Decoding Deception with TAXODIS – a Taxonomy of Disinformation Cues for Fine-grained Text Labeling

TAXODIS is a well-motivated, clearly presented 66-concept taxonomy of linguistic disinformation cues, published as an open SKOS/RDFS resource on Zenodo (DOI, CC-BY, resolvable namespace). For an Ontology Description the resource is relevant and of good quality: the systematic-review methodology is sound, and the SKOS modelling together with the Open Annotation / schema.org linking and the SPARQL examples is convincing. The revision addresses most earlier concerns: novelty over the prior unimplemented version is now explicit in Section 3 (new "theme" dimension, 48 vs 18 leaf concepts); CimpleKG and DBKF are integrated; the SKOS-vs-ontology decision is justified and flagged for future formalization; the overlap with previous publications and the relation to existing claim vocabularies are also clarified.

Two issues remain and motivate a light revision rather than acceptance:

1. Evaluation: the only quantitative evidence is still the inter-annotator agreement (Cohen's Kappa = 0.72), measured on a reduced set of 15 derived labels rather than the full taxonomy; there is no new evaluation of the SKOS implementation, and no validation of annotation correctness against a gold standard or expert judgment: agreement demonstrates consistency, not validity; the AI/explainability use cases (Section 6.2) remain entirely prospective; either a small added demonstration or an explicit framing of these as untested future directions would strengthen the paper; the newly added download count is an adoption signal, not an evaluation

2. Resource and reproducibility: the Zenodo deposit contains the taxonomy (TTL plus two display PDFs) but no README, and none of the annotation data underlying the agreement result, so that evaluation cannot be reproduced from the provided artifacts; adding a README and depositing or persistently linking the annotation data would close this; the repository choice (Zenodo) is appropriate, and the deposit is otherwise complete and well-formed.

Although the authors have addressed to some extent multiple points raised in the previous review regarding the relationship to prior work, related resources, and existing vocabularies, the evaluation and novelty remain major issues that have not been properly addressed in this revision. The paper still lacks (1) a proper evaluation of the proposed taxonomy, and (2) a clear differentiation from the authors' previous work.

Updated comments following the article revision:
1) Novelty and relationship to prior work (not addressed):
As mentioned in the previous review, the article extends two previous publications ([13] and [9]). The new revision adds very limited clarifications: the new dimension is now named explicitly ('theme') and the paragraph in Section 3 mainly states that leaf concepts grew from 18 to 48 compared to the previous paper. As a result, the changes fail to adequately address the concern previously raised and the differentiation from prior work remains insufficient.

2) Conceptual scope (not addressed):
The research question has been extended and now includes "structuring and explicating the linguistic dimensions of disinformation as a coherent conceptual space". However, the design of the resource remains fundamentally annotation-oriented and the taxonomy's structure, methodology, and intended uses are unchanged from the previous version.

3) Related work (partially addressed):
The related work now includes a description of CimpleKG and the authors acknowledge that it includes the extraction of textual features. However, the authors fall short in contextualising these features in relation to their own taxonomy. In particular, in Section 5, the authors claim that "fine-grained linguistic characteristics of disinformation content are not explicitly represented in existing vocabularies," which directly contradicts what they state in Section 2. CimpleKG already extracts and represents emotions, sentiment, political leanings, propaganda techniques, and conspiracy theory mentions at the claim level, features that map directly onto TAXODIS categories. The justification provided in Section 5 is therefore weak and the example queries do not demonstrate what TAXODIS uniquely enables over existing resources. A more precise account of TAXODIS's added value is needed in both the related work and Section 5.

4) Modelling (addressed):
The authors now discuss in more detail the choice of SKOS in Section 3 and acknowledge that richer relationships between concepts could benefit from a more formal ontological representation. They also note that future work could investigate such more complex representations.

5) Methodology and usage (partially addressed):
The authors added a paragraph in Section 5 acknowledging that some queries can already be supported by existing vocabularies and knowledge graphs. However, the queries themselves are unchanged from the previous version, and no new query demonstrating what TAXODIS uniquely enables has been added.

6) Evaluation (not addressed):
The evaluation remains largely unchanged from the previous version of the article. The main addition is a sentence reporting 150 Zenodo downloads as "a complementary indicator of early adoption," which does not constitute an evaluation. There is still no new evaluation of the SKOS implementation or its integration with other resources.

Additional notes: Some references do not appear in the article body and are unrelated to the article [33, 54, 57 and 80]. These references should be removed from the paper.

The authors addressed all the comments and concerns raised during the previous round of reviews. The revisions have increased the quality and clarity of the article.
Therefore, the manuscript now meets the journal's standards for publication. I recommend its acceptance for publication.