Node Labeling in Line Diagrams of Ordered Sets

Gerd Stumme; Marcel N\"ohre

REVIEW 4 major objections 4 minor 12 references

Node Labeling in Line Diagrams of Ordered Sets

T0 review · 4 major / 4 minor · reviewed 2026-07-31 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a two-phase algorithm can place every label on a line diagram of an ordered set without overlaps, including labels that cannot sit beside their node.

desk verdict The paper fills a real gap with an honest, code-backed pipeline, but the headline no-intersection guarantee is not actually delivered, and the evaluation never measures how often it is violated. read the letter →

arxiv 2607.23283 v1 pith:Y5E7OZHF submitted 2026-07-25 cs.CG

Marcel Nöhre , Gerd Stumme This is my paper · ORCID

classification cs.CG MSC 68U0506A0705C6268R10

keywords nodelabelinglinediagramsorderedsetsformalconceptanalysislatticesoverflowlabelsforce-basedrefinementfixed-positionmodel

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

This paper is trying to establish that label placement for line diagrams of ordered sets can be fully automated with publication quality. Its two-phase algorithm first filters an eight-position set of label candidates using geometric and readability criteria; labels that still cannot sit next to their node become overflow labels placed in the free space of interior faces or the unbounded exterior and connected by straight binding lines. Because the edges of such diagrams are nearly vertical, the algorithm can exploit this structure instead of treating the diagram as an arbitrary graph, and the cost models minimize the distance between each node and its label. If the claims hold, authors of order-theory and Formal Concept Analysis figures no longer need to position labels by hand.

What carries the argument

The key mechanism is the two-phase pipeline centered on the fixed-position model and the overflow label construct. The fixed-position model generates eight anchored candidates per label; line-diagram-specific filters (concave hull, node proximity, edge intersection, angular sector) prune them. Remaining unplaced labels become overflow labels: first placed inside faces of the planarized graph via a visibility mask, then on a discrete grid ribbon around the drawing, with each label linked to its node by a straight binding line. An assignment algorithm with iterative cost penalties produces the initial placement, and force-based refinement (repulsion, tangential distribution, spring, steering)

What would settle it

Run the pipeline on a deliberately dense line diagram, such as a concept lattice with many large labels forced into a small bounding box, and inspect the output for any label bounding box that overlaps an edge or any binding line that crosses an unrelated node or label. The paper's own evaluation acknowledges that dense graphs produce more frequent intersections; exhibiting a concrete instance where such an intersection occurs during normal parameter settings would contradict the stated hard-constraint claim.

Watch

Extended reading notes

Core claim

The central claim is that the near-vertical edge structure of line diagrams makes automated label placement tractable and visually clean. The algorithm decomposes the problem into a fixed-position phase (eight candidate positions per label, filtered by concave hull, node proximity, edge intersection, and angular sector tests, then resolved by a conflict-resolution heuristic with readability tie-breakers) and an overflow phase (bounded overflow labels in planarized graph faces, then unbounded overflow labels on a grid in the exterior, assigned via an assignment solver and refined by force-based steps). The same pipeline, with a type-based prefilter and half-plane forces, handles the Formal Co

Load-bearing premise

The load-bearing premise is that a valid, non-overlapping placement always exists for every label within the algorithm's search spaces (the eight candidate positions, the graph faces, and the exterior ribbon), since validity is treated as a hard constraint rather than proven; the paper itself notes that dense graphs lead to longer binding lines and more frequent intersections.

Editorial extensions

If this is right

If the algorithm works as described, line diagrams of ordered sets can be labeled automatically end-to-end, with no manual repositioning and no label scaling or redrawing of the diagram.
Labels that do not fit beside their node remain readable: they are placed in the nearest face or the exterior ribbon and connected by straight binding lines, preserving the node-label association.
The variant for Formal Concept Analysis shows that the same machinery handles dual labeling conventions — attributes above, objects below — with only a pre-filter and a half-plane force.
The runtime analysis identifies concrete default parameters (grid step, search radius, assignment iterations, force iterations) that balance quality and speed for graphs up to roughly 30 nodes, making the approach usable in static publication workflows.
Scalability is bounded by available whitespace, since label count and size compete for the same finite area; dense diagrams will push labels outward and may show more intersections.

