I intended to post this writeup to coincide with the official publication of the 10th anniversary edition of the Applied Ontology journal due to go in print this month, but I have less patience than I thought. The reason for this is that my PhD student, Zubeida Khan, and I got a paper accepted for the anniversary edition, which is online in the issue preprint as An empirically-based framework for ontology modularity . It was one of those I’m-not-sure-but-lets-be-bold-and-submit-anyway papers, with Zubeida as main author. It is her first ISI-index journal article, and congrats to that! (The article bean counting is [somewhat/very] important in academia in South Africa). UPDATE (22-12): from the editorial by Guarino & Musen: it was one of the 2 papers accepted out of the 7 submitted, and IOS Press has awarded a prize for it.
So, what is the paper about? As the blog post’s title suggest: ontology modules. The first part is a highly structured and comprehensive literature review to figure out what all the parameters are for ontology modularisation, which properties modules have, and so on. The second part takes a turn to an experimental approach, where almost 200 ontology modules are classified according to those parameters. Both seek to answer questions like: “What are the use-cases, techniques, types, and annotation features that exist for modules? How do module types differ with respect to certain use-cases? Which techniques can we use to create modules of a certain type? Which techniques result in modules with certain annotation features?” Answers to that can be found in Section 6, with the very short version to the first question (explanations in the paper):
- use-cases: maintenance, reasoning, validation, processing, comprehension, collaborative efforts, and reuse.
- techniques: graph partitioning, modularity maximisation, hierarchical clustering, locality-based modularity, query-based modularity, semantic-based abstraction, a priori modularity, and manual modularity.
- types: ODPs, subject domain-based, isolation branch, locality, privacy, domain coverage, ontology matching, optimal reasoning, axiom abstraction, entity type, high-level abstraction, weighted, expressiveness sub-language, and expressiveness feature modules.
- annotation features: seed signature, information removal, abstraction (breadth and depth), refinement, stand-alone, source ontology, proper subset, imports, overlapping, mutual exclusion, union equivalence, partitioning, inter-module interaction, and pre-assigned number of modules.
This, then, feeds into the third part of the paper that puts the two previous ones together into a framework for ontology modularity, which links the various dimensions (groups of parameters) based on the dependencies that surfaced from the literature review and analysis of modules, and proposes how to proceed in some modularization scenario. That is, it answers the other three main questions. They are easier to show in a figure, imo, like the following one on dependencies between module type and the techniques to create them:With it, you’d be able to answer a case like: “Given that we wish to create an ontology module with a certain purpose or use-case in mind, which modularity type of module could this result in?” and so on to which technique to use, and which properties such a module will have.
Space limitations caused the paper to have only one illustrative example with the Symptom Ontology, but the general idea hopefully will be clear nevertheless. At the moment, before official print, the article is still behind a paywall if you don’t have either institutional or IAOA access, but feel free to contact either Zubeida or me for a copy.
 Khan, Z.C., Keet, C.M. An empirically-based framework for ontology modularity. Applied Ontology, 2015, in print (25p). DOI: 10.3233/AO-150151.