{"id":"7c008de9-89c9-4d41-971c-8d7af94dbf42","arxiv_id":"2507.05767","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"The authors introduce CMO, an ontology that links skills, occupations, and courses, plus a layered architecture, and demonstrate it only on a hand-built example.","lead":"This paper proposes a layered ontology-based framework, the Competency Management Ontology (CMO), for connecting competencies, jobs, and training programs in a shared semantic model. It demonstrates the idea with a small SPARQL-based case study on the French ROME 4.0 directory, but the benefits are described as expected rather than measured.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed 'semantic reasoning' is not in the demonstration: Listing 2 is an exact-URI set difference over hand-inserted triples, so the framework's core automation claim rests on data entry, not inference.","rationale":"The reader's weakest-assumption analysis correctly identifies the reliance on ROME 4.0 data quality and on manually represented competencies. I would emphasize a more internal weakness: even granting perfect input data, the demonstrated gap-detection query does not use the ontology's semantics. A job requirement stated as a super-competency or as a level would not match a learner's sub-competency unless the URIs are identical, so the approach does not yet deliver the 'semantic reasoning' that the abstract and Section 4 claim. This is a correctness risk in the core mechanism, not just an evaluation gap. The paper is explicitly an early framework, and the conclusion acknowledges only a 'première validation exploratoire', so I do not think rejection is warranted. The concern is addressable by adding explicit inference rules or SPARQL property paths, publishing the ontology artifact, and re-running the case study on real or realistic ROME 4.0 instances. Because that is exactly the kind of condition the reader already attached, I would keep the CONDITIONAL verdict unchanged.","tokens_in":11309,"tokens_out":6531,"duration_ms":83389,"concrete_test":"Publish the CMO OWL file, or reconstruct a minimal graph from Figures 2-4 with one consistent predicate vocabulary, and run Listing 2 with a standard OWL reasoner after asserting that a skill Louis Le possesses is a sub-competency of a required skill, e.g. cmo:Python01 cmo:aSousCompetence cmo:Python02. If the missing-skill result changes and Python02 is no longer flagged, the semantic vocabulary is doing load-bearing work. If the result is unchanged because FILTER NOT EXISTS is purely syntactic, then the claimed semantic reasoning reduces to set difference and the central enhancement claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step for the paper's central claim is that CMO's ontology semantics actually drive competence-gap detection and training recommendation. In the only executable demonstration, Listing 2, the gap computation is a syntactic FILTER NOT EXISTS over manually asserted triples: it finds skills linked to M1405 via cmo:includeCompetence and excludes those Louis Le links via cmo:possedeUneCompetence. This uses none of the machinery the paper presents as distinctive: no use of 'a sous-compétence', no transitive or subsumption reasoning, no level/score comparison, and no rule that connects a possessed skill to a required skill unless the URIs match exactly. Table 3 is therefore predetermined by the hand-built profile, and the claimed enhancement from 'semantic reasoning' is not demonstrated. The recommendation step is even less specified: the formation DS25 is asserted in the scenario, not derived by any rule or query. Additionally, Listings 1 and 2 use inconsistent predicate names (cmo:possedeCompetence vs. cmo:possedeUneCompetence), so as printed the two queries cannot both run against one consistent ontology. This is an internal correctness risk, not merely a missing external benchmark.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a French-language ontology-based framework (CMO, Competency Management Ontology) for competence management, aimed at training, recruitment, and career-planning applications. The framework is described as a five-layer architecture (data, semantic, intelligence/recommendation, application/interaction, integration/interoperability) aligned with standards such as ROME 4.0 and ESCO. The authors model competencies, sub-competencies, levels/scores, jobs, trainings, certifications, temporal entities, and professional mobility. The central claim is that by leveraging ontological models and semantic reasoning, the framework enhances automation of competence-to-job matching, personalization of learning recommendations, and career planning. The paper presents a case study in which a learner, Louis Le, is assessed against the Data Scientist occupation (ROME M1405); two SPARQL queries retrieve his competencies and identify missing competencies, and a training (DS25) is recommended. The paper presents result tables and discusses expected impacts and future work.","tokens_in":11553,"tokens_out":4763,"duration_ms":51974,"significance":"If the central claim were substantiated, the proposed framework would offer a useful contribution: a structured, standards-aligned ontology that interlinks competencies, occupations, training programs, and certifications, potentially improving interoperability between HR and learning platforms. The manuscript provides a comprehensive conceptual model and, notably, publishes the actual SPARQL queries and result tables, which makes the demonstration transparent and falsifiable. However, the evidence presented does not support the claim that semantic reasoning or inference plays a role in the demonstrated functionality. The executable part is a direct retrieval over manually inserted triples, and the recommendation step is asserted rather than derived. As a framework proposal with an illustrative mock-up, the paper has merit; as a validation of improved