready To Expire’) which is cited as basis for dismissing a complaint because the complainant did not produce clear evidence to support their allegations (National Arbitration Forum 2022e).

RQ1: To what extent does the UDRP dispute process differ between DRSPs?

RQ1.1: What percentage of disputes are won by the complainant, respondent, or are terminated?

To answer the first sub-question, we analysed case outcomes and their relationship to other factors that previous literature had found were important.

Complainants remain major winners under the UDRP which is similar to Mueller’s (2001) findings. However, compared to Mueller’s findings, complainants’ dominance has increased by 10.9%. On aggregate, when decisions are made, complainants succeed in winning or cancelling contested domain name(s) 90.9% of the time; respondents win 8.9% of the time. The remaining 0.2% of the time the case outcome is a split decision meaning ‘some domain names are transferred to the complainant and some are retained by the respondent’ (Mueller 2001).

To search for nuance in the outcome ratios, we analysed outcomes by DRSP (see Table 5). The outcome ratios of ADNDRC, NAF and WIPO are very similar, except for two differences. First, fewer NAF cases are terminated, cancelled, or settled which is what Mueller (2001) also found. Second, many more ADNDRC cases are terminated, cancelled, or settled. At 98.5%, CAC has a much higher complainant win ratio than the other DRSPs, but this is most likely due to the low number of CAC cases in our dataset. Complainants succeed in winning or cancelling contested domain names around 91% when they bring their complaint to NAF and WIPO versus 94% of the time when the complaint is brought to ADNDRC.

Table 5. Outcomes by DRSP.

	ADNDRC		CAC		NAF		WIPO		All
Complainant wins	2113	88.1%	268	98.5%	27,917	90.8%	36,883	89.1%	67,181	89.8%
Respondent wins	136	5.7%	4	1.5%	2630	8.6%	3873	9.4%	6643	8.9%
Split decision	–	–	–	–	142	0.5%	–	–	142	0.2%
Terminated/cancelled/settled	149	6.2%	0	0.0%	63	0.2%	623	1.5%	835	1.1%
Total UDRP cases	2398		272		30,752		41,379		74,801

Mueller (2001) also examined the impact of a respondent defaulting on the outcome. A respondent default occurs when a respondent does not respond to a complaint. Compared to Mueller’s findings, the respondent default rate in our dataset increased by 34% to 68%, meaning a little over two-thirds of cases went uncontested.

Overall, in contested cases, respondents win 13% of the time (see Table 6) which is above the 8.9% in Table 5, but it has dropped 12% compared to Mueller’s (2001) findings. The respondent win rate for NAF is 24% which is higher than WIPO’s (19.5%); therefore, between those two, Mueller’s (2001) finding holds that WIPO is the DRSP ‘most likely to transfer a domain name when a decision is rendered in contested cases’. Overall, in uncontested cases, respondent win rate is 1.65% (see Table 7) which is in line with Mueller’s (2001) findings.

Table 6. Outcomes in contested cases by DRSP.

	ADNDRC		CAC		NAF		WIPO		All
Complainant wins	205	91.12%	28	100.00%	5884	74.87%	9788	79.15%	15,905	86.28%
Respondent wins	19	8.44%	–	0.00%	1890	24.05%	2413	19.51%	4322	13.00%
Split decision	–	0.00%	–	0.00%	64	0.81%	–	0.00%	64	0.20%
Terminated/cancelled/settled	1	0.44%	–	0.00%	21	0.27%	166	1.34%	188	0.51%
Total contested cases	225		28		7859		12,367		20,479

Table 7. Outcomes in default cases by DRSP.

	ADNDRC		NAF		WIPO		All
Complainant wins	110	98.22%	20,752	97.26%	20,897	96.56%	41,759	97.35%
Respondent wins	1	0.89%	485	2.27%	390	1.80%	876	1.65%
Split decision	–	0.00%	71	0.33%	–	0.00%	71	0.11%
Terminated/cancelled/settled	1	0.89%	29	0.14%	356	1.64%	386	0.89%
Total uncontested cases	112		21,337		21,643		43,092

RQ1.2: How much time elapses between complaint dispute creation and decision release?

