Hi-d maps: An interactive visualization technique for multi-dimensional categorical data

Benjamin A Watson; Radi Muhammad Reza

REVIEW 3 major objections 4 minor 25 references

Hi-d maps: An interactive visualization technique for multi-dimensional categorical data

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper introduces Hi-D maps, a space-efficient interactive visualization that maps multi-dimensional categorical data onto a regular polygon by hierarchical cuts parallel to its sides.

desk verdict A concrete, testable design for categorical visualization whose effectiveness claims are all unverified—worth refereeing, but not worth believing yet. read the letter →

arxiv 2507.07890 v1 pith:WZSTNP3P submitted 2025-07-10 cs.GR

Radi Muhammad Reza , Benjamin A Watson This is my paper · ORCID

classification cs.GR

keywords Hi-Dmapsmulti-dimensionalcategoricaldataspace-fillingvisualizationregularpolygonpartitionhierarchicalinteractivetreemapssunburst

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Dimensional categorical data — demographic records, survey answers, product attributes — is hard to see when each record combines several categories. The paper proposes Hi-D maps, a visualization that packs all such data into one regular polygon, cutting it hierarchically with lines parallel to each dimension's side so that every leaf polygon's area equals the fraction of records in that combination of categories. The motivation is practical: existing space-filling methods such as treemaps lose track of which dimension a cut belongs to, while radial methods such as sunbursts waste space by giving each dimension its own ring. Hi-D maps aim to keep the space efficiency of the former and the dimensional clarity of the latter in a single view, with interaction for re-ordering, zooming, and hiding dimensions.

What carries the argument

The central object is a regular polygon with a user-controlled, ordered sequence of sides, one per dimension. The construction cuts the polygon hierarchically with lines parallel to the chosen side, using binary search to place each cut so the child polygon areas match the sizes of the data partitions; recursion continues until all dimensions are used, and the leaf polygons are the categorical intersections. When the number of dimensions is even, the algorithm keeps the polygon's side count odd and leaves one side unmapped, avoiding parallel sides that would blur the angular encoding. This single recursive splitting rule is what carries both the space efficiency and the dimensional clarity.

What would settle it

A controlled experiment in which naive viewers are shown Hi-D maps of synthetic data and asked to name the dimension each cut represents and estimate relative intersection sizes; if accuracy falls steeply as dimensions grow or as polygon angles become similar, the central clarity claim fails.

Watch

Extended reading notes

Core claim

Hi-D maps is a space-partitioning visualization that maps the full data space of multi-dimensional categorical data onto a single regular 2D polygon of constant circumradius. Each dimension is assigned to one oriented side of the polygon, ordered clockwise, and the polygon is cut recursively with lines parallel to the current side, so the area of each resulting sub-polygon is proportional to the number of data points in that categorical intersection. The display uses orientation, hue, lightness, line thickness, and countable 'marble' glyphs as reinforcing cues, and supports interactive re-ordering, hierarchical browsing, and dimension-hiding. The paper's claim is that this retains the spatial efficiency of treemaps while providing the dimensional clarity of sunburst visualizations: all dimensions share the same space, but their identity is encoded by angle and color rather than by level counting or separate rings.

Load-bearing premise

The paper assumes that viewers can read which dimension a cut represents from its angle and hue, and judge intersection sizes from area, saturation, and marble counts, without training or perceptual bias; no user study in the paper yet tests that assumption.

Editorial extensions

If this is right

If Hi-D maps work as claimed, a viewer can see the relative size of every combination of categories in a single bounded 2D view, without following links or traversing rings.
The same display supports interactive dimension re-ordering, hierarchical filtering by clicking edges, and hiding uninteresting dimensions, so the space partition can be explored rather than just inspected.
Because the polygon is cut recursively, the technique extends naturally to hierarchical data, with innermost polygons as leaves.
The algorithm's cost is dominated by $p^n$ for $n$ dimensions with up to $p$ values each, so the method targets categorical data with few values per dimension, not numerical data.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural next step the paper leaves implicit is a formal comparison of angle-plus-hue dimension identification against level-counting in treemaps; if angle perception degrades more slowly than level counting, the method's clarity claim gains quantitative support.
The use of an odd number of polygon sides whenever dimensions are even suggests a general design rule: reserve one side to avoid parallel ambiguity, a trick that could be exported to other angular space-filling layouts.
Because the paper reports failure around 30 dimensions from circularity and crowding, one testable extension is a hybrid approach that adaptively switches to a different polygon or layout when angular separation falls below a perceptual threshold.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

The paper is a plausible design contribution with a clear algorithm and an honest limitations section, but its central claim—that Hi-D maps add clarity before perceptual limits are reached—is asserted, not demonstrated. If you read it for the algorithm, it's a reasonable layout scheme; if you read it for evidence, there is none.