Reading between the lines

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

The near-vertical structure exploited here suggests the approach could generalize to other upward drawings, such as layered DAGs, where edges are mostly vertical, not just transitive reductions of ordered sets.
The cost trade-off between label proximity and obstacle clearance is presented as fixed by a weight; a natural extension would make this weight adaptive per region or let users specify readability priorities.
Because the algorithm preserves a clear node-label association even when binding lines cross, it could support interactive tools that let authors tweak overflow placements after automatic labeling, rather than requiring fully automatic perfection.
A testable prediction: on a family of increasingly dense lattices with fixed label sizes, the fraction of labels placed as overflow and the number of binder intersections should grow predictably with node density; measuring that curve would calibrate when alternative representations, such as nested diagrams, become necessary.

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

Dear colleague,

Short version: this is a legitimately useful paper for anyone who has to generate line diagrams of ordered sets or concept lattices automatically. It proposes a two-phase labeling pipeline — fixed-position candidates filtered by line-diagram-specific criteria, then overflow labels placed in faces or the exterior with straight binding lines, refined by forces — and it ships a public implementation. The FCA variant with intent labels above and extent labels below is a genuine extension, not a token addition. The authors know the labeling literature and adapt Wagner-Wolff and Kakoulis-Tollis rather than reinventing them.

The main soft spot is exactly what the stress-test note flags. Section 3 defines a valid placement by two hard constraints: no label may intersect any node, edge, or other label, and a binding line may not intersect any unrelated node or label. The abstract repeats that as the thing being studied. But Section 6 concedes that in dense graphs, 'prioritizing non-overlapping positions ... leads to longer binding lines and more frequent intersections.' That means the algorithm can output placements that violate the hard constraints, and the paper never reports how often. The evaluation measures average distances and runtimes but not the fraction of labels or binders that intersect an obstacle. Without that number, the central claim is unverified. This is not fatal — the paper honestly frames the trade-off and the figures look reasonable — but it needs to be quantified.

Other soft spots: the many weights and thresholds (wθ, w∂, wσ, wφ, τrep, τgap, τbind, ...) are all hand-set; the sensitivity analysis shows the output is fairly robust to them, but there's no principled tuning. There is also no comparison with a baseline — maybe a simple greedy or force-based labeling without the line-diagram-specific filters — so the actual benefit of each filter is unclear. The runtime analysis is fine, but the evaluation set (standard FCA graphs with up to 30 nodes) is small; the one dense example in Figure 18 is exactly where the method struggles.

On the plus side, the search-space definition for bounded overflow labels and the half-plane handling for FCA are thoughtful, and the code is public, which makes the paper reproducible. The citation pattern looks fine.

Verdict: this deserves a serious peer review. A careful referee should push for a quantitative violation report in the dense cases, a baseline comparison, and a clearer statement of when the hard constraints are actually guaranteed. I'd read a revised version.

Best,

Referee Report

4 major / 4 minor

Summary. The paper proposes a two-phase algorithm for labeling nodes in line diagrams of ordered sets, with a specialized variant for Formal Concept Analysis (FCA) that enforces intent labels above and extent labels below their nodes. Phase 1 generates eight-position label candidates and filters them using geometric, edge-intersection, and angular-sector criteria, then resolves conflicts with an adapted map-labeling heuristic. Phase 2 treats unplaced 'overflow' labels by first trying bounded placements in graph faces and then moving to an unbounded exterior region, using a grid of candidates, an iterative Hungarian assignment, and a force-based refinement step; binding lines connect overflow labels to their nodes. The stated goal is a placement where no label intersects any node, edge, or other label while minimizing label-to-node distances. The evaluation reports runtime and averaged quality metrics on small FCA graphs, with a qualitative discussion of dense cases.