automation or personalization, it is not yet convincing. The narrow, self-contained case study is explicit about being exploratory, but the abstract and several sections promise more than the demonstration delivers.","major_comments":[{"comment":"The gap-detection query is a syntactic set difference: it selects resources linked to cmo:M1405 via cmo:includeCompetence and filters out those linked to cmo:LouisLe via cmo:possedeUneCompetence. Because the triples were manually inserted in the scenario, Table 3 restates the input data and does not exercise any of the semantic machinery (sub-competence relations, transitive reasoning, level comparison) that Section 4 presents as the framework's contribution. The central claim that semantic reasoning improves automation is therefore not demonstrated by the only executable artifact in the paper. Either extend the example to use the ontology axioms (e.g., inferring gaps via sub-competence or level scores) or soften the claims to exact-match retrieval.","section":"§5.3, Listing 2"},{"comment":"The two queries use different predicate names for the same relation: Listing 1 has cmo:possedeCompetence and Listing 2 has cmo:possedeUneCompetence, while Figure 3 labels the relation 'possède une compétence'. As printed, the queries cannot both be evaluated against a single consistent ontology. The authors should unify the predicate names (and align them with Figure 4) and verify the example with an actual SPARQL engine.","section":"§5.3, Listings 1–2"},{"comment":"The recommended training 'Formation Data Scientist DS25' is asserted into the scenario rather than derived. No query in §5.3 links missing competencies to training offers via cmo:developpeLaCompetence or any prerequisite relation. Hence the paper's claim that the framework 'permet de recommander des formations ciblées' is not backed by the demonstration; a query that selects trainings developing the missing skills would be needed.","section":"§5.2, Fig. 4"},{"comment":"The gap query ignores the level/score information that the ontology explicitly models. FILTER NOT EXISTS tests only the existence of a competence triple; a learner who possesses a required skill at a too-low level (e.g., Python at Niveau01) would not be flagged as lacking the skill unless the required skill is represented as a different URI. The case study sidesteps this by naming Python02 as a distinct competence, but the framework's own level mechanism is not exercised. A level-aware comparison (e.g., required minimum level vs. acquired level) should be part of the demonstration.","section":"§5.3, Listing 2"},{"comment":"The section 'Résultats attendus et impact' is prospective, and the paper's conclusion calls the study 'une première validation exploratoire'. There is no evaluation on real ROME 4.0 data, no comparison with existing matching methods, and no user study. The invented profile and job data in Table 3 mean the results cannot distinguish framework behavior from data-entry effects. If the contribution is a framework proposal, the claims should be framed accordingly; if the contribution is validation of automation benefits, an actual evaluation is required.","section":"§5.4"}],"minor_comments":[{"comment":"'Achitecture' should be 'Architecture' in the section heading.","section":"§4.1"},{"comment":"The PREFIX declarations contain a space before the trailing '#' (e.g., <http://gamaizer.ia/cmo #>), which would cause a parse error in standard SPARQL; the correct IRI should end with 'cmo#' or a valid base IRI.","section":"§5.3, Listings 1–2"},{"comment":"'OpenClassRooms' should be 'OpenClassrooms'.","section":"§5.2"},{"comment":"There is a typo in the conclusion: 'recruitement' should be 'recrutement'.","section":"§6"},{"comment":"Table 2 includes individuals (Henri, Marc, Sophie) who do not appear in the case study; their presence is unexplained and may be intended for future scenarios, but this should be stated.","section":"Table 2"},{"comment":"The figures are difficult to read in the PDF; larger labels and higher resolution would help.","section":"Figures 2 and 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is best understood as a framework proposal with an illustrative mock-up, not as an empirical validation. I would suggest the editor ask the authors to either (a) make the case study actually exercise the reasoning/level-handling features of the ontology and fix the query inconsistencies, or (b) explicitly label the scenario as illustrative and move the automation claims to future work. As submitted, the mismatch between the abstract's promise and the evidence is too large for acceptance, but the conceptual framework is coherent and could be publishable after substantial revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's the short version: this paper describes a new ontology, CMO, for linking competencies, jobs, training, and certifications, but the only demonstration is a hand-built instance that a simple SPARQL query reads back. The paper does not actually show the semantic reasoning it claims as its selling point.\n\nWhat's new: the CMO model itself. The combination of classes and properties (has-subskill, level scales, prerequisite conditions, certification validity, temporal entities) in one OWL-style ontology, plus the five-layer architecture, does not appear in the cited prior work. It's a genuine artifact, though incremental relative to Miranda, Paquette, Draganidis, and others. The figures are clear, and the literature review is honest about what existing taxonomies and ontologies cover.\n\nWhat it does well: the design covers many relevant relations for competence management, and the authors are upfront in the conclusion that this is exploratory validation. The SPARQL queries are simple and readable.