{"id":"21708e88-9950-46b4-a2d4-b034eca9c89f","arxiv_id":"2412.05289","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey of Knowledge Graph visualization tools finds that intuitive UI and performance are widely met but modularity is rare, with embedding-based tools splitting into exploratory and explanatory perspectives.","lead":"This essay reviews 72 papers on Knowledge Graph visualization and separates embedding-based tools into exploratory (embeddings guide navigation) and explanatory (visualizations explain embeddings) camps. It finds that intuitive interfaces and performance are common, while modularity is rare, and suggests modularity and relation visualization as future directions.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'rarely modular' conclusion rests on an undisclosed corpus and unvalidated binary feature coding, so the central synthesis cannot yet be separated from selection and measurement bias.","rationale":"The paper is a structured, honest survey with a reproducible query, and its synthesis is plausible. The most load-bearing condition for the central claim is that the selected corpus and the binary feature assessments faithfully represent the field. That condition is not currently verifiable: the 72-paper list is missing, the exclusion criteria are vague, and the feature coding lacks inter-rater validation. I looked for a stronger internal inconsistency, such as a direct contradiction in Table 2's checkmark-to-column mapping or a category whose definition guarantees the query-support finding, but the tables are too under-specified to establish such a flaw; that under-specification is itself part of the reproducibility problem. The reader's CONDITIONAL verdict captures this accurately, and my concern reinforces the need for the same disclosures rather than moving the verdict. Hence no change is recommended.","tokens_in":13205,"tokens_out":8273,"duration_ms":76167,"concrete_test":"Reconstruct the corpus by running the exact Scopus query from Section 4, applying the stated filters, and asking the authors to release the full list of 72 included papers. Then have two independent coders re-code the four features in Tables 1 and 2 from the retrieved full texts using the Section 3 definitions, and report Cohen's kappa plus the resulting modularity rate. If the re-coded modularity rate exceeds roughly 50% in either table, or if inter-coder agreement falls below substantial (kappa < 0.6), the 'rarely modular' claim is not robust and Section 7 should be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claims in Section 7 — that intuitive UI and performance are 'usually met' and that frameworks are 'rarely modular' — depend entirely on the binary checkmarks in Tables 1 and 2. Those checkmarks are derived from a 72-paper corpus whose membership is not disclosed: Section 4 gives only the Scopus query and exclusion bullets, including 'low-ranking journals or conferences' and 'closed-access papers', without defining any threshold. This is a genuine selection risk: if modular systems tend to be described in lower-ranked venues or in closed-source product documentation, the 'rarely modular' conclusion would be an artifact of the filter. Additionally, modularity and intuitive UI are coded from the papers' descriptions rather than from the artifacts themselves; Section 3 even concedes that intuitive UI is 'naturally subject to personal perspective.' Absence of a modularity discussion in a paper is thus treated as absence of modularity in the tool. Without the list of 72 papers and a coding protocol, the reader cannot verify whether the observed counts (about 4/10 general frameworks and 0/8 embedding tools) represent the field or the authors' selection and judgment.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper is an essay/survey on recent trends in Knowledge Graph (KG) visualization, with a focus on frameworks that use Knowledge Graph Embeddings (KGEs). The authors define four challenges for KG visualization--modularity, intuitive UI, performance, and query support--and use a Scopus-based literature search to select 72 papers, of which 18 frameworks are tabulated: 10 general KG visualization frameworks in Section 5 and 8 embedding-based approaches in Section 6. The embedding-based approaches are further divided into exploratory and explanatory perspectives. The central conclusions, stated in Section 7, are that intuitive UI and performance requirements are usually met, query support is present in general and exploratory tools but not in explanatory embedding tools, and that frameworks are rarely modular, so visualizations are typically fixed. The paper also identifies relation visualization as a peculiar and underaddressed element of KG visualization and suggests it as a future research direction.","tokens_in":13411,"tokens_out":6238,"duration_ms":57032,"significance":"If the survey's synthesis is accurate, the paper provides a useful organizing taxonomy--exploratory versus explanatory use of embeddings--and draws attention to two concrete gaps in current KG visualization practice: modularity and relation-level visualization. The four-challenge framing is clear and could be reused by other researchers. The paper also makes a specific, checkable observation: embedding-based visualization tools generally rely on offline embedding computation to meet performance requirements, yet none of the surveyed tools implements a modular architecture for adding or modifying views. These