To answer the second sub-question, we analysed time to decision along with its relation to case outcome. Mueller (2001) noted the number of days DRSPs need to decide might play a role in a complainant bringing a case to a DRSP. Our dataset only has the date a case was filed for two of the four DRSPs we are examining – CAC and WIPO.

On average, it takes CAC 36 days to decide, compared to 63 for WIPO. The average for the two DRSPs is 63 days. Like Mueller (2001) we turned decision time into a binary variable with ours showing how often it takes DRSPs to decide a case in more or less than 60 days (see Table 8). 98.5% of CAC’s cases were resolved in less than 60 days compared to 53.5% for cases brought to WIPO.

Table 8. Decision times by DRSP.

	60 days or more		Below 60 days		Total
	Cases	Share of DRSP cases	Cases	Share of DRSP cases	Cases	Mean (days)
CAC	4	1.5%	268	98.5%	272	36
WIPO	14,785	46.5%	17,007	53.5%	31,792	63
Total	14,789	46.1%	17,275	53.9%	32,064	63

There are also time differences between case outcomes. On average, it takes 62 days to decide where complainants win, 72 where respondents win, and 63 if the case is terminated, cancelled or settled. Applying our binary variable (see Table 9), 55% of cases where complainants win are decided in less than 60 days compared to 37% in cases where respondents win.

Table 9. Decision times by outcome.

	60 days or more		Below 60 days		Total
	Cases	Share of outcome cases	Cases	Share of outcome cases	Cases	Mean (days)
Complainant wins	12,906	45%	16,054	55%	28,960	62
Respondent wins	1634	63%	966	37%	2600	72
Terminated/cancelled/settled	249	49%	255	51%	504	63
Total	14,789		17,275		32,064

Looking at decision times year-over-year, it appears that from 2009 CAC and WIPO are deciding cases quicker (see Figure 6). From 2000 to 2008, an average of 45% of cases were decided in less than 60 days compared to an average of 58% from 2009 to 2022. This might be due to the maturity of the UDRP. Specifically, since most cases with a case filed date in our dataset are from WIPO, it might show WIPO is learning from its dominance and understands the importance of days to decision in maintaining its market share.

Figure 6. Decision times over time.

RQ2: What are the most prevalent themes as represented by keywords and topics?

RQ2.1: Who are the key complainants and respondents?

To answer the first sub-question, we found unique complainants and respondents from the appropriate metadata variables, lower-cased all the names, and applied regex to remove business entity descriptors such as ‘LLC’, ‘Inc’, et cetera. Complainants and respondents identify at the subsidiary-level so we maintained that.

The top complainant was ‘State Farm Mutual Automobile Insurance Company’ with 664 complaints filed (see Figure 7). Of the top 25 complainants, there is a mix between industries such as financial services, technology, retail, social media, et cetera. In 2022, State Farm Mutual Automobile Insurance Company had direct written premiums of $46.6 billion (16.9% market share) making it the largest private passenger automobile insurer in the U.S. (Insurance Information Institute 2023). In 2022, private passenger automobile insurance was the largest Property/Casualty insurance market in the U.S. (NAIC 2023). To earn and maintain its market share, State Farm heavily markets itself using television advertisements, most notably during National Football League (NFL) games which are the most popular sporting league games in the U.S. (Best 2023). These factors make State Farm a prominent, well-known, lucrative trademark and thus may be the reason it brings forward many complaints.

Figure 7. Key complainants.

Several of the top 25 respondents (see Figure 8) have entity names that look like domain services such as ‘Domain Admin / Whois Privacy’ and ‘Super Privacy Service Ltd c/o Dynadot’ which are both registered corporate entities. Another top respondent, ‘Above.com Domain Privacy’ is a corporate entity that sells domain parking services. Domain parking is the ‘leveraging of advertisements on a parked web domain to generate revenue while that domain is otherwise inactive’ (Scott 2023). Parked websites are ‘completely non-functioning except for a landing page that contains ads’ (Scott 2023). Above.com Domain Privacy has a tagline of ‘maximizing payouts from premium advertisers and parking services’ (Above.com, n.d.) and claims to have ‘over 20 million domains optimized generating over $500 million in revenues’ (Above.com, n.d.). Domain parking is not illegal but, under the UDRP, the practice is usually mentioned to show that the respondent has either no rights or legitimate interests to the domain name.