Significance. If the algorithm achieved the claimed non-intersecting placement reliably, it would fill a genuine gap: automated publication-quality labeling for FCA and order-theory diagrams is underdeveloped, and the paper provides a concrete pipeline, source code, and thoughtful domain-specific adaptations (the FCA above/below conventions are a natural extension). The reproducible code and the attempt to ground parameter choices in a sensitivity analysis are also strengths. However, the central guarantee is not established: the paper provides no correctness proof, and its own Section 6 admits that the algorithm produces 'more frequent intersections' in dense graphs. The significance therefore depends on whether the claims are revised and the actual violation rate is measured; as written, the strongest selling point is unsupported.

major comments (4)

[Abstract; §3; §6] The abstract states the paper studies placements 'such that they do not intersect with any nodes, lines, or other labels,' and §3 defines a valid placement with two hard constraints. Section 6 explicitly concedes that in dense graphs (e.g., the convex-ordinal scale lattice, Figure 18) 'our approach prioritizes non-overlapping positions, which leads to longer binding lines and more frequent intersections.' The conclusion repeats this. This is a direct contradiction of the central claim. The manuscript must either weaken the abstract/problem statement to say the algorithm attempts to avoid intersections, or provide a correctness argument plus an empirical violation audit (e.g., fraction of labels/binders that intersect nodes, edges, or labels) demonstrating zero violations on the evaluated instances.
[§4.5–§4.7] The pipeline is heuristic and no feasibility guarantee is given. The bounded-overflow search space S(ℓ) may be empty, and there is no fallback stated if it is. Section 4.6's iterative Hungarian solver with a penalty matrix 'does not always converge on a singular, stable state,' so the solver keeps the lowest-cost configuration, which can still contain overlaps. The force-based refinement in §4.7 moves labels to reduce overlaps but does not enforce the hard constraints of §3. Without a proof or an empirical audit that the final output satisfies the constraints, the claim that the algorithm produces valid placements is unsupported.
[§6] The evaluation does not measure the central property. The three reported metrics are averages: average distance to drawing, mean minimum label–obstacle distance, and mean minimum binder–obstacle distance. A positive mean minimum distance does not rule out a small number of serious violations, and no violation counts, worst-case values, or percentiles are reported. The sensitivity analysis in Figure 17 only varies parameters and reports aggregate quality, so it cannot detect that some labels or binders intersect. The paper should report, at minimum, the fraction of labels and binders that violate each hard constraint, ideally stratified by graph density and label size.
[§5.3] The FCA variant is described in §5 as respecting the convention that intent labels are placed above and extent labels below their nodes. However, §5.3 introduces a directional penalty 'Chp' that is a soft preference and states it enables the Hungarian algorithm to 'bypass the directional rules in crowded areas.' This contradicts the strict semantic separation claimed in §5 and Figure 14. If the FCA version allows violations of the above/below convention, that must be stated explicitly and quantified; if it does not, the soft penalty should not be used in the default configuration.

minor comments (4)

[§4.5] The definition of S(ℓ) uses set subtraction and the symbol ⊖ without defining whether it is a Minkowski difference or ordinary set difference. The expression mixes candidate points λ and rectangles R(λ); a clearer formalization would help reproducibility.
[§4.3] The angular sector filter defines a set of prohibited anchor positions 'A̸∠' but the symbol is not explained and appears to be a LaTeX artifact. The restr(Q) mapping is also presented as a table without justification; a brief intuition would improve readability.
[General] Several typographical and formatting issues: 'intergrated' in §5.3, 'posessing' in the formal context section, and the affiliation string 'envel⌢pe/h⌢me' appears to be a LaTeX macro leak. These should be cleaned before publication.
[§6] The runtime analysis reports the grid construction as O(n·|P|·logn), but the dependence on label size and label–label conflict resolution is not characterized. A brief asymptotic breakdown of each phase would be useful.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper contributes a constructive labeling algorithm; its optimization objectives are the specification, and no fitted parameter is renamed as a prediction.

