Received: November 25, Date of acceptance: December 15, 2020, Date of publication: December 23 2020, DOI: https://doi.org/10.26826/law-in-context.v37i1.126 Knowledge Graphs: Trust, Privacy, and Transparency from a Legal Governance Approach By Daniel Schwabe1, Professor at Department of Informatics (INF), Pontifical Catholic University, Rio de Janeiro, Orcid: https://orcid. org/0000-0003-4347-2940 Carlos Laufer1, Associate Researcher, TecWeb Lab, at Dept. of Informatics, PUC-Rio, Orcid: https://orcid.org/0000-0002-2606-4067 and Pompeu Casanovas2, Research Professor at La Trobe University Law School, Bundoora, Melbourne, Australia, Orcid: https://orcid. org/0000-0002-0980-2371 1 Pontifical Catholic University, Rio de Janeiro, Brazil 2 La Trobe University, Melbourne, Australia ABSTRACT This paper presents the Knowledge Graph Usage framework, which allows the introduction of Knowledge Graph features to support Trust, Privacy, Transparency and Accountability concerns regarding the use of its contents by applications. A real-world example is presented and used to illustrate how the framework can be used. This article also shows how knowledge graphs can be linked to the elements of legal governance. Thus, it is an invitation to dialogue for legal and Law & Society scholars who might be interested in how the evolution of the web of data and computational sciences intersects with their own discipline. Keywords – Knowledge Graphs, Privacy, Transparency, Trust, Legal Governance, Legal interpretation Acknowledgements: A former version of this paper deprived from its present socio-legal dimension has been presented by Daniel Schwabe and Carlos Laufer with the title “Trust and Privacy in Knowledge Graphs” at The Web Conference (WWW’19), May 13, San Francisco, USA. See: Proceedings of WWW '19. ACM, New York, NY, USA. Daniel Schwabe was partially supported by a grant from CNPq. The present work by Pompeu Casanovas has been carried out for the EU H2020 Programme LYNX, Legal Knowledge Graph for Multilingual Compliance Services. The interested reader can find a description of this project as a Research Note at the end of this volume, LiC 37 (1). Disclosure statement – No potential conflict of interest was reported by the authors. License – This work is under Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) https://creativecommons.org/licenses/by-nc-sa/4.0/ Suggested citation: Schwabe, D., Laufer, C., Casanovas, P. (2020). “Knowledge Graphs: Trust, Privacy, and Transparency from a Legal Governance Approach”, Law in Context, 37 (1): 24- 41, DOI: https://doi.org/10.26826/law-in-context.v37i1.126 Summary 1. Introduction 5. A Summary of the KG Usage Framework 2. A Legal Governance Approach 5.1 KI Representation 2.1 Components of Legal Governance 5.2 Controlling Usage 2.2 A Preliminary Framework 5.3 Rules 3. Background Concepts 5.4 Conflict Resolution 3.1 Trust 3.2 Privacy 6. Example Scenario revisited 3.3 Transparency 7. Conclusions 4. An illustration—Disaster Relief Donation 8. References Law in Context, Vol 37, Issue 1; 2020 24 ISSN: 1839-4183 1. INTRODUCTION data, which is a fundamental part of data in some domains The advent of the Web has enabled the generation of such as life-sciences (Kuhn et al. 2018). a vast amount of information, mostly in the form of in- KGs differ also on the way they are built and populated. terlinked pages with texts containing links (references) A few are curated (e.g., CYC), others rely on crowdsourced to other texts. Users accessed information by loading a information (e.g. Wikidata, and several, perhaps most, page - identified by a URL - in their browsers following extract information from structured, semi-structured or links in the page to find related information. The advent of textual information harvested from the Web). social networks added end-user-generated content, further The multiplicity of sources and various extraction increasing the amount of textual information available. approaches naturally raises the issue of data quality and However, it quickly became evident that search function- confronts the user of the data in the KG with the issue ality was needed to help users find relevant information, of trusting, or not, the information contained in the KG. ushering the creation and growth of search engines. While For some types of information, for example in case of search engines became hugely successful, it also became online reviews and online and social media, this trust can clear that searching strings contained within Web pages have a direct effect on commercial success (e.g. Angella was a limited approach to solve the actual problems that and Johnson 2016). This highlights the fact that data users have, which is to find information that is relevant for ultimately expresses a belief, opinion or point of view of a certain need. As stated by Google, people are interested some agent – the author. in “things, not strings”.1 For example, users would like to From a broader perspective, information (and knowl- distinguish between “Sydney” as a city and “Sydney” as the edge) is said to have become the prime resource in the name of a known person (e.g., “Sydney Pollack”, “Sydney Third Industrial Revolution, also called the Digital Age Poitier”, “Sydney Lumet”), a football club (“Sydney F..C”), – when digital technologies enabled new ways of generat- etc…, among many other possible meanings. ing, processing and sharing information (Castells 2010, Although to this day there isn’t a precise definition of Rifkin 2011), and is becoming even more central as we the term (Ehrlinger and Wöß 2016, Hogan et al, 2020), we move into the Fourth Industrial Revolution (4IR) (Schwab adopt the view that a Knowledge Graph (KG) represents 2017). The 4IR is characterized by a fusion of technologies, a network of interlinked descriptions of entities (objects, which is blurring the lines between the physical, digital, events, concepts etc.)-- a graph-theoretic representation and biological spheres. of human knowledge such that it can be ingested with Increasingly, systems and applications operate in semantics by a machine (Kejriwal 2019). a context in which the flow of information has direct Graph-based databases have been available for a long bearing on daily lives of billions of people, where two time2 and many more continue to be created on a regular fundamental characteristics of the use of such informa- basis – see (Hogan et al, 2020) for a survey. The original tion emerge – Transparency and Privacy. Transparency vision for the WWW was later enriched to form the Se- is the quality that allows participants of a community to mantic Web, instances of which can also be regarded as know what the particular processes and agents are being a KG—see, for example, the survey by Gandon (2018). used in its functioning. It is generally regarded as a means While the graph model or some variant has been to enable checks and balances within this community, used in several KGs, it has already been observed that ultimately providing a basis for trust among participants using only atomic (indivisible) nodes as the “granule” of that community. When the community is regarded as of information is insufficient to express complex types being the entire society, these checks and balances are of information, such as events, or time-varying data. For reflected in its political system to prevent misuse by any example, Wikidata (Vrandečić and Krötzsch 2014) is of the parties involved. organized around Items described by a collection state- One of the mechanisms created to increase transparency ments (Erxleben et al. 2014). Another reason for having in political systems is the enactment of regulations ensur- more complex “granules” is recording provenance (meta) ing the right of its members to access to information in a variety of contexts, ranging from government-produced 1 https://googleblog.blogspot.com/2012/05/introducing-knowledge graph-things-not.html 2 For instance, Wordnet (Miller 1995), DBPedia (Lehmann et al. 2013), Yago (Suchanek et al. 2007), CYC (Lenat 1995), NELL (Carlson et al 2007), ConceptNet (Speer and Havasi 2010) 25 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 information and data to consumer-related information should be introduced, and whether the Privacy Act should regarding goods and products, as well as the right of in- include a ‘right to erasure’. dividuals to freely create, publish and access information. It is our contention that these and similar needed As we will explain later, regulations do not comprehend reforms around the world should consider the technical only legislation, i.e. enacted laws and statutes, but all kinds developments occurring in computer science and seman- of legal instruments of governance—hard law, soft law, tic web studies. In order to deal with the myriad of often policies and ethics. conflicting cross-cutting concerns, Internet applications The free flow of information, on the other hand, may and systems should incorporate adequate mechanisms conflict with another basic human right, that of Privacy to ensure compliance of both ethical and legal princi- (Universal Declaration of Human Rights 1948, art. 12). ples. In order to be effective, we claim that the use of There are many definitions for Privacy (Paci, Squicciarini, Knowledge Graphs ought to provide support for these and Zannone 2018), but in essence they all refer to the concerns—trust, privacy and transparency. In this paper right of an individual to control how information about we propose a framework that enables this