What's actually new: the specific combination of hierarchically cutting a regular polygon with lines parallel to an ordered sequence of sides, with leaf area proportional to intersection size, plus dimension reordering and interactive browsing. I don't see that exact recipe in the treemap/sunburst/parallel sets literature they cite. The algorithm is simple and credible: binary-search area splits realize arbitrary proportions, and the recursive structure is clearly described. The authors also add multiple visual cues (hue, lightness, saturation, marbles, text) and interactive features like overview-detail, dimension hiding, and hierarchical browsing.

The soft spot is the effectiveness claim. The paper states it retains treemap-like space efficiency and sunburst-like dimensional clarity, and that it enables quick and effective analysis—but no user study, no perceptual experiment, and no empirical comparison support any of that. The limitations section concedes failure around 30 dimensions and defers user studies to future work. That's honest, but it means the core value is a hypothesis. I also note the marble-counting cue is a heuristic that may not help viewers compare areas, and the algorithm's exponential dependence on values per dimension limits it to small categorical domains, though they acknowledge that. No code or data are released, so independent replication of the figures isn't possible.

That said, the paper is internally coherent, the writing is clear, and the authors don't overclaim. The binary-search splitting logic is sound; there's no mathematical error I can find. The citation list looks appropriate, and the self-citation to their agent-based animation work is fine.

Who's this for? A reader working on categorical visualization or space-filling layouts could get ideas from the design. It's not yet a demonstrated technique, but it's a concrete, testable design proposal. I'd send it to peer review—a referee could push for a user study or at least a controlled comparison, and the design is specific enough to be worth that effort. I wouldn't cite it as evidence for effectiveness until evaluation exists, but I'd cite it as prior art in related work.

Recommendation: engage with it, but treat the effectiveness claims as open questions.

Best,

Referee Report

3 major / 4 minor

Summary. The paper presents Hi-D maps, a visualization technique for multi-dimensional categorical data. The data space is represented as a regular polygon; dimensions are assigned to polygon sides in a user-controlled order, and the polygon is recursively cut by lines parallel to those sides, with each leaf polygon's area proportional to the number of data records in the corresponding categorical intersection. The paper describes visual encodings (orientation, hue, lightness, saturation, countable 'marble' glyphs, and text), interaction mechanisms (dimension reordering, overview-detail, hierarchical browsing, dimension hiding, and animation), an algorithm with a stated time complexity, and a qualitative comparison with treemaps, sunburst visualizations, and parallel sets. The central claim is that Hi-D maps combine treemap-like spatial efficiency with sunburst-like dimensional clarity and help users analyze data quickly, effectively, and in an engaging manner, while acknowledging that the technique degrades around 30 dimensions.

Significance. The proposed layout is internally coherent and has design appeal. The recursive splitting of a convex polygon by lines parallel to chosen orientations can realize arbitrary leaf-area proportions, so the underlying area-proportionality claim is structurally sound. If the perceptual claims about orientation, hue, lightness, and marble counting are validated, the technique would be a useful addition to the multivariate categorical visualization toolbox, particularly because it keeps all dimensions within a single shared two-dimensional region. The paper is honest about limitations and clearly separates implemented features from planned evaluation. Its main weakness is that the central effectiveness claim is unmeasured: there is no user study, no perceptual experiment, and no empirical comparison with the named alternatives. The paper explicitly defers these to future work, so at present the core contribution is a plausible design hypothesis rather than a demonstrated technique. The absence of released code or supplemental material also limits independent replication of the figures and algorithm behavior.

major comments (3)