are valuable claims for the visualization and semantic-web communities. However, the significance is conditional on the transparency of the survey methodology: the paper's conclusions are counts over binary feature checkmarks in Tables 1 and 2, and the corpus and coding procedure behind those checkmarks are not disclosed in sufficient detail.","major_comments":[{"comment":"Section 4 reports that 656 Scopus records were reduced to 72 included papers, but the paper never lists the 72 papers and never explains why Tables 1 and 2 summarize only 18 frameworks. Because the Section 7 conclusions (intuitive UI and performance are usually met; modularity is rare) are counts over the tabulated frameworks, the missing corpus list and flow diagram prevent the reader from verifying the synthesis. Please provide the complete list of included papers, a PRISMA-style flow diagram, and a statement of how the 54 included-but-not-tabulated papers were used, or why they do not appear in the tables.","section":"Section 4"},{"comment":"The exclusion criteria 'closed-access papers' and 'low-ranking journals or conferences' are not operationalized. The survey gives no venue-ranking source, no rank threshold, and no count of papers excluded per reason. Closed-access exclusion is an availability filter rather than a quality filter, and it creates a concrete selection risk for the central modularity claim: modular systems documented in lower-ranked or closed-access venues would be invisible to the survey. Please define the exact criteria, report the number excluded for each reason, and discuss how the conclusions might change under plausible variations of the thresholds.","section":"Section 4"},{"comment":"The binary checkmarks in Tables 1 and 2 are the quantitative basis for the Section 7 conclusions, but the paper does not provide a coding protocol or per-cell evidence. Section 3 itself concedes that intuitive UI is 'naturally subject to personal perspective,' and the paper states only that the authors inspect which features are present or absent. Without an explicit rubric (e.g., what counts as modular, whether performance claims are taken from the paper or measured), the counts cannot be separated from the authors' judgment. Please add a coding protocol, ideally with an inter-rater reliability check, and a supplementary table that gives the evidence for each checkmark.","section":"Tables 1 and 2, Section 7"},{"comment":"Table 2 includes general-graph embedding tools (CorGIE, GEMVis, BiaScope) that Section 6.2 explicitly says are 'not tailored to KGs.' The conclusion that embedding-based frameworks are rarely modular is then reported as a finding about KG visualization at large. This conflation is load-bearing for the central claim. Please report KG-specific and general-graph tools separately, or justify why the general-graph tools are included in the counts that support the KG-specific conclusion.","section":"Section 6.2, Table 2"}],"minor_comments":[{"comment":"The phrase 'an user' should be 'a user.'","section":"Section 1"},{"comment":"The phrase 'As as example' should be 'As an example.'","section":"Section 3"},{"comment":"The description of GEMVis says 'in a5 view framework'; this should be 'in a 5-view framework,' and the similar phrase '5 view framework' should be revised for clarity.","section":"Section 6.2"},{"comment":"The word 'accomodates' in the CorGIE description should be 'accommodates.'","section":"Section 6.2"},{"comment":"Figure 2 would benefit from labeled axes and explicit counts on the vertical axis; currently the reader cannot read the exact number of papers per year.","section":"Figure 2"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a visualization or semantic-web journal, and the direction of the findings is plausible. My recommendation of major_revision is driven entirely by methodology transparency: the central claims are counts over an undisclosed corpus and unvalidated binary codes. I would not require new experiments, but I would require a supplementary corpus list, a flow diagram, operationalized exclusion criteria, and a coding protocol before the synthesis can be accepted as a reliable survey."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a competent survey with a genuinely useful organizing idea: splitting KG-embedding visualization into exploratory and explanatory perspectives. The four challenges (modularity, intuitive UI, performance, query support) give the reader a clean evaluative frame, and the tables offer a quick map of the field. The discussion of each framework is balanced, and the observation that relation visualization is largely overlooked is a concrete, actionable gap. So there is real value here.\n\nThe soft spots are real but not fatal. The methodology says 72 papers were included, yet Tables 1 and 2 show only 18 frameworks. The other 54 are presumably background or excluded from the tables, but the paper never says. The exclusion criteria are vague: \"low-ranking journals or conferences\" and \"closed-access papers\" are not operationalized, and only Scopus was used. The feature coding is binary checkmarks with no protocol, and the authors freely admit in Section 3 that intuitive UI is subjective; they narrow it to one objective factor, which helps, but the modularity and query-support judgments are still made from paper descriptions rather than hands-on testing. The central conclusion that \"frameworks are rarely modular\" is plausible, but it is only as strong as that undisclosed selection and coding. That is a transparency problem, not evidence of fabrication.