full rationale

The paper's contribution is a layout algorithm, not a derived scientific quantity: Section 4 defines candidate generation, geometric filters, Hungarian assignment, and force refinement, and Section 5 specializes these to FCA labels. There is no equation whose output is equal to an input by construction, and no parameter is fitted to data and then reported as a prediction. The quality metrics in Section 6 (average distance from unbounded labels to the drawing, mean minimum label-obstacle and binder-obstacle distances) overlap with terms in the cost function Cfix and the force model, so the evaluation is partly self-referential as evidence of visual quality; however, this is an evaluation limitation rather than circular reasoning, because those metrics are properties of the output layout, not re-labeled inputs. The paper's self-citations (e.g., [1], [8], [9], [20]) are used for background, drawing algorithms, datasets, or small-label content and are not load-bearing for the labeling claim. The Section 6 admission that dense graphs yield 'longer binding lines and more frequent intersections' is a correctness and scalability caveat that weakens the hard-constraint claim of Section 3, but an unmet guarantee is not a circular argument. No specific circular step can be exhibited, so the honest finding is no significant circularity.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central algorithm depends on a large set of hand-tuned weights and thresholds whose values are not stated; these are free parameters, but they are algorithmic tuning knobs rather than fitted scientific constants. The domain assumptions about vertical edges and face-based overflow are load-bearing.

free parameters (6)

Cost weights wθ, w∂, wσ, wϕ
Weights in the fixed cost Cfix for unbounded overflow labels; values not given in paper.
Penalty weights wλλ, wσλ, wσσ
Weights in the conflict penalty Π(λ_{i,j}) for Hungarian cost matrix updates.
Force parameters τrep, τgap, τbind, k, σ*, whp
Parameters in repulsion, tangential, spring, steering, and half-plane forces; values not given.
Padding δ and extended padding ε
Spatial padding around label bounding boxes.
Grid step size ρ and max search distance d_max = ρ=0.5, d_max=2.0
Defaults from sensitivity analysis in Section 6.
Hungarian top-k candidates and iteration count = top-50, 10 iterations
Defaults from Section 6.

assumptions (5)

domain assumption Line diagram edges are upward and 'more or less vertical', so fixed-position candidates and angular sector filters are appropriate.
Abstract and Section 1; if edges are not predominantly vertical, the angular filter may remove all valid positions.
domain assumption The planarization via dummy vertices at edge crossings yields faces that serve as valid label spaces; the boundary walk of the unbounded face defines the concave hull.
Section 2.2 and 4.5; labels are placed in faces of the planarized graph.
domain assumption Label ink is adequately represented by axis-aligned minimum bounding rectangles for intersection testing.
Section 2.3; conservative approximation, but may mark non-intersecting placements as invalid.
standard math The hybrid map-labeling algorithm of Wagner-Wolff and its rules (safe selection, mutual dependency, clique) are correct for finding maximal conflict-free subsets.
Section 4.4; cited from [27,26].
ad hoc to paper The angular sector filter's restr(Q) mapping correctly identifies free quadrants.
Section 4.3; heuristic restriction mapping specific to this paper, no proof of completeness.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Node Labeling in Line Diagrams of Ordered Sets." pith.science (2026). https://pith.science/paper/Y5E7OZHF

