There are different methods, both automated and manual, to develop an ontology. This work aims to succinctly explain one way to manually develop an ontology. Disclaimer: this is based on my experiences and thereby limited. The reader is encouraged to look at the diverse global content on this and related topics. This work originally began in approximately 2010 in an unpublished written document I call "The Ontology Guidebook".
If you find value in the content of this work, please cite it in your reference citation style of choice or as: "Rovetto, Robert J., "Ontology Guidebook", https://purl.org/rrovetto/ontology-guidebook"
- In-progress. Subject to revision. No claims to completeness
- As an unfunded personal project to date, formal support is needed to sustain development: financial support, employment, work collaborations, studyship options, etc.
- Link to this page: https://purl.org/rrovetto/ontology-guidebook
- Given the similarity between computational ontologies and other artifacts that go by names such as 'controlled vocabulary', 'taxonomy', 'classifcation system', 'semantic data model', 'conceptual model', etc., many of the facts, opinions, and recommendations made in this repository apply to them as well, i.e., to other so-called knowledge organization systmes.
- If you find value in my work, donations are welcome via PayPal or via this GoGetFunding webpage.
- This project is intended to result in a product/output as a book or manual. Sponsors and publishers are desired; please contact.
- Employment opportunities & work collaborations are desired. Click here to schedule a meeting.
- Hire me, Purchase products, or schedule a meeting at this link
- Presentation list of Ontology Education (PPT file)
- Presentation list of Ontology Services (PPT file)
A) Botton-Up development from data (example 1):
Click to expand for further details
- Examine data, identify relevant subject matter; then create categories to describe that data (e.g., for what the data is about). This is specific to generic, or concrete to abstract. It may focus on asserting only those categories that will annotate the data elements you have.
B) Botton-Up development from subject matter (example 2):
Click to expand for further details
- Examine the subjectmatter (e.g., corpus documents, nomenclature, etc.) and create corresponding ontology constructs for the most common concepts found in the subjectmatter.
C) Top-Down development (example):
Click to expand for further details
- Identify and create ontology constructs independent of examining data, but which can annotate the data. E.g., create abstract classes and modify accordingly once you examine the data and the intended meaning of the datasets and data elements.
D) Hybrid development: a combination of both bottom-up & top-down.
A manual computational ontology development methodology includes the following activities: identifying and scoping your target subject-matter, set of concepts, and/or data; subject-matter/domain research; listing or creating a defined set of terms; listing statements and pieces of knowledge (e.g., general facts of the domain) to be captured; a taxonomy and classification system using the specified terminology; computational formalization of the terminological system; computational formalization of the knowledge statements; identifying applications. The development process is iterative and non-linear: some phases (or steps) of the development process may be performed concurrently with others and revisited.
Click to expand for further details
- Identify purpose(s), goal(s), or function(s), application(s) for the ontology. What is the ontology(s) for? How is it intended to be used? What are the desired features and functionalities?
- Identify the subject-matter or set of concepts (the universe/domain of discourse/interest) you wish to model in an ontology.
- Identify the data that the ontology will be applied to (e.g., to annotate), annotate, or be based on. What datasets, databases? What types of data? What is the data about?
Meta-level Consideration: the boundaries or limits of a given target area, and domain are often arbitrary. Some domains, perceived as disciplines, are in fact overlapping. Therefore there is a degree of arbitrariness when scoping and identifying or creating a target domain. One modeler may scope and model the same topic differently than anothe modeler.
Click to expand for further details
- Research the target subject and domain corpora: publications, data, projects, subject-matter experts, educational courses, etc.
Click to expand for further details
- Identify use-cases for the ontology. Where may the ontology be applied? How can the ontology be used? What stakeholders, societal sectors, computational systems, and applications may benefit from it? What are situations in which the ontology could be used?
Click to expand for further details
- Identify or specify the boundaries of the target subject-matter that the ontology(s) will represent.
- NOTE: consider this.
Click to expand for further details
- Identify what will be required to (i) realize the purpose and (ii) apply the ontology to the use-cases.
- State specification details:
- What natural language?