[Abstract and Section 5 (Conclusions and Future Work)] The central claim that Hi-D maps 'help' users analyze data and 'retain the dimensional clarity of sunburst visualizations' is not supported by any evaluation. Section 5 explicitly lists user studies and empirical comparisons as future work, and the abstract's closing statement that 'Hi-D maps may add clarity before those limits are reached' is unfalsifiable as stated. The load-bearing assumptions are that viewers can (a) infer the mapped dimension from cut orientation and hue, (b) read value ordering from lightness, and (c) reliably compare leaf-polygon areas using area, saturation, and marble counts. None of these is tested. I recommend adding a controlled experiment, at minimum measuring dimension-identification accuracy, area-comparison error, and task completion time against treemaps, sunburst, and parallel sets. Alternatively, the comparative and effectiveness claims should be systematically toned down to design hypotheses.
[Section 4, Algorithm 1] The split step assumes every sub-partition has positive size. The paper does not state how zero-count intersections are handled. In real categorical data, many high-dimensional intersections are empty; a zero-count partition would require a zero-area child polygon (a degenerate segment), and the recursive splitting of that degenerate polygon in subsequent dimensions is undefined. The paper should either specify a minimum-area or epsilon-based treatment for empty partitions, or explicitly restrict the method to data with no empty intersections and discuss the consequences for practical applicability.
[Section 3.2 (Comparison)] The claimed advantages over treemaps and sunburst visualizations are asserted qualitatively without operationalized measures. For example, the statement that Hi-D maps 'do not suffer as much from dimensional ambiguity' because they use angle and hue is a design rationale, not an empirical result; the same applies to the claim that cross-dimensional intersections are easier to find than in parallel sets. These comparative statements are central to the paper's motivation. If the evaluation requested in the first major comment is added, it should test these comparisons directly; if not, the comparative language should be revised to 'may' throughout.

minor comments (4)

[Section 4] The complexity expression 'O(n(m + p) +pnnlgl)' is typographically garbled in the submission. It should be rendered as something like O(n(m+p) + p^n * n * log l), or the formula should be replaced with a clearly typeset derivation.
[Section 3 (Design)] The hue assignment is under-specified: the text states hue varies linearly from 0.0 to 0.4 and from 0.6 to 0.9, but does not state how hue values are assigned to individual dimensions, especially when the number of dimensions is odd or even, or how the unused last side of an even-dimensional polygon is treated in the color mapping.
[Section 3 (Design)] The paper uses lexicographic value ordering for all dimensions, which imposes an arbitrary ranking on nominal categorical values. The authors should acknowledge that this ordering may bias interpretation or state that users can redefine value order.
[Section 4, Algorithm 1] The handling of the 'even number of dimensions' case (keeping the number of sides odd and leaving one side unused) is described in prose but not reflected in Algorithm 1 or in a precise initialization rule. Please make the mapping from dimensions to polygon sides explicit in the pseudocode or in an accompanying parameter table.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Hi-D maps layout algorithm is a self-contained construction, and the only self-citation is a non-load-bearing implementation detail.

full rationale

The paper does not derive a prediction from fitted parameters or from a claimed uniqueness theorem. Its central contribution is a space-partitioning algorithm: data are recursively partitioned by categorical values, and each polygon is split by binary search along a fixed-direction line so that child polygon areas match the partition proportions. This is a constructive layout method, not a derivation whose output is equivalent to its input by definition. The claims that Hi-D maps 'retain the spatial efficiency of treemaps and dimensional clarity of sunburst visualizations' (Section 2) are design assertions supported by visual comparison, not by a chain of reductions; the absence of a user study is an evaluation gap, not circularity. The only self-citation is reference [5] for the agent-based marble animation, a peripheral implementation detail that does not support the paper's central claims. The paper's stated limitations, including failure near 30 dimensions and the deferral of user studies and empirical comparisons to future work, further confirm that the effectiveness claims are hypotheses awaiting testing rather than conclusions forced by construction. No load-bearing step in the manuscript reduces to its own inputs or to a self-citation, so the appropriate circularity score is 0.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The paper does not introduce new physical entities or fitted constants. Its free parameters are user interface choices such as color ranges and marble count. Its load-bearing assumptions are perceptual: that area, angle, hue, and thickness cues are read accurately by viewers, and that dimension reordering mitigates ordering bias. These assumptions are stated explicitly but not validated by any empirical evaluation.

free parameters (3)

cut order and cut direction assignment = chosen by the user
The mapping of dimensions to polygon sides is user-controlled and ordered, which affects the visual result, though the paper notes leaf areas are invariant.
hue range and lightness range = hue 0.0-0.4 and 0.6-0.9, lightness 0.4-0.7, saturation 0.25-1.0
These are hand-chosen perceptual constants that affect readability.
number of marbles = 100
The marble count is a fixed design choice used for counting percentages.

assumptions (3)

domain assumption Leaf polygon area proportional to data subset size conveys subset size accurately to viewers.
The paper's design relies on area perception being effective, but it provides no user study, and the conclusion acknowledges perceptual limits.
domain assumption Angle of cut direction plus hue lets viewers identify dimensions without level counting.
This is the core design claim in Section 3 and the comparison in Section 3.2, but it is not empirically tested.
standard math Regular polygon cuts with parallel lines remain numerically stable and visually interpretable.
The geometry is straightforward, but the paper does not discuss numerical edge cases such as very small polygons.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Hi-d maps: An interactive visualization technique for multi-dimensional categorical data." pith.science (2026). https://pith.science/paper/WZSTNP3P