Figure 8. Key respondents.

RQ2.2: What are the most frequently used keywords in disputes?

To answer the second sub-question, we used term frequency analysis across the English UDRP cases. ‘Selling’ was the most frequently occurring word (see Figure 9). Other interesting frequently occurring words are ‘pattern’ which is an often referenced concept in UDRP cases as panelists seek a pattern of a certain behaviour. ‘Google’, ‘insurance’ and ‘farm’ are top complainants (see Figure 7) but they are also part of three regularly cited precedent cases. ‘Typosquatting’ is when a person registers a common misspelling of another entity's domain name, for example googl.com rather than google.com (Microsoft, n.d.). Many UDRP cases reference typosquatting to show bad faith.

Figure 9. Frequently used words in UDRP cases.

RQ2.3: What are the most important keywords used in disputes?

We used TF-IDF to answer the third sub-question. The most important occurring words are trademark names such as ‘payoneer’, ‘wolfram’ and ‘adimpo’ (see Figure 10). Some of these trademarks were involved in their own cases with some then becoming precedent cases.

Figure 10. Most important words in UDRP cases.

RQ2.4: What topics are most prevalent across the dataset?

We used topic modelling to answer the fourth sub-question. For topic modelling one must select the number of topics, ‘k’. We created several models, each with a different value of ‘k’, to find an appropriate value. We chose a value of 50 for ‘k’ as the models it produced had both high exclusivity and high semantic coherence compared to the other models. Exclusivity measures how many topics a word belongs to. Semantic coherence measures how contextually similar words in a topic are. It is ‘maximized when the most probable words in a given topic frequently co-occur together, and [it] correlates well with human judgment of topic quality’ (Silge 2018). See Appendix for a list of all 50 topics from our topic model.

For Figure 11, the x-axis (gamma) is a metric that measures ‘proportion of the document that is made up of words from the assigned topic’ (Sevier 2017). Figure 11 shows that Topic 20 captures the precedent described in the ‘Precedent cases’ section regarding case FA157287 (‘eGalaxy Multimedia Inc. v. ON HOLD By Owner Ready To Expire’). Topic 31 is about domain parking; Topic 28 is about phishing; Topic 17 is about counterfeits; and Topic 45 is about typosquatting. These examples illustrate that our topic model performed well as it captures both abstract concepts, such as precedent, and concrete concepts such as phishing.

Figure 11. Top 20 topics in UDRP cases and their top seven words.

RQ3: To what extent have these topics changed over time?

To answer the third research question, we defined four time periods and looked at the top five topics in each period (see Figure 12). One of the objectives when defining the periods was to capture both from the beginning and post COVID-19 which the period ‘2020–2023’ does. It is also important to reiterate that gamma is based on document proportion and the number of documents (cases) in each period are similar (see Table 10), thus the results are comparable.

Figure 12. Topics in UDRP cases by period.

Table 10. UDRP cases decided by time period.

Period	Number of Cases	Share of total (%)
2000–2006	13,389	18
2007–2013	19,807	27
2014–2019	19,410	27
2020–2023	20,216	28
Total UDRP cases	72,822

Topic 20 captures precedent related to complaints being dismissed. From 2000 to 2013, Topic 20 is not prevalent; from 2014 to 2019 it becomes more prevalent; and during the COVID-19 period it becomes the most prevalent topic with 10.3% of words in cases from that period being related to it. This implies that during the COVID-19 period, there was an uptick in cases being dismissed due to complainants not providing clear evidence to support their allegations.

Also, during the COVID-19 period, we can see there was an increase in domain names used for phishing as shown by Topic 28. Phishing is ‘a technique for attempting to acquire sensitive data, such as bank account numbers, through a fraudulent solicitation in email or on a website, in which the perpetrator masquerades as a legitimate business or reputable person’ (Computer Security Resource Center, n.d.). The perpetrators would try to make their domain names appear like those from reputable brands. This finding corresponds with literature that phishing dramatically increased during COVID-19 due to several factors such as the shift to remote work, government stimulus and vaccine rollout (Hu 2021).

