Open Exploration vs. Goal-Constrained Ontology Building: How to Choose?
The article distinguishes two ontology construction approaches—open exploration for evolving research questions and goal-constrained for defined business targets—explains their differences in problem formation, model evolution, progress evaluation, and stopping criteria, and advises selecting based on current task while separating feedback loops from construction orientation.
Two Ontology Construction Orientations
The article opens with two teams building ontologies: one doing public opinion analysis where topics shift from brand negativity to celebrity gossip, discovering new events and propagation relations as more platforms are added; the other organizing enterprise maintenance records with clear questions about equipment identity, fault classification, timestamps, and personnel. Both are called ontology construction, but their starting points and progress evaluation differ. Requiring the first team to lock final requirements upfront would stall research; letting the second team expand freely would leave business users asking when the promised query capability will be ready.
Definitions
Open exploration ontology construction applies when the overall problem boundary is undetermined or research goals adjust with materials and findings. Problems, concepts, and relations form gradually, letting questions and the model co-evolve. Public opinion and intelligence analysis often face this: initial focus may shift after reviewing materials, revealing which objects to distinguish, which relations to trace, or even whether the original question is worth asking. The ontology both helps answer questions and helps discover the next questions.
Goal-constrained ontology construction centers on relatively clear business goals, scope of use, and completion criteria. Necessary concepts, relations, and constraints are selected and used to check whether the model and application satisfy the agreement. Enterprise maintenance archives and equipment history queries fit this mode: first agree on what to query, then decide what the model must express. The model can still change, but each change must show how it supports the current goal or why the goal itself needs formal adjustment.
The author notes these labels are for discussion convenience, not industry standards. "Relatively clear" goals do not mean all facts are known—maintenance records may be missing or contradictory, equipment status may be unknown—but the team still knows which problem to solve.
Open exploration allows questions to unfold with research; goal-constrained builds around relatively clear agreements; both can revise the model.
Comparison of Approaches
What is known at start: Open exploration has a research direction, materials, or phase questions with an unfolding boundary. Goal-constrained has relatively clear purpose, responsibility scope, and completion criteria.
How problems change: Open exploration: new findings may bring new questions, research pivots, or analysis method adjustments. Goal-constrained: mainly refines agreed questions; expanding scope requires separate confirmation.
Why the model changes: Open exploration: to express new findings, distinguish interpretations, or support new research questions. Goal-constrained: to fill meanings needed for current tasks, correct ambiguities, or adapt to confirmed business changes.
How to assess progress: Open exploration: what evidence supports new understanding, which explanations were excluded, what remains to verify. Goal-constrained: can agreed questions be answered, does output meet requirements, how are open items handled.
When to stop: Open exploration: phase questions have conclusions or clear limits, decide whether to enter next round. Goal-constrained: agreed scope passes review and handover; ongoing operation and new uses get separate responsibility.
Both need materials, expert judgment, and checking; both can be expert-led or AI-assisted. Open exploration is not "let the model grow randomly"; goal-constrained is not "experts define once, never change again." The article references Ontology Development 101 (Noy & McGuinness), which treats ontology development as iterative refinement allowing domain and purpose to change during development, using competency questions to scope. The author's two categories are an organization built on that foundation. Reference: https://protege.stanford.edu/publications/ontology_development/ontology101-noy-mcguinness.html
Beyond a Classification Tree
In public opinion analysis the ontology may look like a tree growing branches: starting with brands, later adding persons, works, events. But one must distinguish what is growing. Adding a celebrity under an existing "Person" type is usually adding instances. Only when current expressions cannot hold needed meanings should new types, relations, or constraints be considered. A new platform may only require field mapping changes.
Moreover, a taxonomy tree captures only part of the structure. Persons participate in events, accounts relay viewpoints, materials support or refute claims—these need cross-branch relations. What truly aids research is not just more branches but knowing what evidence supports those relations and what their limitations are.
Taxonomy is only partial structure; adding instances differs from extending shared definitions; cross-category relations also need material evidence.
Similarly, a maintenance ontology expands. If business later requires distinguishing "repair completed" from "re-qualified after review", the team evaluates whether existing definitions suffice. This does not automatically turn the project into open exploration because the business judgment to support remains clear.
Choosing the Right Orientation for the Current Task
The author asks three questions: Can we now state which judgment to support? Is new material mainly filling answers or potentially changing the questions? What will we present to users at the end of this round?