support. In her/him is used by others. Data Protection and Privacy Section 2 the regulatory framework approach and some have had, and still have, different regulatory regimes concepts stemming from the work already done in the field that are applied in a variety of jurisdictions in both Civil will be defined. It will also describe the preliminary legal and Common Law legal systems. For instance, Privacy is background. Section 3 backgrounds three main general considered a fundamental (human) right by the recent concepts—trust, privacy, and transparency—that will be General Data Protection Regulation (GDPR) enacted in taken into account in our modelling. Section 4 presents Europe in May 2018. It is also assumed as a constitutional an illustration by way of example, and Section 5 discusses right in many EU countries. But this is not so in the USA, the proposed KG representation framework, showing its in which privacy and data protection do not qualify as application in the illustration. Section 6 reflects on the specified fundamental rights under the Constitution. previous example, and Section 7 draws some conclusions The situation in Common Law countries is nuanced, and points to future work. varying with some subtlety between different national jurisdictions. For instance, there is no general law right 2. A LEGAL GOVERNANCE APPROACH to privacy in Australia. Although Australia is a signatory In this section, we argue that the privacy legal regula- to the International Covenant on Civil and Political Rights tory framework does not solely lean on national legislation (adopted by UN General Assembly on 16 December 1966, but encompasses other legal instruments of transnational and in force from 23 March 1976), the international law nature (such as protocols, standards, best practices and right to privacy conferred under Article 17 of the ICCPR ethical principles and values). The field of privacy is spe- has not been enacted into Australia’s domestic law (Watts cially suitable for a broader legal governance, in which and Casanovas 2018). The Privacy Act 1988 still regu- hetero-regulatory, co-regulatory and self-regulatory in- lates information privacy in the Commonwealth public struments tend to coexist in different ways according to sector and the national private sector. It covers personal the specific contexts created by the normative systems at information and sensitive information (such as health stake in regional, national, international and transnational information, ethnicity, sexual preference, trade union economic, social and political spaces (Pagallo et al. 2019; membership). This situation is also evolving, partially Pagallo, Casanovas and Madelin 2019). fuelled by the palliative reaction against Covid-19. From Legal governance can be defined as the processes and 2019 the Federal Government has been engaged in several practices of implementing the set of normative systems trends to update the definition of personal information.3 put in place in specific contexts for a variety of scenarios; The recent Issues Paper (Australian Government 2020) i.e. the process of creating sustainable legal ecosystems. on the reform displays a wide set of issues, including From this standpoint, legality is the result of the coordina- whether a statutory tort for serious invasions of privacy tion of different types of agency (artificial and/or human, 3 I.e. issuing several Recommendations for reform (Australian Government 2019). Among others (i) updating the definition of ‘personal information’ to capture techni- cal data and other online identifiers (Recommendation 16(a)); (ii) strengthening existing notification requirements (Recommendation 16(b)); (iii) strengthening consent requirements and pro-consumer defaults (Recommendation 16(c)); (iv) and introducing a direct right of action to enforce privacy obligations under the Privacy Act (Recommendation 16(e)). Law in Context, Vol 37, Issue 1, 2020 26 ISSN: 1839-4183 using socio-technical, cognitive-socio-technical systems regarding linked data resources. On the other hand, as or normative Multi-agent Systems). Some legal instru- is well known, from 2000 onwards many data breaches ments—such as rules extracted from legal norms—can have been reported—AOL, NETFLIX, Equifax, Cambridge be automated. Others, for instance, ethical values and Analytica etc.5 — and the lawsuits that followed indicate principles or best practices regarding the monitoring of a certain alarm about a general state of uncontrolled regulatory systems, cannot be fully hardcoded, as they surveillance (Norris et al. 2017, Zuboff 2019). Thus, the require human intervention and decision-making (Koops legal dimensions of security, intellectual property, pat- and Leenes, 2014) This is the case with trust, transpar- ents, licenses and, especially privacy and data protection ency, privacy and data protection. Their implementation should be taken into account and applied to the use of requires building institutions to hold complex models of legal resources and the building of new tools for the web legal governance. But this is not saying that they cannot of data (Rodríguez-Doncel et al. 2016). be semi-automated. On the contrary, from this perspec- Legal resources should be differentiated from legal tive, appropriate automation can facilitate human control sources. The former refers to the large number of exist- and monitoring. ing legal vocabularies and documents on the Web of Data. Accordingly, privacy and data protection can be con- The latter refers to the specific content that ‘counts as sidered from the legal governance approach. This means legal’ at regional, national, international and transna- that they can be implemented through the construction, tional levels to be effectively implemented or enforced. development and implementation of a technological Determining what is ‘valid’ law, what counts as legal, is toolkit, comprising data mining, data analytics and the in itself a non-trivial theoretical operation that is usually linked open data tools of the later developments of the performed through the concepts of doctrine, legal theory, Semantic Web—i.e. the Web of Data. The Knowledge and checks-and-balances (Sartor 2005, Peczenik 2011). Graphs approach that we embrace in this article is related For our purposes, it is worth noting that a taxonomy of a to this dimension. legal quadrant, or legal compass (Fig. 3) would suffice (i) to classify and annotate a variety of sources (ii) that are 2.1 COMPONENTS OF LEGAL GOVERNANCE deemed necessary to produce the ‘ecological validity’ of Since 2010, many large-scale RDF datasets have been a regulatory system—i.e. the condition of sustainability created. For instance, in 2017, Freebase 1 had 2.5 billion of legal ecosystems for a cluster of stakeholders (Poblet, triples;4 DBpedia2 had more than 170 million triples. LOD Casanovas and Rodríguez-Doncel 2019). This legal compass (the Linked Open Data cloud) connects more than 3000 (i) reflects and endorses the two sides of the rule of law datasets, with more than 84 billion triples. The number (binding power and social dialogue), (ii) and assumes as of data sources doubles every three years (Zou and Özsu 2017). This has created a data space, strongly intercon- nected, in which law and government can have a leading role, as legal documents and, mostly, legal content and knowledge, are increasingly offered for public consumption (Casanovas et al. 2016). Linguistic resources for the legal domain are increasingly being identified, classified and annotated (Martín-Chozas et al. 2019, Rehm et al. 2020). Ontolex-lemon, the vocabulary for lexical resources in the Web of Data, have been extended to create databanks of legal terminologies that can be used automatically by dictionaries (Rodríguez-Doncel et al. 2015). However, new legal issues arise, such the use by LOD of crowdsourced vocabularies, where there is no author- ity imposing one interpretation over another. There is FIGURE 1. Legal compass for the rule of law. Source: Poblet, no evidence so far of case-law nor out-of-court disputes Casanovas and Rodríguez-Doncel (2019) 4 A triple is a set of three entities that codifies a statement about semantic data in the form of subject–predicate–object expressions. 5 See a short list of the most notorious cases at https://en.wikipedia.org/wiki/Data_breach 27 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 a pre-requisite that they can be partially coded through semantic representations. In the scheme of Fig. 1, institutional strengthening and trust are intended and eventually produced through a vari- ety of sources that must be ordered beyond a determined threshold to build valid regulatory systems. Validity, and specifically legal ecological validity, emerges from the degree of compliance with several requirements—e.g. enforce- ability, efficiency, effectiveness, fairness (justice)—that make the system acceptable and sustainable. Thus, validity qualifies as a second order property that encompasses the whole regulatory system (not only a specific rule or norm). This approach is also compatible with recent surveys on business languages and compliance (Hashmi FIGURE 2. Relationship between Anonymity, Unlinkability and 2018) but shows that legal compliance requirements can Disclosure. Source: Torra (2017, p. 9). Quoted with permission be more complex than those set by regulatory languages. 