@misc{pith2026250707890,
  author       = {Pith},
  title        = {Pith review of: Hi-d maps: An interactive visualization technique for multi-dimensional categorical data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WZSTNP3P}},
  note         = {Machine review of arXiv:2507.07890}
}

read the original abstract

In this paper, we present Hi-D maps, a novel method for the visualization of multi-dimensional categorical data. Our work addresses the scarcity of techniques for visualizing a large number of data-dimensions in an effective and space-efficient manner. We have mapped the full data-space onto a 2D regular polygonal region. The polygon is cut hierarchically with lines parallel to a user-controlled, ordered sequence of sides, each representing a dimension. We have used multiple visual cues such as orientation, thickness, color, countable glyphs, and text to depict cross-dimensional information. We have added interactivity and hierarchical browsing to facilitate flexible exploration of the display: small areas can be scrutinized for details. Thus, our method is also easily extendable to visualize hierarchical information. Our glyph animations add an engaging aesthetic during interaction. Like many visualizations, Hi-D maps become less effective when a large number of dimensions stresses perceptual limits, but Hi-D maps may add clarity before those limits are reached.

Figures

Figures reproduced from arXiv: 2507.07890 by the authors.

**Figure 1.** (a) Our Hi-D maps, (b) treemaps [17], (c) sunburst [23], and (d) parallel sets [4] visualizations of the same data. ABSTRACT In this paper, we present Hi-D maps, a novel method for the visualization of multi-dimensional categorical data. Our work addresses the scarcity of techniques for visualizing a large number of datadimensions in an effective and space-efficient manner. We have mapped the full data-space onto … view at source ↗

**Figure 2.** Hi-D Maps: multiple dimensions mapped to a regular [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** (a) The dimensions can be re-ordered by dragging one [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 5.** Figure 5: The check-boxes at the bottom supports partial display. [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

[1]

D. F. Andrews. Plots of high-dimensional data. Biometrics, pp. 125– 136, 1972

work page 1972
[2]

Ankerst, D

M. Ankerst, D. A. Keim, and H.-P. Kriegel. Circle segments: A technique for visually exploring large multidimensional data sets. In Visualization, 1996

work page 1996
[3]

J. Beddow. Shape coding of multidimensional data on a microcomputer display. In Proceedings of the 1st Conference on Visualization’90, pp. 238–246. IEEE Computer Society Press, 1990

work page 1990
[4]

Bendix, R

F. Bendix, R. Kosara, and H. Hauser. Parallel sets: visual analysis of categorical data. In IEEE Symposium on Information Visualization,

work page
[5]

J. R. Benson, D. Crist, P. Lafleur, and B. Watson. Agentbased visual- ization of streaming text. In I7 Information Visualization Conference, 2008

work page 2008
[6]

W. W.-Y . Chan. A survey on multivariate data visualization.Depart- ment of Computer Science and Engineering. Hong Kong University of Science and Technology, 8(6):1–29, 2006

work page 2006
[7]

S. Chen, S. Chen, L. Lin, X. Yuan, J. Liang, and X. Zhang. E-map: A visual analytics approach for exploring significant event evolutions in social media. In 2017 IEEE Conference on Visual Analytics Science and Technology (VAST), pp. 36–47. IEEE, 2017

work page 2017
[8]

S. Chen, S. Chen, Z. Wang, J. Liang, X. Yuan, N. Cao, and Y . Wu. D-map: Visual analysis of ego-centric information diffusion patterns in social media. In 2016 IEEE Conference on Visual Analytics Science and Technology (VAST), pp. 41–50. IEEE, 2016

work page 2016

Show all 25 references

[9]

Cheng and K

S. Cheng and K. Mueller. The data context map: Fusing data and attributes into a unified display. IEEE transactions on visualization and computer graphics, 22(1):121–130, 2015

2015
[10]

Cheng, W

S. Cheng, W. Xu, and K. Mueller. Colormap nd: A data-driven ap- proach and tool for mapping multivariate data to color. IEEE trans- actions on visualization and computer graphics , 25(2):1361–1377, 2018

2018
[11]

Chernoff

H. Chernoff. The use of faces to represent points in k-dimensional space graphically. Journal of the American Statistical Association , 68(342):361–368, 1973

1973
[12]

Feiner and C

S. Feiner and C. Beshers. Worlds within worlds: metaphors for ex- ploring n-dimensional virtual worlds. In UIST, vol. 90, pp. 76–83, 1990