\n\nSoft spots: the case study is circular. Listing 2 is a FILTER NOT EXISTS over triples manually placed in the ontology. The gap list is a set difference between the job's required skills and Louis's profile; no subsumption, no transitivity, no level comparison, no rules. The recommended training DS25 is asserted, not derived. So the central claim that semantics drive matching and recommendation is unsupported by the evidence shown. Also, Listings 1 and 2 use different predicate names (possedeCompetence vs possedeUneCompetence), so as printed they cannot both run on a consistent ontology. That's a minor but real internal correctness issue. And there's no OWL file, no real data, and no baseline comparison.\n\nIn proportion: these are fixable, and none are fatal if the paper is positioned as a conceptual proposal. But the abstract and introduction promise more than the demonstration delivers.\n\nWho this is for: people building competence ontologies or HR/education recommender systems. They'll find a useful catalogue of classes and relations. I'd read it for that, but I wouldn't cite it as a validated framework without the artifact.\n\nRecommendation: it deserves a serious referee, but I'd expect major revision on the evaluation and the query inconsistencies. The core idea is plausible; the gap is in the demonstration, not the design.","headline":"A coherent but unevaluated ontology framework for competence management; the case study reads back hand-inserted data and never exercises the claimed reasoning.","tokens_in":12095,"tokens_out":2033,"would_cite":false,"duration_ms":23167,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that an ontology linking people, jobs, and courses can automatically compute the exact skills a learner lacks for a target job and recommend the training that fills the gap.","keywords":["competence ontology","knowledge graph","skill-gap detection","SPARQL","semantic reasoning","learning recommendation","career planning","ROME 4.0"],"falsifier":"Run the two SPARQL queries on a cohort of job seekers and compare the machine-generated gap lists with an independent human recruiter's assessment for the same target job; if recruiters identify essential competencies that the FILTER NOT EXISTS query misses, or reject listed gaps as irrelevant, the framework's recommendation is not yet correct. Alternatively, after a learner completes the recommended training and updates the profile, re-running the missing-skills query should return an empty gap; if it does not, the training-to-competence mapping is incomplete.","tokens_in":11116,"feed_emoji":"🎯","tokens_out":4994,"duration_ms":53771,"temperature":0.7,"pith_summary":"This paper proposes a competence-management ontology (CMO) and a layered framework that represents competencies, occupations, and training programs as a structured knowledge graph. The central claim is that putting jobs, people, and courses in one semantic model turns skill-gap detection into a simple database query: list the competencies a job requires, subtract the competencies a person already has, and the remainder is the gap. The paper demonstrates this with a concrete case in which a learner with basic Python wants to become a data scientist, and the ontology automatically identifies the missing competencies and recommends a targeted training course. If the framework works as claimed, it would let job seekers, recruiters, and training providers share one dynamic, interoperable description of skills, making retraining faster and recruitment more precise.","feed_headline":"Ontology maps job requirements straight to missing skills","feed_subtitle":"A SPARQL gap query turns a learner profile and a target occupation into a personalised training plan.","key_machinery":"The load-bearing object is the Competency Management Ontology (CMO), an OWL/RDF model in which competencies are typed as social, cognitive, technical, or linguistic, linked by sub-competency relations, and associated with national and international reference directories. The decisive mechanism is the SPARQL gap query: one query retrieves a learner's competencies with their levels, and a second query selects every competence required by a target occupation and filters out those already held, leaving exactly the missing set. A layered architecture routes data through a semantic layer into a recommendation layer, so the computed gap can be mapped to a training course that develops the missing competencies.","core_discovery":"The paper's central discovery is that competence management can be reformulated as an ontological model where competencies, occupations, training courses, certifications, and learner profiles are all instances linked by typed relations. In this model, a job's requirements are expressed as a set of competencies through the includeCompetence property, and a person's holdings are expressed through possedeUneCompetence; the skill gap is then the set difference between these two sets, computed by a SPARQL query with FILTER NOT EXISTS. The case study shows this operation identifying three missing competencies for a learner targeting the data-scientist occupation and linking those gaps to a specific six-month training course. On this basis, the paper argues that the framework can automate competence-to-job matching, personalize learning recommendations, and support career planning.","pith_inferences":["The paper stops at one case study; a direct extension would be to run the gap query across many occupations and many learner profiles, then measure whether the recommended courses actually close the detected gaps when learners re-query after training.","Because the ontology models professional mobility only schematically, it could be extended to chain occupations: the gap between a current job and a distant target job could be split into intermediate steps, each covered by a training module.","The temporal data in the model suggests a time-aware employability measure: a skill-gap score could decay as certifications expire or as job requirements in the reference directory are updated, which the paper notes