[Paper Review] OPPO: An Ontology for Describing Fine-Grained Data Practices in Privacy Policies of Online Social Networks

Sanonda Datta Gupta, Torsten Hahmann|arXiv (Cornell University)|Sep 27, 2023

Privacy, Security, and Data ProtectionSocial Sciences3 citations

TL;DR

OPPO is an OWL2-based ontology that formalizes fine-grained data practices from online social network (OSN) privacy policies using BFO, IAO, OBI, and OMRSE as foundational ontologies. It enables automated reasoning and transparency assessment by modeling detailed practices such as data retention duration, storage locations, and security mechanisms, validated via SPARQL queries on a Telegram policy dataset with 15 out of 27 competency questions successfully expressed and answered.

ABSTRACT

Privacy policies outline the data practices of Online Social Networks (OSN) to comply with privacy regulations such as the EU-GDPR and CCPA. Several ontologies for modeling privacy regulations, policies, and compliance have emerged in recent years. However, they are limited in various ways: (1) they specifically model what is required of privacy policies according to one specific privacy regulation such as GDPR; (2) they provide taxonomies of concepts but are not sufficiently axiomatized to afford automated reasoning with them; and (3) they do not model data practices of privacy policies in sufficient detail to allow assessing the transparency of policies. This paper presents an OWL Ontology for Privacy Policies of OSNs, OPPO, that aims to fill these gaps by formalizing detailed data practices from OSNS' privacy policies. OPPO is grounded in BFO, IAO, OMRSE, and OBI, and its design is guided by the use case of representing and reasoning over the content of OSNs' privacy policies and evaluating policies' transparency in greater detail.

Motivation & Objective

• To address the lack of detailed, axiomatized modeling of data practices in existing privacy policy ontologies.
• To improve transparency assessment of OSN privacy policies by formalizing granular practices such as retention duration, storage location, and security mechanisms.
• To enable automated reasoning over privacy policies using formal ontological representations grounded in established upper ontologies.
• To support compliance evaluation and cross-policy comparison by structuring data practices with precise semantic constraints.
• To provide a reusable, extensible ontology for researchers, regulators, and privacy advocates to query and analyze policy content systematically.

Proposed method

• OPPO is developed using the METHONTOLOGY framework and guided by 45 competency questions derived from GDPR, CCPA, and real OSN privacy policies.
• The ontology is formalized in OWL2 and built upon BFO, IAO, OBI, and OMRSE, with selective reuse of DPV and OWL-Time for data types and temporal constraints.
• It introduces 60 new classes and 17 object properties, including specialized subclasses for security mechanisms (e.g., EncryptionMechanism, AuthenticationMechanism) and data practices (e.g., retention duration, storage location).
• Logical consistency is verified using the HermiT OWL2 reasoner, and expressivity is validated by encoding 15 out of 27 competency questions as SPARQL queries over a Telegram policy dataset.
• A minimal import version (OPPO_Ontology_Minimal_Import.ttl) isolates OPPO-specific concepts for clearer modeling in ontology editors.
• The ontology is hosted on GitHub and evaluated in GraphDB using the OWL-RL reasoning profile to validate query execution and result accuracy.

Figure 1: OPPO’s high-level concepts and how they extend concepts from BFO, IAO, OBI, and OMRSE. Concepts in yellow are introduced by OPPO, while all other concepts are being reused.

Experimental results

Research questions

• RQ1 How can fine-grained data practices—such as retention duration, storage location, and security mechanisms—be formally modeled in a way that supports automated reasoning?
• RQ2 To what extent can an ontology capture the transparency of privacy policies by representing detailed data practices with semantic precision?
• RQ3 Can the ontology express and answer competency questions about data practices using SPARQL queries over real-world policy data?
• RQ4 How does the integration of established upper ontologies (BFO, IAO, OBI, OMRSE) support the formalization and reusability of privacy policy concepts?
• RQ5 What is the feasibility of using OPPO to compare transparency levels across multiple OSN privacy policies?

Key findings

• The OPPO ontology successfully models 60 new classes and 17 object properties, with 271 logical axioms, forming a consistent and logically sound formalization in OWL2.
• The ontology is logically consistent, with no inconsistencies or unintended class equivalences detected using the HermiT reasoner.
• Fifteen out of twenty-seven competency questions were successfully encoded as SPARQL queries and executed over a Telegram policy dataset in GraphDB, with results matching expected outcomes.
• The ontology correctly identified that Telegram stores four distinct data types for a maximum of 12 months, demonstrating its ability to capture retention practices with precision.
• The minimal import version effectively isolates OPPO-specific content, enhancing clarity and usability in ontology editors like Protégé.
• The validation process confirms that OPPO is expressive enough to represent and reason over detailed data practices, supporting transparency and compliance analysis.

Figure 2: The core concepts in OPPO and the relations between them.