Presentation Description
This talk is about application profiles as a framework for structuring information applications and why this is important in the context of knowledge organization systems and artificial intelligence.
Date: August 19, 2026
Venue: International Society for Knowledge Organization – ISKO 2026
Location: São Paulo, Brazil
Broad Topic: Automation
Presenter: Joseph Busch, Principal Analyst, Taxonomy Strategies
Watch this Presentation:
Download Presentation: MP4 | PDF
Download Paper: PDF
Download Slide Script: PDF
[Slide 1]
Today I will discuss application profiles as a framework for structuring information applications. The key argument is that application profiles are not usually treated as knowledge organization systems in themselves, but they contain two kinds of vocabularies that are central to knowledge organization: schema vocabularies and value vocabularies. I will also connect this argument to current questions about generative AI reliability and transparency.
[Slide 2] Agenda
The talk moves from concepts to infrastructure and then to contemporary AI applications. I will begin by defining application profiles, then distinguish schema vocabularies from value vocabularies. I will then discuss how profiles are constructed, why documentation may be as important as interoperability, and finally how the same framework can improve AI systems.
[Slide 3] Central Claim
The central claim is deliberately nuanced. Application profiles are containers or frameworks. They describe how to use existing standards for a particular application. But inside those profiles we find two types of vocabularies: the schemas that define elements and the controlled vocabularies that standardize values. Those vocabularies are the knowledge organization components.
[Slide 4] Research Questions and Objectives
These are the research questions and objectives of the paper. Firstly, What is the role of knowledge organization systems in application profiles? Secondly, in a resource description format (RDF) environment, do we need application profiles at all? Thirdly, What is the role of KOS in application profiles in the AI environment?
[Slide 5] Application Profiles: Definitions
Three definitions frame the discussion. Heery and Patel emphasize mixing and matching metadata schemas. SEMIC uses a data-centric semantic web vocabulary. W3C emphasizes use of base specifications for a community goal. Across the definitions, the common theme is structured reuse rather than creating an entirely new standard.
[Slide 6] Namespaces and Reuse
Namespaces are more than technical labels. They identify authority, supply persistent identifiers, and define the data elements being reused. In application profiles, namespaces let communities assemble a profile from established vocabularies such as Dublin Core or FOAF (Friend of a Friend) rather than writing a local schema from scratch.
[Slide 7] Schema vs. Value Vocabularies
This distinction is the core concept behind the paper. Schema vocabularies structure description: what fields exist, what they mean, and how they may be used. Value vocabularies structure meaning: what terms should be used, what synonyms or broader terms exist, and how concepts are related. Both are necessary for reliable information applications.
[Slide 8] Metadata Schemas
Metadata schemas are prescriptive. They tell us what kinds of descriptive data should be captured and how those elements are structured. Dublin Core is the recurring example because it provides a widely used element set for resource description and has been central to application profile work.
[Slide 9] Controlled Vocabularies
Controlled vocabularies address the language of description. They constrain variation so that systems and users do not scatter meaning across inconsistent labels. This is especially important in retrieval, where synonyms, hierarchy, and related terms can have measurable consequences for recall and precision.
[Slide 10] Kinds of Value Vocabularies
The paper distinguishes several forms of controlled vocabulary. Taxonomies organize hierarchical structure. Thesauri add synonym control and associative relations. Authority files stabilize names. While ontologies provide richer formal semantics. These are not interchangeable, but each can play a role in an application profile.
[Slide 11] Constructing Application Profiles
The Dublin Core application profile tradition makes profile construction concrete. A profile is not only a list of metadata fields. It includes requirements, a model of the domain, a description set, and guidance for usage and syntax. This is where the profile becomes a documentation practice.
[Slide 12] Figure 1. Singapore Framework for Dublin Core Application Profiles
This figure visualizes how an application profile connects functional requirements, domain models, description sets, syntax guidelines, and usage guidance. The important point for this talk is that schemas and value vocabularies are embedded in a broader design and documentation framework.
[Slide 13] Why Application Profiles?
The original promise of application profiles is interoperability. If different systems follow the same profile, they can exchange information more easily. But the paper also emphasizes communities of practice: profiles express community-specific requirements and practices, not merely technical constraints.
[Slide 14] Examples of Application Profile Contexts
The paper situates application profile work across several communities. LRMI addresses learning resources. DCAP and DCTAP address Dublin Core application design. NKOS AP supports descriptions of knowledge organization systems. Machine-actionable data management plans (DMPs) illustrate profile use in research data management.
[Slide 15] Interoperability—and Its Limits
Nevile and Mendez raise an important challenge: in an RDF environment, do we need application profiles at all? The paper’s answer is pragmatic. Even if broad semantic interoperability can occur without a profile, profiles are still valuable because they document requirements and decisions for communities and implementers.
[Slide 16] Documentation as Scholarly Infrastructure
This is one of the paper’s strongest contributions. Application profiles are not only interoperability devices; they are scholarly and technical documentation infrastructure. They make assumptions explicit. They record why particular schema elements or vocabularies are used. That documentation is essential for transparency and reuse.
[Slide 17] Why This Matters for Generative AI
The paper then brings the argument into the present AI context. Generative AI treats many kinds of content as processable data. But its outputs are variable and often opaque. Schema and value vocabularies can supply structure, terminology, and context that make AI applications more reliable and explainable.
[Slide 18] Figure 2. Integrating Vocabularies with Generative AI
This figure shows a simple vocabulary-enhanced approach to prompting. Thesauri can add synonyms and quasi-synonyms. Taxonomies can broaden or narrow concepts. Metadata schemas can provide context such as content type, geography, language, time, or process. The broader point is that vocabulary work can become part of the AI pipeline.
[Slide 19] RAG and Knowledge Graphs
Retrieval-augmented generation is presented as a way to make prompts more responsive to domain-specific questions. Knowledge graphs can support retrieval with equivalent terms and typed related concepts. Vocabularies can be used before generation for retrieval and indexing, or after generation for annotation and verification.
[Slide 20] From Vector Search to Vocabulary-Aware AI
The paper cautions that RAG systems may rely on vector similarity without vocabulary-based indexing. That can be useful, but it is not the same as semantic control. A more reliable approach maps content to vocabularies and then uses those vocabularies to improve retrieval quality and explainability.
[Slide 21] Implications for Knowledge Organization
For knowledge organization, the implication is that vocabulary work is not legacy infrastructure. It is directly relevant to AI, data exchange, and accountable systems. Application profiles can connect KOS theory with practical engineering and can serve as community knowledge artifacts.
[Slide 22] Conclusion
To conclude, application profiles are best understood as structured documentation frameworks that integrate schema and value vocabularies. Their value lies not only in interoperability but also in making requirements and design choices explicit. In the AI context, this framework can help move from opaque statistical generation toward more transparent, vocabulary-aware information applications.
[Slide 23] Selected References
This reference slide provides an abbreviated bibliography for the presentation. I invite discussion and point you back to the standards and scholarly sources that anchor the argument.