\n\nThe claim of being the first survey on this angle is not verified against prior surveys. That is a minor issue.\n\nWho is this for? Someone who wants a compact, current map of KG visualization tools and a sense of where the open problems are. It is not ground-breaking, but it is a credible synthesis.\n\nMy recommendation: send it to peer review. Ask for an appendix listing the 72 papers, a defined venue-ranking threshold, and a coding protocol before it is accepted. The taxonomy alone is worth publishing.","headline":"A useful survey with a clean taxonomy, but the headline claims on modularity need more transparency about the corpus and coding before they can be taken at face value.","tokens_in":13854,"tokens_out":2405,"would_cite":true,"duration_ms":23570,"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":"This essay argues that knowledge-graph visualization has largely met usability and performance goals, but rarely achieves modularity, and that embedding-based tools split into exploratory and explanatory camps with differing feature…","keywords":["knowledge graph visualization","knowledge graph embeddings","visual analytics","graph embedding explanation","modularity","exploratory and explanatory visualization","survey"],"falsifier":"A replication that broadens the corpus, for instance by adding lower-ranked venues, open-source repositories, and non-English work, and then finds that a substantial share of the additional tools are modular or visualize relations would directly undercut the paper's two main conclusions. Concretely, if more than half of, say, 30 newly included tools had plugin-based architectures or relation-specific views, the claim that frameworks are rarely modular and relations are generally overlooked would need to be revised.","tokens_in":13019,"feed_emoji":"🕸️","tokens_out":5798,"duration_ms":50426,"temperature":0.7,"pith_summary":"This essay reviews roughly a decade of research on visualizing knowledge graphs and knowledge-graph embeddings, organizing the field around four requirements: modularity, intuitive user interface, performance, and query support. The authors find that most reviewed frameworks meet the usability and performance requirements, and that query support is commonly present in general and exploratory tools, but that few frameworks are modular and visualizations are typically fixed. They also distinguish exploratory uses of embeddings, which help users navigate a graph, from explanatory uses, which help users understand what the embeddings captured, and find the two perspectives have different feature profiles. The reason to care is that the survey locates where the field is mature and where it is not, pointing to modularity, relation-focused views, and query support for explanatory tools as the concrete directions for future work.","feed_headline":"Knowledge-graph tools: intuitive but rarely modular","feed_subtitle":"An essay covering 2014–2024 finds usability and speed largely solved, while modularity and relation views lag.","key_machinery":"The paper's analytic engine is a two-part classification. First, each surveyed framework is scored against four requirements that the authors define: modularity, meaning the architecture is extensible and adaptable; intuitive UI, meaning the interface is clear and largely independent of the graph's structure; performance, meaning the handling of large graphs; and query support, meaning the user can explore without learning a query language. Second, embedding-based visualization approaches are split into an exploratory perspective, where embeddings guide navigation and summarization, and an explanatory perspective, where visualizations expose what embedding models captured. These two axes organize the review and drive its conclusions.","core_discovery":"On the paper's own terms, the central claim is that the state of knowledge-graph visualization can be summarized as: user interfaces are generally intuitive and performance demands are usually met, but modular architectures are rare, so most visualizations are fixed rather than extensible. In the embedding-specific literature, the authors separate tools that use embeddings to guide exploration of a graph from tools that explain the embeddings themselves; the exploratory tools generally provide query support, while the explanatory ones do not, and nearly all tools ignore relation embeddings and relation-focused visual features. The essay further claims that this asymmetry, together with the neglect of relations, is the main open frontier for KG visualization.","pith_inferences":["Editorial inference: The modularity gap may be partly a market artifact: many tools are closed-source industrial products with fixed feature sets, so a survey restricted to published frameworks may underrepresent modular open-source toolkits.","Editorial inference: Relation embeddings are a distinctive output of KGE models, so designing views that visualize relations by type, path, or learned vector direction could be a testbed for whether KG visualization can differentiate itself from general graph visualization.","Editorial inference: The challenge framing could be extended by treating explainability of embeddings not as a separate perspective but as a queryable view, so users could ask why an entity is close to another and get a visual answer.","Editorial inference: A direct empirical test