- What computable language (knowledge representation and reasoning language / implementation langauges)?
- Stylistic conventions, e.g., naming conventions for terms/labels, etc.
- Identify or create competency questions: queries that the ontology (and datasets) can be asked computationally.
- Select, Acquire or Develop resources and tools, e.g., ontology editor software, knowledge representation languages,
- Corpora (research documents, domain knowledge), data sources, contributors, partners, subject-matter experts, etc.
- Identify and select supplemental tools: ontology engeineer tools, ontology development environments, automated reasoners, etc.
NOTE: an example is the NEON Methodology.
Click to expand for further details
- Examine the target subject matter. Gain a big-picture understanding. Try to identify essential concepts, themes, challenges, etc.
- Identify generic knowledge, statements or beliefs to be represented in the ontology. These are often universal statements of a broad nature.
Click to expand for further details
- Identify & list specific things in the universe of discourse you want the ontology to represent, e.g., particulars objects, activities, etc.; categories of objects, activities, etc.
- Identify & list concepts and terms needed to either: (i) describe the things in universe of discourse, or (ii) described the concepts and terms themselves.
- Create concepts and terms where needed, e.g., where finer distinctions non-found in the domain literature are needed to ontologically model the domain.
- List statements of knowledge or belief to be represented in the ontology. This may take the form of a set of sentences describing some general rule, fact, principle, or aspect about the target domain and its contents.
- WHEN CREATING AN ONTOLOGY FOR A SPECIFIC DATABASE: list the key terms and pieces of knowledge found therein, e.g., database column/field and row names, etc. Analysis and discussion may be needed.
- For a candidate category, ask: Is there a data element or dataset that can be annotated with this category? If not, then you may not need it (but it will be situational).
Click to expand for further details
- Identify or prescribe the intended meaning for each of the list of things, concepts and terms: determine the intended semantics for the ontology's terminology.
- You may begin with some intended meaning or concept, and the form a term that most clearly expresses it.
- Research definitions (of key terms) from corpus material: dictionaries, academia, journal publications, subject-matter experts, etc.
- Identify undefined (primitive) terms.
- Write definitions in natural-language that most closely expresses that intended meaning.
- For all primitive terms, state they are undefined, but also provide a clarifying note to provide the user with some sense of meaning.
Click to expand for further details
- In this process, identify how the things or terms denoting them are related (if at all). Distinguish between relationships between the word (term) and what it denotes. E.g., dogs have four legs. vs. the word 'dog' means [...].
- Create or select terms for how they are related (i.e., for the relationships between them).
- Define these relational terms.
- Use abstract distinctions and structuring relations, e.g., category vs. instance, class vs. member, class and subclass, broader & narrower, etc.
Click to expand for further details
- Select or create a formalism/implementation language that is computable. Select or create an ontology editor tool with a built-in artificial language.
- Consider also translating the natural-language definitions and knowledge statements into symbolic logic definitions, e.g., in first-order predicate calculus (FOL).
- Translate the natural language meanings into the chosen formalism, or use your chosen ontology editor tool’s language.
- Formal definitions may use the labels to better foster automation.
Click to expand for further details
- Before deployment into actual use-cases or operational data owned by stakeholders or others, collect or create test data to test and validate the functionality, goals, and utility of the ontology. This will serve to identify any bugs (technical problems or limitations).
- Apply or use the ontology to that test data by using supplemental tools (See Stage 5): run any reasoner to test for functionalities such as classification, correct inference, etc.
Robert J. Rovetto rrovetto(at)terpalum.umd.edu
- International Association for Ontology and its Applications (IAOA), Education Technical Committee member;
- IAOA Semantic Web & Applied Ontology Special Interest Group
- Ontolog Profile
Attribution is required. Cite as: Rovetto, Robert J., "Ontology Guidebook", https://purl.org/rrovetto/ontology-guidebook
© 2023, Robert John Rovetto. All right reserved. Not authorized for commercial use unless explicitly negotiated with the author. Citation/attribution required. No warranty. Presented "AS IS". Author and copyright holder is not liable. All content, work and products are subject to revision. No claims to completeness or complete accuracy.