Discussion

This study has helped clarify the role the ICANN UDRP has played in managing the intense global competition engendered by the trademark dilemma. We see the UDRP as substantially contributing to the private, non-state, ‘governance’ of the internet and to the stability of the broader DNS. We have drawn upon international regime theory to help clarify the importance of the UDRP in establishing principles, norms, values, and decision-making processes, placing it at the centre of a ‘strong’ global regime with substantial mechanisms to enforce its decisions. Over the 23 years of the UDRP the number of disputes it handles annually has grown. This process is certainly not without criticism.

One of the strongest, and most fundamental criticisms of the UDRP emerges from the concern that the entire regime was captured from its inception by the dominant trademark holders who put in place a system that inherently benefited them. Given this foundation, the ongoing concern is that existing trademark holders continue to dominate the UDRP process. Our literature review discussed the substantial participation by trademark holders and the business community in the formation of both ICANN and the UDRP. However, the corporate participation in the process was counterbalanced with substantial participation by key broad-based non-governmental organisations. However, we know that ‘participation’ in these global regime formation processes by non-governmental actors does not automatically equal ‘influence’ (Cogburn 2017). We have highlighted that the sustained regime formation process of the UDRP has clearly benefited established trademark holders relative to new entrants. However, we have argued that the stability engendered by this regime benefits the longer-term growth of the internet, which benefits everyone. This stability concern is so strong among most internet governance stakeholders that it even dominated the highly politicised IANA transition process, which represented a fundamental shift of internet governance decision-making out of direct U.S. Government control, and into the hands of the global multistakeholder community. We argue that the protection of consumers from fraud and abuse on the internet, including phishing scams, as well as protecting the rights of legitimate trademark holders, had to be balanced against interests related to innovation, new entrants, and participation from the global south.

Our analysis of default versus contested cases is one possible method to discern whether the UDRP process’ decisions are justifiable given that trademark holders played a key role in its establishment. It can be presumed that good actors are more likely to contest UDRP complaints and bad ones are more likely to default. We found most respondents default and in such cases trademark owners win 97.35% of the time. However, in cases where a respondent contests, the complainant win rate drops by 11–86.28%. From this analysis, cases with bad actors favour trademark owners compared to those with good ones hence, although there is definite room for improvement in the system, showing some justifiability in decision rates.

Moreover, a closer analysis of successfully contested cases – that is, cases where the respondent did not default and won – in our dataset provides evidence of not only non-trademark interests winning but also strong UDRP jurisprudence that essentially protects those interests. An example of such jurisprudence is called ‘Plan B’ (or ‘Plan B case’) which establishes a complainant as turning to the UDRP process in an improper attempt to deprive a respondent of a disputed domain name after failing to purchase, or lease, the domain name by means of commercial negotiation (Isenberg 2017; WIPO Arbitration and Mediation Center 2021). Plan B is based on another UDRP jurisprudence called ‘Reverse Domain Name Hijacking’ (abbreviated as ‘RDNH’) which the UDRP defines as ‘using the [UDRP Policy] in bad faith to attempt to deprive a registered domain-name holder of a domain name’ (ICANN 2015). RDNH is usually invoked by respondents in their contentions to a complaint and allows the respondent to provide an argument of a complainant acting in bad faith that the appointed case panelist(s) will rule on in addition to the regular ruling of cancelling, suspending or transferring, the disputed domain name.

Our argument is that the UDRP has created a very stable low cost and quick resolution process of private ordering for domain name disputes. We found the average time to resolve a UDRP case was 63 days. In addition, the present rate of trademark owner success beneficially protects both consumer interests and trademark owners’ own interests primarily because it protects both parties against bad actors. In the absence of the UDRP process there would be global confusion resulting in much longer resolution times which would benefit bad actors who, from our topic modelling analysis, primarily use domains for domain parking, phishing/impersonation, and counterfeiting, which negatively impacts both parties.

The UDRP is not a failure of private ordering. In contrast, with its promulgation of global norms, principles, values, decision-making processes, and enforcement mechanisms for managing disputes related to domain-name ownership, it plays a central role in the global regime for governing the DNS. The UDRP regime is not state centric, but instead, it involves a multiplexity of state, private sector, international organisation, and civil society actors, which contributes substantially to the ongoing stability of the internet and the DNS.