@misc{pith2026260723283,
  author       = {Pith},
  title        = {Pith review of: Node Labeling in Line Diagrams of Ordered Sets},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y5E7OZHF}},
  note         = {Machine review of arXiv:2607.23283}
}

read the original abstract

We propose a flexible, two-phase algorithm for labeling line diagrams of ordered sets, in which the nodes of direct neighbors in the order relation are connected by a straight, upward-pointing line. In contrast to the labeling of diagrams of arbitrary graphs, we benefit from the fact that all edges in line diagrams of ordered sets are more or less vertical. In this paper, we study the placement of all labels such that they do not intersect with any nodes, lines, or other labels while minimizing the distances between the nodes and their labels. Our approach starts by filtering the fixed-position model using line diagram-specific readability criteria. For labels that cannot be placed adjacent to their node (overflow labels), we exploit the free space in the graph's interior or the infinite space surrounding the drawing and link the labels with their respective node by straight binding lines that should not cross other nodes or labels if possible. To balance quality and runtime, we derive an initial placement of the overflow labels using a cost function over a sparse grid of candidates, followed by a force-based refinement step to fine-tune the layout. Furthermore, we demonstrate the flexibility of this approach by applying it to the visual constraints of line diagrams in the field of Formal Concept Analysis (FCA), where certain labels have to be placed above their node and others below. This special version of the algorithm shows that a pre-filtering in the first phase and minimal adjustments for the cost function and force-based model are sufficient to handle the dual labeling requirements of concept lattices.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 4 canonical work pages

[7]

doi:10.1145/276884.276923

Association for Computing Machinery. doi:10.1145/276884.276923. 17 Donald E. Knuth and Michael F. Plass. Breaking paragraphs into lines.Soft- ware: Practice and Experience, 11(11):1119–1184,

arXiv
[1955]

wiley.com/doi/abs/10.1002/nav.3800020109, arXiv:https://onlinelibrary.wiley.com/ doi/pdf/10.1002/nav.3800020109,doi:10.1002/nav.3800020109

URL: https://onlinelibrary. wiley.com/doi/abs/10.1002/nav.3800020109, arXiv:https://onlinelibrary.wiley.com/ doi/pdf/10.1002/nav.3800020109,doi:10.1002/nav.3800020109. 19 S.T. Leutenegger, M.A. Lopez, and J. Edgington. Str: a simple and efficient algorithm for r-tree packing. InProceedings 13th International Conference on Data Engineering, pages 497–506, ...

arXiv 1997
[1967]

Efficient processing of spatial joins using r-trees.SIGMOD Rec., 22(2):237–246, June 1993.doi:10.1145/170036.170075

3 Thomas Brinkhoff, Hans-Peter Kriegel, and Bernhard Seeger. Efficient processing of spatial joins using r-trees.SIGMOD Rec., 22(2):237–246, June 1993.doi:10.1145/170036.170075. 4 Jon Christensen, Joe Marks, and Stuart Shieber. An empirical study of algorithms for point- feature label placement.ACM Trans. Graph., 14(3):203–232, July 1995.doi:10.1145/21233...

arXiv 1993
[1981]

wiley.com/doi/abs/10.1002/spe.4380111102, arXiv:https://onlinelibrary.wiley.com/ doi/pdf/10.1002/spe.4380111102,doi:10.1002/spe.4380111102

URL: https://onlinelibrary. wiley.com/doi/abs/10.1002/spe.4380111102, arXiv:https://onlinelibrary.wiley.com/ doi/pdf/10.1002/spe.4380111102,doi:10.1002/spe.4380111102. 18 H. W. Kuhn. The hungarian method for the assignment problem.Naval Research Logistics Quarterly, 2(1-2):83–97,

work page doi:10.1002/spe.4380111102
[1991]

doi:10.1145/109648.109680

Association for Computing Machinery. doi:10.1145/109648.109680. 12 H. Freeman and R. Shapira. Determining the minimum-area encasing rectangle for an arbitrary closed curve.Commun. ACM, 18(7):409–413, July 1975.doi:10.1145/360881.360919. 13 Ralph Freese. Automated lattice drawing. In Peter Eklund, editor,Concept Lattices, pages 112–127, Berlin, Heidelberg,

arXiv 1975
[1999]

