Dark Pattern Ontologies

• Dark Pattern Ontologies are formal frameworks that classify and analyze manipulative UI practices using structured hierarchies and semantic rules.
• They integrate normative, empirical, and meta-ontological approaches to support automated detection, regulatory mapping, and ethical design.
• Applications span legal assessments, empirical audits, and design guidance, while challenges include context sensitivity and dynamic evolution of patterns.

Dark pattern ontologies provide formal frameworks for representing, organizing, and reasoning about manipulative user interface (UI) design practices known as dark patterns. These ontologies are crucial for regulatory analysis, automated detection, cross-domain harmonization, empirical studies, and the formalization of ethical design guidance. Ontologies in this context describe types, properties, interrelationships, and operational semantics of dark patterns in a machine-readable and normatively precise form, supporting both theoretical inquiry and real-world application.

1. Foundations and Principal Ontological Approaches

Dark pattern ontologies have evolved to address the classification inconsistencies, domain bias, and limited interoperability observed in purely taxonomic or natural-language dark pattern research (Li et al., 2024, Caragay et al., 2023, Lewis et al., 2024). The primary ontological formalisms represented in the literature are OWL/RDF class hierarchies (enabling formal reasoning, strict validation, and interlinking), description logic, and directed graphs with network community structure.

Key approaches:

• Normative, autonomy-based ontologies: Structure dark patterns according to violations of decisional autonomy, with categories parameterized by regulatory criteria and conditions for materiality (e.g., severity threshold, bias-exploitation) (Brenncke, 2023).
• Empirical and interactional ontologies: Model systematically observed pattern categories, with a focus on operational features, influence mechanisms, and cognitive biases, frequently supporting automated detection or field-scale auditing (Mathur et al., 2019, Li et al., 2024).
• Concept catalog and expectation-based ontologies: Frame patterns as deviations from standard application concepts, linking normative harm to the breach of user expectations and benefit asymmetries (Caragay et al., 2023).
• Meta-ontologies via taxonomy graph integration: Construct harmonized representations by merging distinct taxonomies using network analysis, community detection, and formal equivalence/synonymy relationships (Lewis et al., 2024).

2. Core Class Structures, Properties, and Semantic Relations

Ontologies across major works define multi-level class hierarchies, object/data properties, and key annotation properties. Abstracting across leading frameworks (Brenncke, 2023, Caragay et al., 2023, Mathur et al., 2019, Li et al., 2024, Lewis et al., 2024):

Common Core Classes

Class	Role/Definition
DarkPattern	The root concept; subclassed into types, categories, or autonomy violations
PatternCategory	High-level grouping semantics (e.g., Sneaking, ConsentManipulation, Friction, SocialInfluence)
DarkPatternType	Operational patterns, typically leaf nodes with specific behavioral definitions
SubCategory	Intermediate categories (in fine-grained frameworks such as DPAF)
ConsumerBias/CognitiveBias	Mechanistic drivers (e.g., DefaultEffect, AnchoringBias, SunkCostFallacy)
ChoiceArchitectureFeature	UI/control element as instantiated in digital environments
Concept, StandardConcept, DarkConcept	Abstract functional units, standard abstractions, or violative deviations (Caragay et al., 2023)

Common Semantic/Object Properties

Property	Domain → Range	Role/Constraint
hasSubPattern	AutonomyViolationCategory → DarkPattern	Top-down structuring of patterns
isA, belongsToTaxonomy	DarkPattern → Taxonomy/Taxon	Provenance, cross-taxonomy mapping
exploitsBias	DarkPattern → ConsumerBias	Links cognitive mechanism to pattern
violatesAbilityToMakeModelSelfDeterminedDecision	ChoiceArchitectureFeature → ModelSelfDeterminedDecision	Captures normative impairment of user autonomy
partOf, implementsPattern	DarkPattern → DarkPattern	Shows composition, implementation, or employment relationships
inlfuenceMechanism	DarkPatternType → {Deception, Steering, Coercion}	Rhetorical and cognitive mechanism
hasCategory, subCategoryOf	DarkPatternType → DarkPatternCategory, PatternCategory	Taxonomic placement
severity, hasSeverity	DarkPattern → decimal/annotation	Quantifies materiality or harm
annotation properties	Various	Definitions, examples, harm scenarios, scenario tags