2019b) and (ii) to the existing data protection engineer- 2.2 A PRELIMINARY FRAMEWORK ing approaches. As already stated, there are many differences be- Following Pfitzmann and Hansen (2010), Torra (2017) tween the USA and EU approaches for regulating privacy. organized their terminology for data minimization into Whereas EU laws consider ‘privacy’ as a human right, four interrelated categories, depicted in Fig. 3 as Venn a constitutional fundamental right, and a fundamental diagrams. Henceforth, Torra extended and refined their EU right under the Article 7 and 8 of the EU Charter of original data minimization strategy. Fundamental Rights (2000 and 2012) (González-Fuster 2014, Blasi 2016),6 US “values it as a liberty over and Pfitzmann and Hansen (2010) originally provided the against the state” (Blasi 2014). Thus, data protection of following three definitions: (i) Anonymity of a subject personal data should be differentiated from privacy tout means that “the subject is not identifiable within a set of court. In contrast, as noticed by many scholars, the United subjects, called ‘anonymity set’”; (ii) From an adversary States does not provide for an overall legal expectation of (intruder, attacker…) perspective, anonymity of a subject privacy. The collection and processing of personal data is means that “the adversary cannot achieve a certain level regulated based on the type of data at stake. Thus, data of identification for the subject s within the anonymity related to healthcare is subject to the Health Insurance Portability and Accountability Act (Kennedy–Kassebaum set”; (iii) Unlikability of two or more items of interest Act, 1996) commonly known as HIPAA, and financial data (IoI) from an attacker’s perspective means that “within is governed by the Financial Services Modernization Act the system (comprising these and possibly other items), (Gramm-Leach-Bliley Act, 1999), known as GBLA.7 the attacker cannot sufficiently distinguish whether those Legal requirements might be quite detailed, and differ- IoIs are related or not.8 Torra (ibid. 10) provides three ent according to national and jurisdictional frameworks. additional ones: (iv) Disclosure, that “takes place when Related to privacy, to link the knowledge graph approach attackers take advantage of the observation of available to legal governance, we will endorse a broader conceptual data to improve their knowledge on some confidential in- stance, linking the main concepts to be modelled (i) to formation about an IoI”; (v) Identity Disclosure, “when the the middle-out approach defined by Pagallo et al. (2019a, 6 The Charter of Fundamental Rights of the EU Union (2000-2012) states in Art. 7: “Everyone has the right to respect for his or her private and family life, home and communications”. Art. 8.1 reads: “Everyone has the right to the protection of personal data concerning him or her”, and Art. 8.2: “Such data must be processed fairly for specified purposes and on the basis of the consent of the person concerned or some other legitimate basis laid down by law. Everyone has the right of access to data which has been collected concerning him or her, and the right to have it rectified.” 7 https://www.hipaajournal.com/comparison-of-european-and-american-privacy-law/ 8 It is worth noting that unlinkability is deemed a sufficient non-necessary condition, as it implies anonymity. However, Torra (2009, 9) points out that there might be a case in which linkability is possible, but anonymity is not. Law in Context, Vol 37, Issue 1, 2020 28 ISSN: 1839-4183 TABLE 1: Fair Information Principles Practices. Source: Langheinrich (2001). There should be no secret record keeping. This includes both the publication of the existence of such 1. Openness and transparency collections, as well as their contents. 2. Individual participation The subject of a record should be able to see and correct the record. 3. Collection limitation Data collection should be proportional and not excessive compared to the purpose of the collection. 4. Data quality Data should be relevant to the purposes for which they are collected and should be kept up to date. 5. Use limitation Data should only be used for their specific purpose by authorized personnel. 6. Reasonable security Adequate security safeguards should be put in place, according to the sensitivity of the data collected. 7. Accountability Record keepers must be accountable for compliance with the other principles. adversary can correctly link a respondent to a particular TABLE 2. Privacy Design Strategies. Source: from Hoepman record in the protected data set”; (vi) Attribute Disclosure (2014) and ENISA (2015) “when the adversary can learn something new about an attribute of a respondent, even when no relationship can be established between the individual and the data”. These concepts stand on the shoulders of the Fair Information Practice Principles (FIPPs) proposed by US Secretary's Advisory Committee on Automated Personal Data Systems in a 1973 Report, Records, Computers and the Rights of Citizens. This Report was followed by the US Privacy Protection Study Commission Report on Personal Privacy in an Information Society (1977). Legal experts—Alan Westin (1967)—, AI and law experts—such as Layman E. Allen—, and computer scientists—Willis H. Ware (RAND Corporation)—were involved in their development. Table 1 summarises the so-called FIPPs. In 2004, Kim Cameron, Chief Identity Architect of Microsoft, wrote and blogged what he would call the “7 Laws of the Internet”. His approach was to set a metasystem identity layer, i.e. “to develop a formal understanding of (PbD), the process for embedding privacy principles into the dynamics causing digital identity systems to succeed design specifications architectures. However, computa- or fail in various contexts, expressed as the Laws of Iden- tional modelling, the specific engineering paths to make tity. Taken together, these laws define a unifying identity it happen was—and still is—a model to be assembled or metasystem that can offer the Internet the identity layer to be built. Hoepman (2018) puts it in the following way: it so obviously requires.9 “Privacy by design is a system development philoso- In the following years, Ann Cavoukian (2006, 2007) phy that says that privacy should be taken into account aligned this Internet Identity Metasystem Layer with throughout the full system development lifecycle, from FIPPs. The result was the proposal of Privacy by Design its inception, through implementation and deployment, 9 “We need a unifying identity metasystem that can protect applications from the internal complexities of specific implementations and allow digital identity to become loosely coupled. This metasystem is in effect a system of systems that exposes a unified interface much like a device driver or network socket does. That allows one-offs to evolve towards standardized technologies that work within a metasystem framework without requiring the whole world to agree a priori.” Cameron (2004). 29 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 all the way until the system is decommissioned and no than a final solution, we show the essential aspects that longer used. In software engineering terms this makes need to be addressed in any representation that is incor- privacy, like security or performance, a software quality porated in Knowledge Graphs to support the regulatory attribute or non-functional requirement.” framework around Privacy and Transparency. Hoepman (2014) represented the existing engineer- ing strategies (see also ENISA 2015) as design strate- 3. BACKGROUND CONCEPTS gies that describe a fundamental approach to achieve Before detailing our framework, we briefly present a certain design goal. Thus, a privacy design strategy our definition for each of the basic concerns. We have could be understood as a "design strategy that achieves detailed each of them in other publications, as referenced (some level of) privacy protection as its goal". The PbD in each sub-section. analytical framework can be mainly set through eight strategies: minimize, hide, separate, aggregate, inform, 3.1 TRUST control, enforce and demonstrate (Table 2). The strate- The issue of trust has been prevalent in the Internet gies are assembled into two classes: (i) the first four since its popularization in the early 90s—see the Grandi- are data-oriented strategies and (ii) the latter four are son and Sloman (2000) survey. Attention has focussed on process-oriented strategies. The European Union Agency the lower layers of the Internet Architecture, emphasizing for Network and Information Security (ENISA) embraced authentication, which deals primarily with verification of this approach in 2015, as there was an increasing need to identity. More recently, with the advent of the Web and create a common framework for policy makers, legisla- social networks, the cybersphere and society generally, tors, service providers, data protection authorities and have become heavily influenced by information (and misin- standardization bodies. It is similar to the framework formation) that flows in news sites and social networks in eventually assumed in GDPR provisions. Colesky, Hoep- the Internet. There are many studies carried out in several man and Hillen (2016) correlated and map the strategies disciplines attempting to characterize and understand against privacy patterns, adding the notion of ‘tactics’ as the spread of information in the cybersphere, as well as an additional level of abstraction between them. how this affects society—see Marwick and Lewis (2017) Nevertheless, this remains an open field. We have al- for an overview. ready quoted the extension of the Pfitzmann