but does not implement.","A testable consequence of the semantic approach is that if job-requirement labels in the reference directory change, the ontology can propagate the change to affected profiles and course mappings, whereas a static taxonomy would require manual updates."],"forward_implications":["A learner who wants to change jobs can be told exactly which competencies are missing and pointed to a course that covers them, shortening requalification paths.","Recruiters can query for candidates by required competencies rather than by job titles, because profiles and jobs share the same semantic vocabulary.","Training providers can adapt course offerings to the gaps actually observed in a population, since the ontology exposes both job requirements and course coverage.","Profiles can stay current automatically: completing a course, passing a certification, or finishing a job updates the profile, and the gap query can be run again with the new state.","Because courses, certifications, and work experiences carry start dates, durations, and validity periods, the same framework can track how competencies evolve over time."],"supporting_citations":[{"why":"Supplies the central precedent of an ontology-based model for competence management in human-resources systems, which this framework extends by linking competencies to jobs and training.","marker":"[11]"},{"why":"Provides the organizational-memory background for structuring learning and competence knowledge in a reusable framework.","marker":"[1]"},{"why":"Documents the limitations of traditional competency-based workforce development that motivate the move to an ontological representation.","marker":"[3]"},{"why":"Shows an earlier ontology-based tool for competency management and learning paths, a direct precursor to the recommendation layer.","marker":"[6]"},{"why":"Defines an ontology and software framework for competency modeling, grounding the educational side of the model.","marker":"[12]"},{"why":"Supplies the skill-driven job-transition recommendation approach that the intelligence layer cites for career planning.","marker":"[4]"},{"why":"Supports the recommendation engine with a constraint-based recommender built on RDF knowledge graphs.","marker":"[8]"}],"fun_headline_variants":["SPARQL gap query turns job specs into training plans","Ontology links job needs to personal learning paths","Skill gap query maps jobs to missing competencies","Semantic framework matches profiles to job requirements","Ontology automates job-skill matching and training advice"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire skill-gap result is only as good as the two input lists—the reference directory's list of required competencies for the job and the ontology's stored list of competencies for the person—and the paper does not assess the accuracy or completeness of either list.","fun_headline_variants_meta":{"raw":{"variants":["SPARQL gap query turns job specs into training plans","Ontology links job needs to personal learning paths","Skill gap query maps jobs to missing competencies","Semantic framework matches profiles to job requirements","Ontology automates job-skill matching and training advice"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000161,"raw_usage":{"total_tokens":1182,"prompt_tokens":840,"completion_tokens":342,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":456,"completion_tokens_details":{"reasoning_tokens":269}},"tokens_in":456,"tokens_out":342,"duration_ms":3877,"temperature":1.0,"reasoning_tokens":269,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:18:16.244653+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the two SPARQL queries on a cohort of job seekers and compare the machine-generated gap lists with an independent human recruiter's assessment for the same target job; if recruiters identify essential competencies that the FILTER NOT EXISTS query misses, or reject listed gaps as irrelevant, the framework's recommendation is not yet correct. Alternatively, after a learner completes the recommended training and updates the profile, re-running the missing-skills query should return an empty gap; if it does not, the training-to-competence mapping is incomplete.","supporting_citations":[{"cited_title":"Référentiels des compétences et des mé- tiers : une approche ontologique","cited_arxiv_id":null,"evidence_quote":"Supplies the central precedent of an ontology-based model for competence management in human-resources systems, which this framework extends by linking competencies to jobs and training."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the organizational-memory background for structuring learning and competence knowledge in a reusable framework."},{"cited_title":"fait partie du référentiel de compétences","cited_arxiv_id":null,"evidence_quote":"Documents the limitations of traditional competency-based workforce development that motivate the move to an ontological representation."},{"cited_title":"A systematic review of evidence-based general competency models : Development of a general competencies taxonomy","cited_arxiv_id":null,"evidence_quote":"Shows an earlier ontology-based tool for competency management and learning paths, a direct precursor to the recommendation layer."},{"cited_title":"A constraint-based recommender system via rdf knowledge graphs","cited_arxiv_id":null,"evidence_quote":"Defines an ontology and software framework for competency modeling, grounding the educational side of the model."},{"cited_title":"Formation Data Scientist DS25","cited_arxiv_id":null,"evidence_quote":"Supplies the skill-driven job-transition recommendation approach that the intelligence layer cites for career planning."},{"cited_title":"Skill-driven recommendations for job transition pathways","cited_arxiv_id":null,"evidence_quote":"Supports the recommendation engine with a constraint-based recommender built on RDF knowledge graphs."}],"review_version":1}