Organize specified maintenance archives for history queries → Goal-constrained. Leave: query questions, data scope, definition basis, check and handover conditions.
Continuously research brands, persons, events with direction shifting per findings → Open exploration. Leave: current research direction, materials, observations, hypotheses, and this round's questions.
Verify a specific claim in a particular event → Can carve out a goal-constrained task within open research. Leave: claim to verify, evidence scope, judgment criteria, unknowns.
Investigate recurring equipment faults of unknown cause → Can run open exploration within maintenance business. Leave: known phenomena, candidate explanations, evidence needed, next research steps.
Open research can carve out clear tasks; existing business can also run exploration; choose orientation by current work.
Selection cannot rely solely on industry name. A public opinion platform may have fixed statistical tasks; a maintenance project may encounter exploratory problems. Look at the current round of work, not just the system name.
The two orientations can complement each other. Exploration discovering "retweets counted as publisher's own views" can spawn a focused task to verify viewpoint attribution; after solving it, overall research continues. Conversely, after maintenance query completion, if business wants to study fault patterns, that should be a separate research scope with its own investment and judgment methods. Do not quietly turn unfinished queries into "exploration" or lower acceptance criteria to pass original sign-off. If goals truly change, keep original goal, change reason, and confirmation record.
Feedback Loops Are a Separate Concern
Both orientations share basic modeling actions: check reusable definitions, organize concepts and relations, map materials to model, check and modify, possibly over several rounds. The difference is goal-constrained works around agreed questions; open exploration must also judge which questions are worth pursuing while researching. How to start, where to push, when to pause and consolidate results naturally differ. Detailed processes for each will follow in subsequent articles.
Both orientations share basic modeling actions, but problem formation differs, so work arrangement and phase-end conditions differ.
Feedback Loops Are a Separate Concern
The previous question was: are problems already clear or must we research to find them? Feedback loops ask: when review or use reveals issues, is there someone to verify, handle, and re-check effectiveness?
The article borrows "open/closed loop" to describe whether feedback is fed back, discussing only checks within agreed scope.
Open loop: Proceed with existing materials or plans but do not bring review or use results back to verify issues, decide handling, and re-check effects. The model may keep updating, but no one checks whether updates solved the corresponding problems.
Closed loop: Feed results back into work, verify issues per evidence, decide whether to adjust and where, then check if handling worked. Handling may not change the ontology—could be data fixes, query changes, or retaining original definitions after verification.
Example: In maintenance, equipment listed with only a "user feedback" record is not fully handled. After verification finds the query missed an existing exclusion condition, correcting the query and re-checking the original error and related samples closes that issue.
Both orientations can do this well or miss it. Public opinion questions keep changing, yet current viewpoint attribution can still be re-checked; maintenance queries have clear goals but may go unhandled after launch. Maintenance duration and delivery count do not determine whether feedback is connected.
Therefore, "open/closed loop" cannot be used as synonyms for the two construction orientations. Saying an issue is closed-loop requires stating what was checked, who handled what, and which parts remain outside this responsibility scope.
Distinguishing from the Open World Assumption
Another technical term to separate: the "open world assumption" in knowledge representation (e.g., OWL) concerns what inferences can be drawn when facts are missing. It is not the same as open exploration. A goal-clear maintenance query cannot automatically deem equipment normal just because maintenance records are absent. Reference: https://www.w3.org/TR/owl2-primer/
Future articles (05-07) will discuss how feedback is verified, how modifications are adopted, and how effects are evaluated—arrangements applicable to both orientations.
Here open/closed loop only describes whether results enter verification, handling, and re-check; both construction orientations can arrange feedback loops, with responsibility and check scope still needing clarification.
References
Natalya F. Noy, Deborah L. McGuinness, Ontology Development 101: A Guide to Creating Your First Ontology , Section 3 and Step 1. Used for iterative development, revisable scope, and competency questions. The article's two orientation names, comparisons, cases, and operational suggestions are the author's organization, not classifications prescribed by the guide. URL: https://protege.stanford.edu/publications/ontology_development/ontology101-noy-mcguinness.html
W3C, OWL 2 Web Ontology Language Primer (Second Edition) , Section 2. Used to explain the knowledge representation meaning of the open world assumption, not to define the article's construction orientations. URL: https://www.w3.org/TR/owl2-primer/
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