and Hansen The original vision for the Semantic Web included (2010) minimization strategy by Torra (2017). There are a “Trust” layer, although its emphasis was more on au- many other possibilities. For instance, Rastogi, Gloria thentication and validation with static trust measures for and Hendler (2017) have developed a general method data. There have been many efforts in representing trust, to enhance privacy in the cloud, deploying mobile appli- including computational models.10 As proposed initially cations dynamically on a scalable on-demand hardware in Almendra and Schwabe (2006), and later in Laufer and software platform. From a legal governance perspec- and Schwabe (2017) and Schwabe, Laufer and Busson tive, Casanovas et al. (2014) have proposed an indirect (2019), the approach used here is based on the work of strategy for security platforms in which coding should Gerck (1997) and Castelfranchi and Falcone (2001), tak- be combined with the construction of ad hoc ‘anchoring’ ing the view that trust is “knowledge-based reliance on institutions to monitor and control the outcomes of the received information”, that is, an agent (i.e., a person or a information processing flow. software program) decides to trust (or not) based solely In the following sections, we illustrate how these on her/his knowledge, and the decision to trust implies considerations can be supported using Semantic Web the decision to rely on the truth of received or on already formalisms to represent the relevant dimensions. Rather known information to perform some action. 10 A general survey can be found in Pinyol and Sabater-Mir (2013); Artz and Gil (2017) presents an excellent earlier survey for the Semantic Web; and Sherchan, Nepal, and Paris (2013) surveys trust in social networks. In the Linked Data world, it is clear that facts in Semantic Web should be regarded as claims rather than hard facts (Bizer and Cyganiak 2009), which naturally raises the issue of trust on those claims. Law in Context, Vol 37, Issue 1, 2020 30 ISSN: 1839-4183 In terms of a Knowledge Graph, an agent wishing to 3.3 TRANSPARENCY perform an action must first filter those information Generally speaking, according to Meijer (2013, 430) and items it deems “trusted”, i.e., it will use them to perform others (Schudson 2015), transparency can be defined as the intended action. Since it is not possible to “half-act”, “the availability of information about an actor that allows in this sense trust is binary—either the agent trusts the other actors to monitor the workings or performance of information, or it does not. A more extensive discussion the first actor.” It contemplates the “capacity of outsiders can be found in Schwabe, Laufer and Busson (2019). to obtain valid and timely information about the activities Trusted information as the basis for supporting privacy of government or private organizations”.12 Transparency and transparency is discussed next. presupposes the involvement of an observed and an ob- server (Berstein 2017). In contrast to privacy, that is con- 3.2 PRIVACY cerned with information about individuals, transparency As discussed in section 2.2, there are many definitions concerns any type of information, although it may make for privacy.11 Tavani (2007) classified privacy theories a difference if the producer (author) is an organization into four broad types: the nonintrusion, seclusion, limi- or an individual (Heimstäd and Dobusch 2018). tation, and control theories. For our research, we adopt Transparency is also related to privacy. In the so-called the Restricted Access/Limited Control (RALC) Theory “Attention Economy” (GDPR, CE 2016), for example, the proposed by Moor and Tavani (2001). RALC presupposes information about users and consumers are a primary that an adequate theory of privacy needs to differentiate source of value, and companies actively seek to obtain as the concept of privacy from both the justification and the much information about users as possible. This can be in management of privacy. Accordingly, the RALC framework direct conflict with privacy rights of users, who have the has three components: an account of the concept of privacy, right to control how the information about them is used an account of the justification of privacy, and an account by others. In this state of affairs, transparency can support of the management of privacy. “RALC requires that one the resolution of potential conflict with privacy policies: must have protection from intrusion and interference the disclosure of information about the company’s pro- and information access, it addresses concerns not only cesses and procedures associated to the individual can about protecting informational privacy (as described in contribute to foster trust, showing that it is compliant the control and the limitation theories) but also about with these regulations and thus to stimulate authorization protection against the kinds of threats described in the on the part of the users. This has been subject to recent nonintrusion and the seclusion theories as well.” (Tavani regulations such as the General Data Protection Regula- 2007, 10). tion enacted by the European Union. Based on this theory, we define Privacy as “controlled From a more abstract approach, both Privacy and access to information related to an agent”. In order to Transparency relate to controlling actions over informa- ensure privacy, it is necessary to answer three questions: tion, and who can define such controls. As such, they can • Q1: What types of Actions are allowed (and controlled) be regarded as two points in the same control dimension. over Knowledge Items (KIs)? Privacy tends to limit or restrict actions over information • Q2: What are the relation types between some Agent items, whereas Transparency tends to allow (in some and a KI that entitle this Agent to establish a Privacy cases, mandate) actions over them, which explains the Rule governing Actions over that KI? natural tension that exists between the two. • Q3: How to resolve conflicts between applicable rules? 11 See for example Smith, Dinev and Xu (2011). Paci, F., Squicciarini, A., Zannone (2018) and Such and Criado (2018) present surveys on multi-party privacy. 12 https://www.britannica.com/topic/transparency-government#accordion-article-history 31 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 4. AN ILLUSTRATION - DISASTER RELIEF between transparency and privacy rules. George’s relation DONATION with ReliefOrg should be accessible in the KG, according The goal of this example is to illustrate the complex to not-for-profit legislation. On the other hand, George’s and interdependent nature of trust, privacy and transpar- privacy rules would prevent this access. Since transpar- ency; it will later be used to show the expressive power ency legislation in this case has higher precedence than of the framework and how it incorporates the various personal rules, this association can then be used by Ed’s concerns at play. rules, and therefore ReliefOrg would not be accepted as Consider a scenario in which a disaster has occurred, a recipient of donations by Ed. such as the fire that burned the roof of the Notre Dame Consider now a slightly different scenario where George Cathedral, or the bush fires in Australia. In response to is not an officer of ReliefOrg, and Ed wants to check the several demands, Ed wants to donate some money to help financial integrity of ReliefOrg. He retrieves the financial with the disaster relief actions. Ed has received several report for ReliefOrg from the KG, but wants to make sure donation requests from different organizations and needs it has been audited, so he looks for a certification of the to choose one of them to make the donation. However, financial report. He finds out it has been audited by Au- given his past experience, Ed wants to make sure these ditInc, which is unknown to him, so he needs to verify that are legimate request as opposed to frequently occurring it has an accreditation certificate from a public authority. scams. In other words, he wants to make sure that the If such a certificate is available, Ed analyses the financial donated money will actually be used for the relief actions, report finding nothing wrong in principle. rather than being misused, e.g., funding the organization’s However, tipped by a friend, Ed learns that George may basic infrastructure, or employed in another action, or in fact be one of the owners of AuditInc. He then checks even pocketed by unscrupulous officers of the receiving AuditInc to see if its owners are listed, and whether George organization. is one of them. Given George’s privacy rule plus the fact Ed formulates a rule saying that he only trusts organiza- that AuditInc is not a not-for-profit organization, his pri- tions that openly publish who are their financial officers, vacy rule would prevent access to the owner relationship, and their financial records. Financial records must be so Ed would not know that George is one of the owners, validated by accredited audit organizations. Furthermore, thus deciding to contribute. Note that in this scenario, because of personal reasons, Ed does not want to contrib- George’s privacy rule would apply not only to AuditInc’s ute to an organization in which George is an officer. This information, but also to information furnished by others, rule can be regarded as an application-related rule, akin for example, a photo in a social network where George to what is referred to as a “business rule” in traditional appears