URL:https://www.sciencedirect.com/science/ article/pii/S092577219900005X,doi:10.1016/S0925-7721(99)00005-X. 26 F. Wagner, A. Wolff, V. Kapoor, and T. Strijk. Three rules suffice for good label placement. Algorithmica, 30(2):334–349, Jun 2001.doi:10.1007/s00453-001-0009-7. 27 Frank Wagner and Alexander Wolff. A combinatorial framework for map labeling. In ...

work page doi:10.1016/s0925-7721(99)00005-x 2001
[2008]

24 Kozo Sugiyama, Shojiro Tagawa, and Mitsuhiko Toda

Association for Computing Machinery.doi:10.1145/1377980.1377986. 24 Kozo Sugiyama, Shojiro Tagawa, and Mitsuhiko Toda. Methods for visual understanding of hierarchical system structures.IEEE Transactions on Systems, Man, and Cybernetics, 11(2):109–125, 1981.doi:10.1109/TSMC.1981.4308636. 25 Marc van Kreveld, Tycho Strijk, and Alexander Wolff. Point labeli...

arXiv 1981
[2014]

23 Thierry Stein and Xavier Décoret

doi:10.1109/PacificVis.2014.21. 23 Thierry Stein and Xavier Décoret. Dynamic label placement for improved interactive explora- tion. InProceedings of the 6th International Symposium on Non-Photorealistic Animation and Rendering, NPAR ’08, page 15–21, New York, NY, USA,

work page doi:10.1109/pacificvis.2014.21 2014

Show all 12 references

[2019]

9 Dominik Dürrschnabel, Tom Hanika, and Gerd Stumme

URL:http://arxiv.org/abs/1906.06208, arXiv:1906.06208. 9 Dominik Dürrschnabel, Tom Hanika, and Gerd Stumme. Dataset of user study for dimdraw. Zenodo, October

1906 arXiv
[2020]

10 Martin Fink, Jan-Henrik Haunert, André Schulz, Joachim Spoerhase, and Alexander Wolff

Dataset.doi:10.5281/zenodo.4075207. 10 Martin Fink, Jan-Henrik Haunert, André Schulz, Joachim Spoerhase, and Alexander Wolff. Algorithms for labeling focus regions.IEEE Transactions on Visualization and Computer Graphics, 18(12):2583–2592, 2012.doi:10.1109/TVCG.2012.193. 20 No...

2012 doi
[2024]

R-trees: a dynamic index structure for spatial searching.SIGMOD Rec., 14(2):47–57, June 1984.doi:10.1145/971697.602266

15 Antonin Guttman. R-trees: a dynamic index structure for spatial searching.SIGMOD Rec., 14(2):47–57, June 1984.doi:10.1145/971697.602266. 16 Konstantinos G. Kakoulis and Ioannis G. Tollis. A unified approach to labeling graphical features. InProceedings of the Fourteenth Ann...

1984
[2026]

21 Jack A

URL: https://arxiv.org/abs/2603.16366, arXiv: 2603.16366. 21 Jack A. Orenstein. Spatial query processing in an object-oriented database system.SIGMOD Rec., 15(2):326–336, June 1986.doi:10.1145/16856.16886. 22 Tim Pattison, Derek Weber, and Aaron Ceglar. Enhancing layout and in...

1986 arXiv

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.