Overall, we believe the stability garnered by the UDRP outweighs the concerns about it not being sufficiently representative. Recommendations for how to revise or update this process must be carefully balanced to enhance what the UDRP currently represents, without taking it backwards. For example, the recommendation to enable ICANN to receive all complaints and then randomly assign them to a DRSP might further eliminate the prospects of venue shopping, thus strengthening the domain-name dispute resolution system. However, other recommendations, such as making the complaints private would perhaps engender more distrust with the perceived lack of transparency.

Acknowledgements

Dr Cogburn, Mr Ochieng, and Ms Wong would like to thank the staff of the DNSRF for their ongoing support of this research project, especially Emily Taylor, Mark Robertshaw and Georgia Osborn. We would also like to thank the anonymous reviewers who provided such helpful insights during the review process. They would also like to thank American University Vice Provost for Innovation and Research, Dr Diana Burley for introducing our research team to the DNSRF. Also, we would like to thank the Internet Governance Lab Faculty Co-Director, Dr Nanette Levinson. We also thank Fiona Alexander, Distinguished Policy Strategist in Residence in the School of International Service and Distinguished Fellow at the Internet Governance Lab. We would also like to thank Eduardo Diaz, Alfredo Calderon, and the entire team working on the North American School of Internet Governance (NASIG) for inviting us to present the penultimate version of this paper during NASIG 2023 at American University.

Disclosure statement

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

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

This work was supported by the DNS Research Federation [grant number 37282].

Notes on contributors

Derrick L. Cogburn

Dr Derrick L. Cogburn is a Professor at American University in Washington, DC, with a joint appointment in the School of International Service, serving in the Department of Environment, Development & Health; and in the Kogod School of Business, serving in the Department of Information Technology & Analytics. He is Founding Executive Director of the AU Institute on Disability and Public Policy (IDPP), Faculty Co-Director of the Internet Governance Lab (IGL), and Director of COTELCO, the Collaboration Laboratory. Dr Cogburn’s multifaceted research interests are located at the intersection of information technology, global governance, and socio-economic development. He uses computational text mining and data analytics approaches, including unsupervised and supervised machine learning and artificial intelligence methodologies, to better understand large-scale text-based datasets in the areas of cybersecurity, disability policy, and distributed collaboration in knowledge work. He has published widely, with his most recent books being: Researching Internet Governance: Methods, Frameworks, Futures (MIT Press 2020); Transnational Advocacy Networks in the Information Society: Partners or Pawns? (Palgrave-McMillian 2017); Making Disability Rights Real in Southeast Asia: Implementing the CRPD in ASEAN (Lexington 2016); and The Turn to Infrastructure in Internet Governance (Palgrave-McMillan 2016). He is Editor of the Palgrave Macmillan book series Information Technology and Global Governance.

Theodore Andrew Ochieng

Theodore Andrew Ochieng is a Senior Research Associate in the American University Internet Governance Lab and a graduate of the Analytics Master’s Program at the Kogod School of Business at American University. He has several years of experience helping organisations leverage data to solve problems by increasing data utility through supporting various stages of the modelling and analytics pipeline, such as improving ETL pipelines, data cleaning, feature engineering, and developing and deploying models.

Haiman M. Wong

Haiman M. Wong is a Doctoral Candidate in Purdue University’s Doctor of Technology program. Her research interests centre on the intersection of technology innovation and leadership, natural language processing, rhetoric, political science, law and internet governance. Her work experience spans across the finance, education, cybersecurity, political consulting, non-profit and technology sectors, where she worked in data science, communications and media relations, global cybersecurity risk and threat intelligence management, cybersecurity incident response and digital transformation. She is also a Senior Research Associate in the American University Internet Governance Lab and an alumna of the Master of Data Science program at American University’s College of Arts and Sciences.