in the annual Christmas party of AuditInc with software development. a caption mentioning his role as one of the partners, or Let us assume that there is a law that stipulates that perhaps with a badge on his neck identifying him as such. not-for-profit organizations must publicly identify their We next describe our proposed KG Usage framework officers and that George is an officer of ReliefOrg, an NGO and subsequently analyse this example showing how it dedicated to raising funds for and helping disaster relief can be represented by it. efforts. George, being a very reserved person, has a privacy rule that stipulates that his association with any organi- 5. A SUMMARY OF THE KG USAGE FRAMEWORK. zation, including ReliefOrg, should not be made public. Figure 3 shows a diagram of the use of information within Ed receives a request for donation from ReliefOrg and a KG. “Using a KG” is represented as a Request made by needs to decide whether he should donate or not. The first some Agent for an Action over a Knowledge Item (KI). We step Ed follows is to verify if the officers of ReliefOrg are assume the existence of an underlying RDF graph, which published in the KG.13 Here we can see a potential conflict would be equivalent to the “traditional” definition of KGs. 13 See for instance https://permid.org or https://opencorporates.com as examples of KGs with this type of information. Law in Context, Vol 37, Issue 1, 2020 32 ISSN: 1839-4183 An RDF graph is simply a collection of statements, each The provenance graph of the nanopublication will in the form of a triple such as contain provenance information about the assertions . An RDF named graph in the assertion graph (e.g.; what image or natural lan- is one for which there is an identifier to refer to it.14 The guage processing software was used, recorded location actual KG is formed by defining several named graphs info, whether the assertions were inferred using some over this underlying RDF graph as a way to structure inference engine, etc.). The provenance graph can be the RDF triples into Knowledge Items, similarly to Items, used to represent, to the desired level of detail, the sup- Statements and Qualifiers in Wikidata (Erxleben et al. porting information for the assertions. For example, if 2014). Thus, a Knowledge Graph represents a collection an automated face recognition algorithm was used, the of interlinked descriptions of KIs – real-world objects, provenance information represented in the provenance events, situations or abstract concepts – where: graph of the nanopublication may inform which algorithm • Descriptions have a formal structure that allows was used, which parameters were used in this particular both people and computers to process them in an case, and a confidence factor of the algorithm about the efficient and unambiguous manner; correctness of the extraction. Another use of provenance can be seen in the case of a statement stating that, for • Entity descriptions refer to one another, forming a example, placeOfBirth . The network, where each entity represents part of the provenance information may include documentation to description of the entities related to it.15 support its truthfulness, such as a reference to a birth We propose to represent the KG as a collection of certificate that states that indeed the place of birth of Knowledge Items (KIs), each of which as a nanopublica- Barack Obama is Hawaii. tion16 (Groth, Gibson and Velterop 2010). A nanopublica- The publication info graph will contain metadata about tion “offers a supplementary form of publishing alongside traditional narrative publications”, consisting of three parts representable by RDF graphs: “(i) an assertion (a small, unambiguous unit of information), (ii) the provenance of that assertion (who made that assertion, where, when, etc.), (iii) the provenance of the nanopublication itself (who formed or extracted the assertion, when, and by what method).” (Golden and Shaw 2016) 5.1 KNOWLEDGE ITEM REPRESENTATION A KI, as all nanopublications, comprises an assertion graph, a provenance graph and a publication info graph. The assertion graph of a KI contains a set of asser- FIGURE 4. Request Evaluation Algorithm tions about its content. The assertions in this graph are a subset of the assertions in the underlying RDF graph. the creation of the KI itself (as opposed to the information As an example, if a KI refers to an event, the assertion contained in its assertions sub-graph), such as its author, graph would contain statements about the participants creation date, etc… and their roles, location information, date information, depictions (photos, videos, …) and so on. 14 https://www.w3.org/TR/rdf11-concepts/#section-rdf-graph 15 https://www.ontotext.com/knowledgehub/fundamentals/what-is-a-knowledge-graph/ 16 http://nanopub.org . “A scholar can promote small pieces of information within her work using the practice of nanopublication. Nanopublications include useful and usable representations of the provenance of structured assertions. These representations of provenance are useful because they allow consumers of the published data to make connections to other sources of information about the context of the production of that data.” (Golden and Shaw 2016) 33 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 5.2 CONTROLLING USAGE In our framework, Privacy and Transparency refer to Request for actions over some information (in a KI), for which an Authorization must be granted, according to the Rules set forth by the relevant stakeholders. Stake- holders include persons “related” to the KI, as well as institutional agents such as “the State” (whose rules are stated as laws). Rules may be based (make use of) on any information available in the KG. We refer to both Privacy and Transparency rules collectively as Usage Rules. Figure 4 introduces the basic algorithm to evaluate a Request. The first step in granting an Authorization for the action is to determine which KIs in the KG are trusted by, the Agent requesting the Action. This is realized by a Trust Engine that collects applicable rules and includes them in a Trusted sub-graph of the KG. We refer to the set of trusted KIs by Agent α as the Trusted KG(a). Once the Trusted KG has been determined, a set of usage rules are evaluated, which entails determining applicable rules (UsageRuleSet), evaluating them (EvalUsageRuleSet) and resolving conflicts if they arise. In order to evaluate a usage rule, it is also necessary to evaluate the trust rules FIGURE 5. Usage process for information in a KG of the author of that usage rule, which needs to be based on trusted KIs. Thus, a set of Trusted KG (ai) is defined, In order to apply a set of usage rules RS for a request one for each Author ai of an applicable usage rule. The for an Action on a KI by an Agent (as specified in step 2. result of this process defines the final Authorization for the above), one must perform the following steps: Action in the Request that was made. If the Authorization 4. Sort (reorder) the rules in RS in decreasing order of is “Allowed”, the Action is carried out using the TrustedKG precedence (i.e., the rule with highest precedence of the Agent that requested the Action. comes first) Stated in words, the algorithm stipulates that, to evalu- 5. Assume there is a default authorization – i.e., if noth- ate a request for an Action by an Agent over a Knowledge ing is said, an Action is allowed (or denied) by default. Item (KI), the following steps should be performed (line numbers correspond to those in Figure 4). 6. Consider each rule R in the set RS in turn. For each one, • Determine the authorization value by applying this 1. Collect the applicable rules referring to the KI and rule R for the Action on the KI requested by Agent. store them in variable RS • If the authorization value can be determined, then 2. Apply the collected rules in RS for the requested Action return it as the result of the algorithm. It is possible by Agent on the KI. that the relevant rules are unable to determine an 3. If the Action was allowed, then collect the trust rules authorization value, e.g., because of insufficient for Agent and store them in variable TS, then use these information in the KG. collected rules to find out the set of trusted informa- 7. If no value could be determined, return the default tion items by Agent in the KG (stored in TGA), and authorization as the result of the algorithm. perform the requested Action on the KI, using the trusted information in TGA Law in Context, Vol 37, Issue 1, 2020 34 ISSN: 1839-4183 To determine the final Authorization value, a conflict Rules are of the form antecedent => consequent, both resolution strategy must be employed, which is in turn of which are sets of statements (Almendra and Schwabe subject to Governance Rules. This algorithm abstracts the 2016). The antecedent of privacy rules may refer to any essential decisions that must be made, to wit: statements in the KG, including 1. Who can formulate a rule for a given KI? – in line 1; • Any statements in the subgraphs in the KI’s nanopub- 2. How are conflicts between rules resolved? – in line 5-8 lication (assertions, provenance and publication info); 3. What are the allowed actions over KIs? – in line 4. • The identity of the agent requesting permission; We detail possible answers to these questions in the • The type of Action; following sub-sections. • Information in the KG serving as contextual informa- tion, such as 5.3 RULES • Date/time of the request; The UsageRuleSet(KI) function