1990
[13]

Grinstein, R

G. Grinstein, R. Pickett, and M. G. Williams. Exvis: An exploratory visualization environment. In Proc. Graphics Interface, vol. 89, pp. 254–261, 1989

1989
[14]

Grinstein, M

G. Grinstein, M. Trutschl, and U. Cvek. High-dimensional visual- izations. In Proceedings of the Visual Data Mining Workshop, KDD, vol. 2, p. 120. Citeseer, 2001

2001
[15]

P. E. Hoffman. Table visualizations: a formal model and its applica- tions. University of Massachusetts Lowell, 2000

2000
[16]

Inselberg and B

A. Inselberg and B. Dimsdale. Parallel coordinates: a tool for visualiz- ing multi-dimensional geometry. In Proceedings of the 1st conference on Visualization’90, pp. 361–378. IEEE Computer Society Press, 1990

1990
[17]

Johnson and B

B. Johnson and B. Shneiderman. Tree-maps: A space-filling approach to the visualization of hierarchical information structures. IEEE, 1991

1991
[18]

Kandogan

E. Kandogan. Star coordinates: A multi-dimensional visualization technique with uniform treatment of dimensions. In Proceedings of the IEEE Information Visualization Symposium, vol. 650, p. 22. Citeseer, 2000

2000
[19]

D. A. Keim. Visual techniques for exploring databases. In Knowledge Discovery in Databases (KDD’97), 1997

1997
[20]

D. A. Keim, M. Ankerst, and H.-P. Kriegel. Recursive pattern: A technique for visualizing very large amounts of data. In Proceedings of the 6th Conference on Visualization’95 , p. 279. IEEE Computer Society, 1995

1995
[21]

D. A. Keim, M. C. Hao, J. Ladisch, M. Hsu, and U. Dayal. Pixel bar charts: A new technique for visualizing large multi-attribute data sets without aggregation. In IEEE Symposium on Information Visualization: INFOVIS 2001, pp. 113–120, 2001

2001
[22]

Rao and S

R. Rao and S. K. Card. The table lens: merging graphical and symbolic representations in an interactive focus+ context visualization for tabular information. In Proceedings of the SIGCHI conference on Human factors in computing systems, pp. 318–322. Citeseer, 1994

1994
[23]

Stasko and E

J. Stasko and E. Zhang. Focus+ context display and navigation tech- niques for enhancing radial, space-filling hierarchy visualizations. In IEEE Symposium on Information Visualization 2000. INFOVIS 2000. Proceedings, pp. 57–65. IEEE, 2000

2000
[24]

W. Wright. Information animation applications in the capital markets. In Proceedings of Visualization 1995 Conference, pp. 19–25. IEEE, 1995

1995
[2005]

INFOVIS 2005., pp. 133–140. IEEE, 2005

2005

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

D. F. Andrews. Plots of high-dimensional data. Biometrics, pp. 125– 136, 1972

work page 1972

[2] [2]

Ankerst, D

M. Ankerst, D. A. Keim, and H.-P. Kriegel. Circle segments: A technique for visually exploring large multidimensional data sets. In Visualization, 1996

work page 1996

[3] [3]

J. Beddow. Shape coding of multidimensional data on a microcomputer display. In Proceedings of the 1st Conference on Visualization’90, pp. 238–246. IEEE Computer Society Press, 1990

work page 1990

[4] [4]

Bendix, R

F. Bendix, R. Kosara, and H. Hauser. Parallel sets: visual analysis of categorical data. In IEEE Symposium on Information Visualization,

work page

[5] [5]

J. R. Benson, D. Crist, P. Lafleur, and B. Watson. Agentbased visual- ization of streaming text. In I7 Information Visualization Conference, 2008

work page 2008

[6] [6]

W. W.-Y . Chan. A survey on multivariate data visualization.Depart- ment of Computer Science and Engineering. Hong Kong University of Science and Technology, 8(6):1–29, 2006

work page 2006

[7] [7]

S. Chen, S. Chen, L. Lin, X. Yuan, J. Liang, and X. Zhang. E-map: A visual analytics approach for exploring significant event evolutions in social media. In 2017 IEEE Conference on Visual Analytics Science and Technology (VAST), pp. 36–47. IEEE, 2017

work page 2017

[8] [8]

S. Chen, S. Chen, Z. Wang, J. Liang, X. Yuan, N. Cao, and Y . Wu. D-map: Visual analysis of ego-centric information diffusion patterns in social media. In 2016 IEEE Conference on Visual Analytics Science and Technology (VAST), pp. 41–50. IEEE, 2016

work page 2016