Appendix. All 50 topics in UDRP cases and their top seven words

Table

Topic	Gamma	Top 7 Terms
Topic 48	5.9%	event, position, relies, extent, earlier, explained, concerned
Topic 36	4.7%	complete, application, federal, purchase, purchased, remain, meet
Topic 25	4.3%	searle, compagnie, tropic, hustle, concrete, martin, mktg
Topic 42	4.2%	assocs, prop, tex, operate, adequately, bond, diverting
Topic 19	4.1%	receipt, database, copies, assigned, telephone, appointment, requesting
Topic 20	4.0%	expire, conclusory, ready, multimedia, dismiss, subjective, unsubstantiated
Topic 31	3.4%	revenue, sponsored, parking, advertising, traffic, generate, advertisements
Topic 33	3.2%	risk, implied, social, jurisdictions, carries, targeting, dictionary
Topic 47	3.2%	methods, agency, famous, kaisha, kabushiki, infringing, continuous
Topic 28	3.1%	fraudulent, phishing, scheme, fraud, send, impersonate, employees
Topic 14	3.0%	pattern, selling, disrupting, renting, misleadingly, banking, credit
Topic 17	2.9%	sold, sells, selling, fashion, accessories, unauthorised, store
Topic 49	2.9%	aol, prods, hanna, commentaries, barbera, entm, gerberg
Topic 9	2.8%	software, media, entertainment, video, popular, news, articles
Topic 16	2.5%	ireland, britain, community, trading, london, goodwill, proprietor
Topic 45	2.3%	misspelling, typosquatting, intentional, typographical, error, errors, typing
Topic 13	2.3%	advertising, promoting, promotion, recognition, television, extensively, media
Topic 38	2.2%	europe, north, distribution, stock, equipment, asia, markets
Topic 21	2.1%	reproduces, paris, unaware, covering, famous, renewed, duly
Topic 2	1.9%	difference, john, virtually, travel, hotels, enterprises, deficient
Topic 18	1.9%	entmt, paul, hanna, prods, barbera, commentaries, differences
Topic 30	1.8%	jr, receives, referral, receiving, charles, manager, michigan
Topic 12	1.7%	substantive, proceed, delay, translation, familiar, informed, merits
Topic 44	1.6%	negative, impossible, consensus, task, requiring, bears, abusive
Topic 41	1.6%	falls, remains, listing, fall, illustrative, derived, forms
Topic 40	1.4%	telstra, nuclear, marshmallows, passive, plausible, conceive, illegitimate
Topic 35	1.4%	republic, switzerland, japan, italy, kong, federation, korea
Topic 11	1.4%	pharmaceutical, roche, hoffmann, medical, healthcare, health, aventis
Topic 37	1.4%	data, americas, oki, asd, disclaimer, disclose, distributor
Topic 15	1.4%	standard, standing, involves, viewed, components, threshold, assess
Topic 27	1.3%	texas, associates, cayman, recognise, comply, consented, scope
Topic 1	1.3%	inactive, distinctiveness, actively, contemplated, passive, degree, message
Topic 5	1.3%	balance, probabilities, onus, evaluation, publicly, affect, assessment
Topic 6	1.2%	david, grounds, additions, rise, concrete, supra, indicative
Topic 7	1.2%	alaska, airlines, air, dawn, disruptive, limb, alia
Topic 3	1.2%	discretion, conducted, ability, possibility, adduced, fairness, exercise
Topic 26	1.2%	brand, sweden, decades, construction, subsidiaries, brands, famous
Topic 43	1.2%	stores, retail, mart, wal, choice, department, confused
Topic 32	1.1%	preliminary, controlled, relate, registrants, instant, control, claiming
Topic 46	1.0%	insurance, automobile, farm, smith, mutual, illinois, anderson
Topic 8	1.0%	sports, properties, matthew, games, contend, fourth, partner
Topic 39	1.0%	alter, hyphens, air, devices, punctuation, seeks, white
Topic 22	0.9%	terry, realm, peppard, noting, pertinent, save, alterations
Topic 10	0.8%	morris, philip, express, precedent, fame, inconceivable, takes
Topic 29	0.8%	criteria, brown, expression, rely, chosen, arises, approval
Topic 4	0.7%	airlines, empire, turkey, consistent, opinion, turkish, note
Topic 50	0.7%	northern, goodwill, profit, evidenced, inclusion, supported, tucows
Topic 23	0.6%	fulfilled, croatia, condition, peter, infers, consistently, summarised
Topic 24	0.5%	digital, dr, chen, wang, juris, brand, master
Topic 34	0.4%	modern, michael, richard, citing, vanguard, management, ventures