call determines what • Current location of agents involved; are the applicable rules given a KI. For trust rules, it is Seen as nanopublications, the actual specification of determined by the author of the request. For privacy the rule is given in its assertion graph, using a notation rules this corresponds to answering the question “who such as N3Logic (Berners-Lee et al. 2006) or SWRL.17 has the right to define a Privacy Rule that controls actions Governance Rules are Rules that include other Rules (in over this KI?” The definition of Privacy itself indicates either of antecedent and consequent), so in this sense that it must be any agent that is somehow related to the they are meta-rules – i.e. rules about rules. information contained in the KI. Any useful instantiation of the framework must spell out what are the accepted 5.4 CONFLICT RESOLUTION relation types, which can include: The Sort-by-Precedence (RS, decreasing) function • An identification property for any agent that is included call sorts the enabled rules in descending precedence in the KI – for example, some person appearing in a order, typically combining several kinds of information posted photo or video; to establish order relations among rules. • Any relation denoting referral to a person included Some possible complementary order relations that in the KI – for example, some person cited in a post; can be employed are: • Any creation or authorship relation; • Hierarchical relations between users – rules defined by a higher-ranked user take precedence over rules • Any agent related to the creation of the KI – for ex- defined by lower-ranked ones. For example, rules ample the author of a video posted by someone else; established by laws (authored by state agents) take • Any agent who has legal jurisdiction over an agent precedence over rules stated by individuals (com- identified or mentioned in the KI; mon citizens); • The Agent representing the legal system(s) that • Hierarchical relations between relation types. Rules has(have) jurisdiction over the KG, over the Agent defined by users related to the KI through a higher- or over the Action request. ranked relation type take precedence over rules The presence of such relations can directly occur in defined by users related via a lower-ranked relation the KG (i.e., as a typed edge), or as a composition of valid type. For example, one may state that relation type relations. Furthermore, the rules may (or may not) allow “identifies” takes precedence over type “mentions”. the use of inferred relations in the KG. Thus, in a video where a person A is identified, and person B is mentioned in a conversation, person A’s rules would take precedence over person B’s rules. 17 https://www.w3.org/Submission/SWRL/ 35 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 Since hierarchies are partial orders (which means that officer . officer there may be items without an ordering between them), . they may not define precedence completely, so further RuleEd1 conflict resolution strategies are still needed. Such and {?O type org:Organization; hasOfficial: ?Ofc; hasFinan- Criado (2018) identified six categories of strategies that cialRecord ?FR. ?Ofc type foaf:Person. can be employed. Most require user involvement at run- ?FR assertions ?FRa. ?FRa log:semantics ?FRaS; ?FRAS time, but the aggregation-based class of strategies can be easily incorporated into an algorithm. Strategies in this log:includes class define an aggregation function such as consensus, { ?FR auditedBy ?Aud. ?Aud type AuditCo}. majority, minimum fixed number of votes, permit-overrides, ?FR provenance ?PFr. ?PFr log:semantics ?PFrS. ?PFrS deny-overrides, etc, and replace the set of conflicting rule log:includes {prov:hasPrimarySouce ?DOCS}. ?DOCS as- results by a single aggregated result. sertions ?DOCSa. ?DOCSa log:semantics ?DOCSaS. An alternative to the aggregation approach is to de- ?DOCSaS log:includes {AuditCo certifiedBy ?CA. }. compose the KI into finer-grained elements so that each author one is subject to only a single rule. This makes sense log:semantics ?TGEd. ?TGEd log:includes {?CA type when the Action to be performed can be stated as the CertificationAgency}} composition of the same operation performed on each => { :add {?O type:NonProfit}. } element independently. For the elements where the Ac- tion is denied, the resulting composite object is altered. For example, in a group photo showing several people, Stated in plain language, the KG contains statements some may allow publication other may deny publication. affirming that NonProfit, AuditCo and CertificationAgency In this case, it would be possible, to blurr out the face of are a kind of Organization, and Donate is an Action. It also the persons who have denied publication. affirms that ReliefOrg is a NonProfit organization; AuditInc is a Certification Agency and that George is an officer of 6. EXAMPLE SCENARIO REVISITED both ReliefOrg and AuditInc. In this section we re-examine the illustrative example RuleEd1 reflects Ed’s trust criteria – Ed’s Trusted Graph scenario in light of the KG usage framework, showing will include a NonProfit organization only if its financial how the most important aspects are represented. The records include a list of its officers, and the financial re- first rule captures Ed’s requirement about Non-Profits: cord itself is audited by a certified Certification Agency. he only trusts organizations that openly publish who This is a trust rule since it refers to KIs which are are their financial officers, and their financial records. needed as input for Ed to allow taking the “Donate” action. Financial records must be validated by accredited audit Ed’s second rule is in fact a business rule – he does not organizations. want to contribute to an organization in which George is The trust rule below captures this, expressed using an officer. While this could be embedded in application N3Logic with extensions. We state under KG some state- code using the KG, we express it here as a rule as well, to ments we assume to be present in the KG: facilitate the discussion. KG RuleEd2 NonProfit subClassOf org:Organization. AuditCo sub- { author ; log:semantics ?TGEd. ClassOf org:Organization. ?TGEd log:includes CertificationAgency subClassOf org:Organization. {?O type:NonProfit, officer:}. }. type NonProfit. type ruleEd1 assertions ?ARule1. ?Arule1 log:semantics . ?Arule1S.?ARule1S log:includes Law in Context, Vol 37, Issue 1, 2020 36 ISSN: 1839-4183 {{ type Donate, recipient . intends } Stated in words, RuleTransp (which assumes there is a => { type Authorization, rule , action statement affirming that NonProfitAct is a Law) says that , value “Denied”}} any operation to read who are the officers of an organiza- Stated in words, RuleEd2 stipulates that if Ed’s trusted tion of type NonProfit will be allowed. Furthermore, it also graph contains an Organization for which George is an states that the provenance graph of RuleTransp contains officer, and Ed intends to take an action of type “Donate” a reference to the proper legal document, the text of the whose recipient is this Organization, then the request law () that is being interpreted by this rule. will be denied. In order to manage the conflict between RuleGeorge1 Next, we look at George’s privacy rule. It is a privacy and RuleTransp, there is a meta-rule stipulating that the rule because it refers to an action over personal informa- latter has precedence over the former. This precedence tion about George. relation is used in the Sort-by-Precedence function call RuleGeorge1 in line 5 in Fig. 4. {G :is {?O type:NonProfit; officer:. author ; log:semantics ?TGG. KG ?TGG log:includes {?O officer:. PersonalPrivacyRule subClassOf Rule. Legislation author }. subClassOf Rule. ruleGeorge1 assertions ?AGeorge1. ?AGeorge1 MetaRule1 log:semantics ?AGeorge1S. ?Ageorge1S log:includes { assertions ?AMR1. ?AMR1 log:semantics { type Read; object ?G. ?A intends } ?AMR1S. ?AMR1S log:includes => { type Authorization; rule ; {{?R1 type PersonalPrivacyRule. ?R2 type Legislation} action ; value “Denied”}} => {?R2 precedes ?R1}} Stated in words, RuleGeoge1 says that if a read operation Stated in words, this rule says that rules of type Legislation is intended over the KG containing the information that have precendence over rules of type PersonalPrivacyRule. George is an officer of some organization, the authoriza- We must also define more precisely what is a Personal- tion for that operation will be denied. PrivacyRule – it is any rule that uses a PersonalInformation The stipulation expressed in the transparency legislation property in its antecedent, which must also be defined. for non-profits to divulge its officers can be expressed as KG ?RuleP assertions ?RulePA.?RulePA log:semantics NonProfitAct type Law. ?RulePAS;?RulePAS log:includes {{?p1 ?r ?p2. RuleTransp (?p1 rdf:type Person OR ?p2 rdf:type Person)}} { author . as- => {?r rdf:type PersonalInformationRelation} sertions ?ATransp. ?ATransp log:semantics ?ATranspS. ?ATranspS log:includes{?G is {?O type Nonprofit; officer ?Ofr. Stated in words, this rule says that any relation involv- type Read; object ?G. ?A intends } ing a person is a PersonalInformationRelation. This is => { type Authorization; rule ; ac- admittedly a simplification, as in actual situations these tion ; value “Allowed”}}. relations that characterize “personal information” should provenance ?PTransp. ?PTransp be further elaborated. log:semantics ?PTranspS. ?PTranspS log:includes { prov:HasPrimarySource } 37 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 {?RulePP assertions ?RulePAS. ?RulePPAS log:semantics careful policies, loopholes in the regulations could allow ?RulePPASS; undesired actions to take place. ?RulePPAS log:includes{?RulePP antecedent ?RuleP- As ongoing and future work, we are investigating PAA. ?RulePPAA log:semantics ?RulePPAAS implementation architectures to allow efficient and scal- ?RulePPAAS log:includes {{?p ?r ?q.