Formal Constraints

• Disjointness: Category, subcategory, and type classes are pairwise disjoint for exclusive classification (Mathur et al., 2019, Li et al., 2024).
• Cardinality: Each pattern instance often linked to exactly one category (Mathur et al., 2019).
• Materiality thresholds: Applied via severity properties (e.g., only patterns with severity ≥ 0.5 constitute autonomy violations (Brenncke, 2023)).
• Bias-exploitation: Only patterns linked to a cognitive bias are considered autonomy-violating (Brenncke, 2023).
• Benefit asymmetry: For deviation-based formalisms, $$Benefit_{provider}(\Delta) > Benefit_{user}(\Delta)$$ required for a pattern to be classified as dark (Caragay et al., 2023).

3. Representative Ontology Frameworks and Taxonomies

The following summarize core frameworks and structure:

Normative Autonomy Violation Ontology (Brenncke, 2023)

Defines six top-level subclasses of AutonomyViolationCategory (e.g., UnderminingOfMandatedInformation, Deception, InducingContractualAgreementWithoutReflection, NegativeFriction, NonNeutralPresentationOfChoiceOptions, Manipulation), each linked to specific patterns such as DripPricing, NaggingPattern, SalientAgreeButton. Formal constraints distinguish between mere influence and true autonomy violation via severity and bias-exploitation conditions.

Empirical Taxonomy Blueprint (Mathur et al., 2019)

Provides seven categories (Sneaking, Urgency, Misdirection, SocialProof, Scarcity, Obstruction, ForcedAction), each with exact type classes and annotation properties (e.g., hasAsymmetric, hasDeceptive). Influence mechanisms and cognitive biases are explicit object properties; a full OWL/Turtle serialization is provided.

Concept Catalog Ontology (Caragay et al., 2023)

Defines dark concepts via deviation $\Delta$ from a Standard Concept (SC), formalized as missing, renamed, or misrepresented state/actions. The classification condition is $$Dark(C′, SC) \iff (\Delta \neq \emptyset) \wedge Benefit_{provider}(\Delta) > Benefit_{user}(\Delta)$$. Subclass and synchronization relations permit fine-grained mapping of domain-agnostic and domain-specific functional concepts.

Directed Graph Meta-ontology and GHS (Lewis et al., 2024)

Models patterns and taxonomies as a directed multigraph, with PatternCategory and Pattern nodes, linked via belongsToTaxonomy, implementsPattern, subCategoryOf, and isSynonymOf. Community detection (Louvain algorithm) reveals ten stable high-level communities, harmonizing overlapping taxonomies and enabling glyph-based visual representation for regulatory and communication purposes.

DPAF 68-Pattern Ontology (Li et al., 2024)

Implements an OWL hierarchy: six top-level categories, 32 subcategories, and 68 DarkPattern classes. Each class is annotated via rdfs:comment (definition), dp:hasSeverity (harm tag), dp:appearsInScenario (usage context). Example instance data link to scenario screenshots, providing concrete referents for ontology classes.

4. Interoperability, Harmonization, and Analytical Methods

Ontologies have increasingly supported taxonomy harmonization, enabling comparison and integration of disparate classification systems (Lewis et al., 2024). Major methods include:

• Synonym detection and pattern merging: Patterns from independent taxonomies are treated as equivalent iff definitions, names, or examples match, with incoming/outgoing edges merged in the ontology graph.
• Network community analysis: Directed ontology graphs support algorithmic detection of clusters ("meta-categories"), with modularity $Q$ quantifying structural coherence:

$$Q = \frac{1}{2m}\sum_{i,j}\Bigl[A_{ij} - \frac{k_i\,k_j}{2m}\Bigr]\delta(c_i,c_j),$$

where $A_{ij}$ is the adjacency matrix, $c_i$ denotes cluster assignment (Lewis et al., 2024).