of the survey's ranking would be a task-based user study comparing modular and fixed tools on the same KG, measuring extension time and user insight."],"forward_implications":["If the survey's picture is right, a new KG visualization framework that wants to add value should focus on modularity and relation-oriented views, since those are the least satisfied requirements.","Explanatory embedding tools, which today generally lack query support, would become more useful for instance-level analysis if querying were added.","Because most embedding-based tools compute embeddings offline, performance is rarely the bottleneck; the bottleneck is adaptivity and extensibility.","Semantic zooming and relation-specific encodings are named as concrete underused features that future systems could explore.","The exploratory/explanatory split suggests that hybrid tools, like the one reviewed tool that combines both perspectives, remain an underexplored design space."],"supporting_citations":[{"why":"Supplies the framing of users, challenges, and visualization opportunities for knowledge graphs that the paper's four requirements build on.","marker":"[30]"},{"why":"Reviewed as a modular SPARQL-result visualization framework; one of the few positive modularity data points in the survey.","marker":"[42]"},{"why":"Reviewed as a web-based visualizer with a modular back-end and front-end architecture; evidence that modularity is achievable.","marker":"[39]"},{"why":"Reviewed as the only tool explicitly combining exploratory and explanatory embedding perspectives; supports the claim about hybrid integration being rare.","marker":"[16]"},{"why":"Reviewed as an exploratory embedding-based guidance tool using TransR; supports the claim that exploratory tools generally provide query support.","marker":"[54]"},{"why":"Reviewed as an explanatory tool for graph embeddings with multiple linked views; supports the claim that explanatory tools lack query support.","marker":"[36]"},{"why":"Reviewed as an explanatory tool for diagnosing unfairness in graph embeddings; additional evidence for the explanatory-perspective feature profile.","marker":"[43]"}],"fun_headline_variants":["KG tools: modularity is the missing piece","Embedding visualizers skip relation views","Exploring vs explaining embeddings: a split","Graph viz lags on modular and relation views","Visualizing knowledge graphs: relations ignored"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the papers returned by the survey's Section 4 search query and its exclusion criteria form a representative sample of knowledge-graph visualization research; if significant tools from lower-ranked venues or non-indexed sources are missing, the conclusion that modularity is rarely met may not generalize.","fun_headline_variants_meta":{"raw":{"variants":["KG tools: modularity is the missing piece","Embedding visualizers skip relation views","Exploring vs explaining embeddings: a split","Graph viz lags on modular and relation views","Visualizing knowledge graphs: relations ignored"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000147,"raw_usage":{"total_tokens":1128,"prompt_tokens":833,"completion_tokens":295,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":449,"completion_tokens_details":{"reasoning_tokens":230}},"tokens_in":449,"tokens_out":295,"duration_ms":3715,"temperature":1.0,"reasoning_tokens":230,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:19:50.572767+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A replication that broadens the corpus, for instance by adding lower-ranked venues, open-source repositories, and non-English work, and then finds that a substantial share of the additional tools are modular or visualize relations would directly undercut the paper's two main conclusions. Concretely, if more than half of, say, 30 newly included tools had plugin-based architectures or relation-specific views, the claim that frameworks are rarely modular and relations are generally overlooked would need to be revised.","supporting_citations":[{"cited_title":"VizKG: A framework for visualizing SPARQL query results over knowledge graphs","cited_arxiv_id":null,"evidence_quote":"Reviewed as a modular SPARQL-result visualization framework; one of the few positive modularity data points in the survey."},{"cited_title":"Stunning Doodle: a Tool for Joint Visualization and Analysis of Knowledge Graphs and Graph Embeddings","cited_arxiv_id":null,"evidence_quote":"Reviewed as the only tool explicitly combining exploratory and explanatory embedding perspectives; supports the claim about hybrid integration being rare."},{"cited_title":"KGScope: Interactive Visual Exploration of Knowledge Graphs with Embedding- based Guidance","cited_arxiv_id":null,"evidence_quote":"Reviewed as an exploratory embedding-based guidance tool using TransR; supports the claim that exploratory tools generally provide query support."},{"cited_title":"Visualizing graph neural networks with corgie: Corresponding a graph to its embedding","cited_arxiv_id":null,"evidence_quote":"Reviewed as an explanatory tool for graph embeddings with multiple linked views; supports the claim that explanatory tools lack query support."},{"cited_title":"BiaScope: Visual unfairness diagnosis for graph embeddings","cited_arxiv_id":null,"evidence_quote":"Reviewed as an explanatory tool for diagnosing unfairness in graph embeddings; additional evidence for the explanatory-perspective feature profile."}],"review_version":1}