(?r rdf:type able usage control over existing KGs, as well as exploring PersonalInformationRelation)} the applicability to various domains. This architecture => {?RulePP type PersonalPrivacyRule}} can also be fitted into wider legal governance models that have been developed in the literature in the last five years and into more general Online Dispute Resolution Stated in words, this rule says that a Personal Privacy governance models (Ebner and Zeleznikow 2016). While Rule is a rule that involves (uses) a Personal Information the algorithm and the example presented here consider Relation in its antecedent. Notice that this is also a meta- that the authorization for an action would be determined rule, since its objects are rule themselves. in an automated fashion, it is entirely possible a version The second scenario does not require any additional where the algorithm could reach a state in which it is rules. It simply results in an “allowed” authorization, not able to resolve conflicts between rules. At this stage, because no conflict arises between RuleGeorge1 and it could refer the information and their supporting facts RuleTransp when applied to , since it is not of to a human being, to allow them to bring in additional type NonProfit, and therefore RuleTransp is not applicable. criteria which would be outside the scope of the rules in This illustrates a possible loophole in the legislation which the KG to reach a final decision. We are investigating the could be avoided if the NonProfit legislation prohibited an extension of the framework to accommodate this mode. organization from being audited by another organization having common officers. 8. REFERENCES 1. A. Hogan, E. Blomqvist, M. Cochez, C. d’Amato, G. de Melo, C. 7. CONCLUSIONS Gutierrez, J. E. L. Gayo, S. Kirrane, S. Neumaier, A. Polleres, R. We have presented a usage framework explicating Navigli, A. N. Ngomo, S. M. Rashid, A. Rula, L. Schmelzeisen, the various types of specifications that must be made to J. F. Sequeda, S. Staab, and A. Zimmermann. “Knowledge capture privacy, transparency and trust concerns. The Graphs”. arXiv. abs/2003.02320. 2020. framework also provides a better understanding of the 2. Almendra, V. D. S., and Schwabe, D. 2006. Trust policies relations between these concerns. for semantic web repositories. In Proceedings of 2nd In- ternational Semantic Web Policy Workshop (SWPW’06), Trust entails determining which data items will be used at the 5th International Semantic Web Conference, ISWC, to perform an action; privacy and transparency involve pp. 17-31. controlling who can perform an action over a data item. 3. Angella J., Kim, K. and Johnson, K.P. 2016. “Power of Trust is thus more fundamental, as privacy and transpar- consumers using social media: Examining the influences ency rules must be based on trusted data. Furthermore, of brand-related user-generated content on Facebook”, privacy and transparency are approached as being differ- Computers in Human Behavior, 58: 98-108. http://dx.doi. ent points along the control dimension, thus explaining org/10.1016/j.chb.2015.12.047 . the natural tension between the two. 4. Artz, D., Gil, Y. 2007. “A survey of trust in computer sci- We have shown how legal requirements, and other types ence and the Semantic Web”, Web Semantics: Science, Services and Agents on the World Wide Web, 5 82): 58- of norms, which ultimately regulate the functioning of any 71. http://www.sciencedirect.com/science/article/pii/ application that uses the KG, can be incorporated into the S1570826807000133 KG itself. The various nuances and interdependencies of 5. Australian Government. 2019. Department of the Treasury. these concerns were illustrated in a running example. One Regulating in the digital age: Government Response and interesting point in the example is the fact that in spite of Implementation Roadmap for the Digital Platforms Inquiry Law in Context, Vol 37, Issue 1, 2020 38 ISSN: 1839-4183 (Government Response, December). https://treasury.gov. 18. Castells, M. 2010. The Rise of the network society (2 ed.). au/publication/p2019-41708 Cambridge, MA, USA: Blackwell Publ. Inc. 6. Australian Government. 2020. Attorney-General’s Depart- 19. Cavoukian, A. 2006. “7 Laws of Identity: The Case for ment. Privacy Act Review. Issues Paper. October, https:// Privacy-Embedded Laws of Identity in the Digital Age”, apo.org.au/sites/default/files/resource-files/2020-10/ Technology, Ontario Information and Privacy Commis- apo-nid309253.pdf sioner, October, pp. 1-24. 7. Berners-Lee, T., Connolly, D., Kagal, L., Scharf, Y., Hendler, J. 20. Cavoukian, A. 2010. “Privacy by Design. The 7 Foundational 2008. “N3Logic: A Logical Framework for the World Wide Principles. Implementation and Mapping of Fair informa- Web”, Theory Pract. Log. Program. 8: 249–269. tion Practices. Information and Privacy Commissioner, 8. Bernstein, E. 2017. “Making Transparency Transparent: Ontario, Canada. The Evolution of Observation in Management Theory”, 21. Colesky, M., Hoepman, J.H. and Hillen, C. 2016, “A critical Academy of Management Annals 11 (1): 217 – 66. analysis of privacy design strategies”. In 2016 IEEE Security 9. Bizer, C. and Cyganiak, R. 2009. “Quality-Driven Informa- and Privacy Workshops (SPW), IEEE, pp. 33-40. tion Filtering Using the WIQA Policy Framework”, Web 22. Council of the European Union, European Parliament, Semantics: Science, Services and Agents on the World Wide 2016. Regulation (EU) 2016/679 of the European Parlia- Web. 7: 1-10. http://www.websemanticsjournal.org/index. ment and of the Council. General Data Protection Regula- php/ps/article/view/157/155 tion”, https://eur-lex.europa.eu/legal-content/EN/TXT/ 10. Blasi-Casagran, C. 2014. “The EU Role in Shaping Global PDF/?uri=CELEX:32016R0679 . Data Privacy Standards”, PrivOn2015 https://sites.google. 23. Davenport, T.H. and Beck, J.C. 2001. The attention economy: com/site/privon2015/accepted-papers Understanding the new currency of business. Harvard 11. Blasi-Casagran, C. 2016. Global data protection in the field Business Press. of law enforcement: an EU perspective. London: Routledge. 24. Ebner, N. and Zeleznikow, J. 2016. “No sheriff in town: 12. Cameron. K. 2004. December- “The Seven laws of the Inter- governance for online dispute resolution”, Negotiation net”. http://www.identityblog.com/stories/2004/12/09/ Journal, 32(4): 297-323. thelaws.html 25. Ehrlinger, L. and Wöß, W. 2016. “Towards a Definition 13. Carlson, A., Betteridge, J., Wang, R.C., Hruschka Jr., E.R., of Knowledge Graphs”, SEMANTiCS, CEUR 1695, http:// Mitchell, T.M. 2010. “Coupled semi-supervised learning ceur-ws.org/Vol-1695/paper4.pdf for information extraction”. In: Proceedings of the Third 26. Erxleben, F., Günther, M., Krötzsch, M., Mendez, J., and ACM International Conference on Web Search and Data Vrandečic, D. 2014. “Introducing Wikidata to the Linked Mining, pp. 101–110. Data Web”, Proc. ISWC 2014, Part I. LNCS, vol. 8796. Cham: 14. Casanovas, P., Arraiza, J., Melero, F., González-Conejero, J., Springer. Molcho, G. and Cuadros, M. 2014. “Fighting Organized Crime 27. European Union. 2012. Charter of Fundamental Rights of Through Open Source Intelligence: Regulatory Strategies The European Union. (2012/C 326/02) 26.10.2012 Official of the CAPER Project”. In Hoekstra, R. ed., Legal Knowledge Journal of the European Union C 326/391. https://eur-lex. and Information Systems: JURIX 2014. Amsterdam: IOS europa.eu/legal-content/EN/TXT/?uri=CELEX:12012P/TXT Press, pp. 189-198. 28. Danezis, G., Domingo-Ferrer, J., Hansen, M., Hoepman, J.-H, 15. Casanovas, P., Jorge González-Conejero, J., de Koker, L. 2017. Metayer, D.L., Tirtea, R. and Schiffner, S. 2014. Privacy and "Legal compliance by design (LCbD) and through design Data Protection by Design – From policy to engineering. (LCtD): preliminary survey." TERECOM@JURIX 2017, Technical report, ENISA, December. ISBN 978-92-9204- 16. Casanovas, P., Palmirani, M., Peroni, S., Van Engers, T. and 108-3, DOI 10.2824/38623. https://www.enisa.europa. Vitali, F. 2016. “Semantic web for the legal domain: the eu/activities/identity-and-trust/library/deliverables/ next step”. Semantic Web, 7(3): 213-227. privacy-and-data-protection-by-design 17. Castelfranchi, C., Falcone, R. Social Trust: A Cognitive 29. Gandon, F. 2018. “A Survey of the First 20 Years of Research Approach. In: Castelfranchi, C., Yao-Hua Tan (Eds.) Trust on Semantic Web and Linked Data”, Revue des Sciences et and Deception in Virtual Societies. Dartmouth: Springer- Technologies de l’Information - Série ISI : Ingénierie des Verlag (2001). Systèmes d’Information, Lavoisier, https://hal.inria.fr/ hal-01935898/document 39 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 30. Gerck, E. 1998. Toward Real-World Models of Trust: Reli- Scientific Linked Data. CoRR, abs/1809.06532. https:// ance on Received Information. Report MCWG-Jan22-1998, arxiv.org/pdf/1809.06532.pdf http://www.safevote.com/papers/trustdef.htm 43. Langheinrich, M. 2001. “Privacy by design—principles 31. Golden, P. and Shaw, R., 2016. “Nanopublication beyond of privacy-aware ubiquitous systems. In G.D. Abowd, B. the sciences: the PeriodO period gazetteer”. PeerJ Computer Brumitt, and S. Shafer (Eds.), International conference on Science, 2, p.e44. https://peerj.com/articles/cs-44/ Ubiquitous Computing, LNCS 2201, Berlin, Heidelberg: 32. González-Fuster, G. 2014. The emergence of personal data Springer, pp. 273-291. protection as a fundamental right of the EU, LGT Vol. 16. 