• Automated consistency checking: Description-logic constraints enforce disjointness, cardinality, acyclicity among classes and pattern relationships (Mathur et al., 2019).
• Benefit trade-off assessment: Automated or manual audits leverage catalog/ontology queries (e.g., SPARQL) to identify violations of standard concept presence, synchronizations, or benefit asymmetries (Caragay et al., 2023).

5. Applications and Evaluation

Dark pattern ontologies have direct applications in:

• Legal and regulatory assessment: Ontologies support both legislative mapping (e.g., EU frameworks regulating for autonomy (Brenncke, 2023)) and operational regulatory tools (e.g., app-store labelling using glyphs in a Globally Harmonized System (Lewis et al., 2024)).
• Empirical measurement and detection: Ontologies underpin the coverage analysis of detection tools and datasets—e.g., coverage rates of 44–45.5% for existing tools under the DPAF ontology (Li et al., 2024).
• Design guidance and auditing: Concept catalog formalisms enable design-time checks for the presence and mapping of critical state/actions, systematic annotation of deviations, and design pattern documentation (Caragay et al., 2023).
• Research synthesis: Harmonized ontologies serve as reference structures for comparing, merging, and extending the dark pattern classification landscape (Lewis et al., 2024).

Coverage evaluations show that compact core ontologies (e.g., seven categories, fifteen types in Mathur et al.) can account for essentially all field-observed patterns in major empirical crawls, while more detailed ontologies (e.g., 68 patterns in DPAF) reveal limitations in detection tool and dataset comprehensiveness (Li et al., 2024, Mathur et al., 2019).

6. Challenges and Future Directions

Persistent challenges include:

• Granularity and context sensitivity: Extensive cataloging (e.g., DPAF’s 68 patterns) increases coverage but creates classification ambiguity; context shifts and evolving UI metaphors (as in the transition from ShoppingCart to PersonalShopper) require ontologies to support redefinition and user expectation recalibration (Caragay et al., 2023, Li et al., 2024).
• Empirical validation: There is limited empirical evidence linking specific ontological deviations to concrete measures of harm, satisfaction, or regulatory adjudication. Coverage rates by automated classifiers and datasets indicate significant room for improvement (Li et al., 2024).
• Dynamic extension: Ontology structures must support continuous integration of new patterns as malicious design evolves (e.g., in XR, live-stream sales), as proposed in network-driven frameworks (Lewis et al., 2024).

A plausible implication is that robust, modular, and harmonized ontological infrastructure will be required to enable systematic regulatory, technical, and ethical interventions at scale as interface complexity and manipulative practices proliferate.

7. Representative OWL/RDF Fragments and Schematic Table

Below is a schematic summary of representative class and property structures drawn from the literature:

Ontology	Top-level Categories	Example Properties
Normative/Autonomy	6 (AutonomyViolationCategory)	hasSubPattern, exploitsBias, severity
Empirical (Mathur)	7 (Category), 15 (Type)	hasCategory, hasAsymmetric, influenceMechanism
Graph-based (GHS)	10 communities via Louvain	belongsToTaxonomy, implementsPattern, isSynonymOf
Concept Catalog	StandardConcept, DarkConcept	diff, benefit attribution, mapping (M)
DPAF	6 (Category), 32 (Subcategory), 68	dp:hasSeverity, dp:appearsInScenario

Representative OWL/Turtle serialization patterns, data properties, and formal axioms are found in (Brenncke, 2023, Caragay et al., 2023, Mathur et al., 2019, Li et al., 2024).

---

These ontological resources collectively provide a rigorous foundation for research, regulatory, and practical efforts to identify, understand, and mitigate the risks of manipulative UI design. They enable formal reasoning about pattern mechanisms, cross-domain harmonization, and empirical assessment, and they underpin ongoing advances in both detection methodologies and ethical-by-design frameworks.