[1] [7]

doi:10.1145/276884.276923

Association for Computing Machinery. doi:10.1145/276884.276923. 17 Donald E. Knuth and Michael F. Plass. Breaking paragraphs into lines.Soft- ware: Practice and Experience, 11(11):1119–1184,

arXiv

[2] [1955]

wiley.com/doi/abs/10.1002/nav.3800020109, arXiv:https://onlinelibrary.wiley.com/ doi/pdf/10.1002/nav.3800020109,doi:10.1002/nav.3800020109

URL: https://onlinelibrary. wiley.com/doi/abs/10.1002/nav.3800020109, arXiv:https://onlinelibrary.wiley.com/ doi/pdf/10.1002/nav.3800020109,doi:10.1002/nav.3800020109. 19 S.T. Leutenegger, M.A. Lopez, and J. Edgington. Str: a simple and efficient algorithm for r-tree packing. InProceedings 13th International Conference on Data Engineering, pages 497–506, ...

arXiv 1997

[3] [1967]

Efficient processing of spatial joins using r-trees.SIGMOD Rec., 22(2):237–246, June 1993.doi:10.1145/170036.170075

3 Thomas Brinkhoff, Hans-Peter Kriegel, and Bernhard Seeger. Efficient processing of spatial joins using r-trees.SIGMOD Rec., 22(2):237–246, June 1993.doi:10.1145/170036.170075. 4 Jon Christensen, Joe Marks, and Stuart Shieber. An empirical study of algorithms for point- feature label placement.ACM Trans. Graph., 14(3):203–232, July 1995.doi:10.1145/21233...

arXiv 1993

[4] [1981]

wiley.com/doi/abs/10.1002/spe.4380111102, arXiv:https://onlinelibrary.wiley.com/ doi/pdf/10.1002/spe.4380111102,doi:10.1002/spe.4380111102

URL: https://onlinelibrary. wiley.com/doi/abs/10.1002/spe.4380111102, arXiv:https://onlinelibrary.wiley.com/ doi/pdf/10.1002/spe.4380111102,doi:10.1002/spe.4380111102. 18 H. W. Kuhn. The hungarian method for the assignment problem.Naval Research Logistics Quarterly, 2(1-2):83–97,

work page doi:10.1002/spe.4380111102

[5] [1991]

doi:10.1145/109648.109680

Association for Computing Machinery. doi:10.1145/109648.109680. 12 H. Freeman and R. Shapira. Determining the minimum-area encasing rectangle for an arbitrary closed curve.Commun. ACM, 18(7):409–413, July 1975.doi:10.1145/360881.360919. 13 Ralph Freese. Automated lattice drawing. In Peter Eklund, editor,Concept Lattices, pages 112–127, Berlin, Heidelberg,

arXiv 1975

[6] [1999]

URL:https://www.sciencedirect.com/science/ article/pii/S092577219900005X,doi:10.1016/S0925-7721(99)00005-X. 26 F. Wagner, A. Wolff, V. Kapoor, and T. Strijk. Three rules suffice for good label placement. Algorithmica, 30(2):334–349, Jun 2001.doi:10.1007/s00453-001-0009-7. 27 Frank Wagner and Alexander Wolff. A combinatorial framework for map labeling. In ...

work page doi:10.1016/s0925-7721(99)00005-x 2001

[7] [2008]

24 Kozo Sugiyama, Shojiro Tagawa, and Mitsuhiko Toda

Association for Computing Machinery.doi:10.1145/1377980.1377986. 24 Kozo Sugiyama, Shojiro Tagawa, and Mitsuhiko Toda. Methods for visual understanding of hierarchical system structures.IEEE Transactions on Systems, Man, and Cybernetics, 11(2):109–125, 1981.doi:10.1109/TSMC.1981.4308636. 25 Marc van Kreveld, Tycho Strijk, and Alexander Wolff. Point labeli...

arXiv 1981

[8] [2014]

23 Thierry Stein and Xavier Décoret

doi:10.1109/PacificVis.2014.21. 23 Thierry Stein and Xavier Décoret. Dynamic label placement for improved interactive explora- tion. InProceedings of the 6th International Symposium on Non-Photorealistic Animation and Rendering, NPAR ’08, page 15–21, New York, NY, USA,

work page doi:10.1109/pacificvis.2014.21 2014