44. Laufer, C, Schwabe, D. 2017.” On Modeling Political Sys- Cham: Springer Science & Business. tems to Support the Trust Process, Proceedings of the 33. Grandison, T., and Sloman, M. 2000. “A survey of trust in Privacy Online”, PrivON 2017, co-located with the 16th internet applications”, IEEE Communications Surveys & International Semantic Web Conference (ISWC 2017) Tutorials 3 (4): 2-16. Vienna, Austria, Out. 2017, p.1 – 16 http://ceur-ws.org/ 34. Groth, P., Gibson, A., and Velterop. J. 2010. "The anatomy Vol-1951/PrivOn2017_paper_7.pdf of a nanopublication." Information Services & Use 30 (1- 45. Lehmann, J., Isele, R., Jakob, M., Jentzsch, A., Kontokostas, 2): 51-56. D., Mendes, P.N., Hellmann, S., Morsey, M., van Kleef, P., Auer, 35. Hashmi, M., Casanovas, P. and de Koker, L. 2018. “Legal S., and Bizer, C.. 2013. “DBpedia-A large-scale, multilingual Compliance Through Design: Preliminary Results of a knowledge base extracted from Wikipedia”, Semantic Web Literature Survey”. TERECOM2018@JURIX, Technolo- Journal 6(2): 167–195. gies for Regulatory Compliance http://ceur-ws.org/ 46. Lenat, D.B. 1995. “CYC: a large-scale investment in knowl- Vol-2309/06.pdf edge infrastructure”, Communications ACM 38(11): 33–38. 36. Hashmi, M., Governatori, G., Lam, H.P. and Wynn, M.T. 2018. 47. Martín-Chozas, P., Montiel-Ponsoda, E. and Rodríguez- “Are we done with business process compliance: state of Doncel, V. 2019. “Language resources as linked data for the the art and challenges ahead”, Knowledge and Information legal domain” In, G. Peruginelli, S. Faro (Eds), Knowledge Systems, 57(1): 79-133. of the Law in the Big Data Age, Amsterdam: IOS Press, 37. Heimstäd, M., Dobusch, L. 2018. “Politics of Disclosure: pp. 170-180. Organizational Transparency as Multiactor Negotiation”, 48. Marwick, A. andLewis, R.. 2017. “Media Manipulation and Public Administration Review 78 (5): 727-738 Disinformation Online.” Data & Society Research Institute, 38. Hoepman, J.-H. 2018. “Privacy Design Strategies” (The New York. https://datasociety.net/pubs/oh/DataAndSo- Little Blue Book). https://repository.ubn.ru.nl/bitstream/ ciety_MediaManipulationAndDisinformationOnline.pdf handle/2066/195397/195397.pdf?sequence=1 49. McCrae, J.P., Bosque-Gil, J., Gracia, J., Buitelaar, P. and Cimi- 39. Hoepman, J.-H. 2014. "Privacy Design Strategies (extended ano, P. 2017. “The Ontolex-Lemon model: development abstract)", in N. Cuppens-Boulahia et al. (Eds.), ICT Systems and applications”. In Proceedings of eLex 2017 Conference, Security and Privacy Protection - 29th IFIP TC 11 Interna- pp. 19-21. tional Conference, SEC 2014, Marrakech, Morocco, June 50. Meijer, A. 2013. “Understanding the Complex Dynamics 2-4, 2014. Proceedings. Springer 2014 IFIP Advances in of Transparency”, Public Administration Review 73 (3): Information and Communication Technology, pp. 446-459. 429 – 439. 40. Kejriwal, M. 2019. Domain-Specific Knowledge Graph Con- 51. Miller, G. A. 1995. “WordNet: Smith, H.J., Dinev, T., Xu, H struction. Cham: Brief Springer International Publishing A lexical database for English”, Communications ACM 38 41. Koops, B.J. and Leenes, R. 2014. “Privacy regulation can- (11): 39–41. not be hardcoded. A critical comment on the ‘privacy by 52. Norris, C., De Hert, P., L'Hoiry, X., Galetta, A. (Eds.) 2017. design’provision in data-protection law”. International The Unaccountable State of Surveillance. Cham: Springer. Review of Law, Computers & Technology, 28(2): 159-171. 53. Paci, F., Squicciarini, A., Zannone, N. 2018. “Survey on 42. Kuhn, T., Meroño-Peñuela, A., Malic, A., Poelen, J.H., Hurl- Access Control for Community-Centered Collaborative bert, A.H., Centeno, E., Furlong, L.I., Queralt-Rosinach, N., Systems”. ACM Comput. Surv. 51 (6): 1–6. Chichester, C., Banda, J.M., Willighagen, E.L., Ehrhart, F., 54. Pagallo, U., Aurucci, P., Casanovas, P., Chatila, R., Chazerand, Evelo, C.T., Malas, T.B., & Dumontier, M. 2018. Nanopub- P., Dignum, V., Luetge, C., Madelin, R., Schafer, B. and Valcke, lications: A Growing Resource of Provenance-Centric P., 2019. AI4People-On Good AI Governance: 14 Priority Law in Context, Vol 37, Issue 1, 2020 40 ISSN: 1839-4183 Actions, a SMART Model of Governance, and a Regulatory 66. Schudson, M. 2015. The Rise of the Right to Know: Politics Toolbox. https://www.eismd.eu/pdf/AI4PEOPLE%20 and the Culture of Transparency, 1945-1973. Cambridge, On%20Good%20Ai%20Governance%202019.pdf MA: Harvard University Press. 55. Pagallo, U., Casanovas, P. and Madelin, R. 2019. “The 67. Schwab, K., 2017. The fourth industrial revolution. Crown middle-out approach: assessing models of legal governance Business. in data protection, artificial intelligence, and the Web of 68. Schwabe, D., Laufer, C., and Busson, A. 2019. “Building Data”. The Theory and Practice of Legislation, 7 (1): 1-25. Knowledge Graphs About Political Agents in the Age of https://www.tandfonline.com/doi/full/10.1080/20508 Misinformation, http://arxiv.org/abs/1901.11408 840.2019.1664543 69. Sherchan, W. Nepal, S., and Paris, C. 2013. “A Survey of 56. Peczenik, A. 2004. Scientia Juris. In A Treatise of Legal trust in social networks”, ACM Comput. Surv. 45, 4, Article Philosophy and General Jurisprudence, Vol. 4. Dartmouth: 47 , August. Springer. 70. Smith, H.J., Dinev, T., Xu, H. 2011. “Information Privacy 57. Pfitzmann, A. and Hansen, M. 2010. “A terminology for Research: An Interdisciplinary Review”, MIS Quarterly talking about privacy by data minimization: Anonymity, 35: 989–1015. unlinkability, undetectability, unobservability, pseudonym- 71. Speer, R. and Havasi, C. 2012. "Representing General Re- ity, and identity management”. V.034. http://www.maroki. lational Knowledge” In ConceptNet 5, Proceedings of the de/pub/dphistory/2010_Anon_Terminology_v0.34.pdf Eight International Conference on Language Resources and 58. Pinyol, I.; Sabater-Mir, J.; Computational trust and reputation Evaluation (LREC'12), ELRA, pp. 3679-3686. models for open multi-agent systems: a review. Artificial 72. Such, J.M. and Criado, N. 2018. “Multiparty Privacy in Social Intelligence Review 40:1–25 DOI 10.1007/s10462-011- Media”, Communications ACM. 61: 74–81. 9277-z (2013) 73. Suchanek, F.M., Kasneci, G., and Weikum, G.. 2007. “YAGO: 59. Poblet, M., Casanovas, P. and Rodríguez-Doncel, V. 2019. a core of semantic knowledge unifying WordNet and Linked Democracy: Foundations, tools, and applications. Wikipedia”. In 16th International Conference on World Springer Law Brief n. 750. Cham: Springer Nature. Wide Web, pp. 697–706. 60. Rastogi, N., Gloria, M.J.K. and Hendler, J. 2017. “Security 74. Tavani, H.T. and Moor, J.H. 2001. “Privacy Protection, Con- and Privacy of performing Data Analytics in the cloud-A trol of Information, and Privacy-enhancing Technologies”, three-way handshake of Technology, Policy, and Manage- SIGCAS Comput. Soc. 31: 6–11. ment”. arXiv preprint arXiv:1701.06828. 75. Tavani, H.T. 2007. “Philosophical Theories of Privacy: 61. Rehm, G., Galanis, D., Labropoulou, P., Piperidis, S., Welß, Implications for an Adequate Online Privacy Policy”, M., Usbeck, R., Köhler, J., Deligiannis, M., Gkirtzou, K., Fis- Metaphilosophy 38: 1–22. cher, J. and Chiarcos, C., 2020. “Towards an Interoperable 76. Torra, V. 2017. Data Privacy: Foundations, New Develop- Ecosystem of AI and LT Platforms: A Roadmap for the ments and the Big Data Challenge. Cham: Springer Nature. Implementation of Different Levels of Interoperability”. arXiv preprint arXiv:2004.08355. 77. United Nations, Universal Bill of Human Rights, Resolution A/RES/217(III) http://unbisnet.un.org:8080/ipac20/ 62. Rifkin, J. (2011). The third industrial revolution: How ipac.jsp?session=14O243550E15G.60956&profile=voti lateral power is transforming energy, the economy, and ng&uri=full=3100023~!909326~!67 the world. New York: Palgrave Macmillan. 78. US Privacy Protection Study Commission. 1977. Personal 63. Rodriguez-Doncel, V., Santos, C., Casanovas, P. and Gomez- Privacy in an Information Society. https://epic.org/privacy/ Perez, A. (2016). “Legal aspects of linked data–The Euro- ppsc1977report/ pean framework”. Computer Law & Security Review, 32 (6): 799-813. 79. US Secretary's Advisory Committee on Automated Personal Data Systems. 1973. Records, Computers and 64. Rodríguez-Doncel, V., Santos, C., Casanovas, P. and Gómez- the Rights of Citizens. https://aspe.hhs.gov/report/ Pérez, A. (2015) “A Linked Term Bank of Copyright-Related records-computers-and-rights-citizens Terms”. In JURIX-2015, Amsterdam: IOS Press, pp. 91-100. 80. Vrandečić, D., Krötzsch, M. 2014. “Wikidata: a free col- 65. Sartor, G. 2005. Legal reasoning. In A Treatise of Legal laborative knowledge base”, Communications ACM 57 Philosophy and General Jurisprudence. Vol. 5. Dartmouth: (10): 78–85. Springer. 41 Law in Context, Vol 37, Issue 1, 2020 ISSN: 1839-4183 81. Westin, A. 1967. Privacy and Freedom. New York: Atheneum. 83. Zuboff, S., 2019. The Age of Surveillance Capitalism: The 82. Zou, L., and Özsu, M.T. 2017. "Graph-based RDF data Fight for a Human Future at the New Frontier of Power. management", Data Science and Engineering 2 (1): 56-70. Barack Obama's Books of 2019. NY: Profile Books. Law in Context, Vol 37, Issue 1, 2020 42