Comparing graphs to identify similarities is a fundamental task in visual analytics of graph data. To support this, visual analytics systems frequently employ quantitative computational measures to provide automated guidance. However, it remains unclear how well these measures align with subjective human visual perception, thereby offering recommendations that conflict with analysts’ intuitive judgments, potentially leading to confusion rather than reducing cognitive load. Multimodal Large Language Models (MLLMs), capable of visually interpreting graphs and explaining their reasoning in natural language, have emerged as a potential alternative to address this challenge. This paper bridges the gap between human and machine assessment of graph similarity through 3 interconnected experiments using a dataset of 1881 node-link diagrams. Experiment 1 collects relative similarity judgments and rationales from 32 human participants, revealing consensus on graph similarity while prioritizing global shapes and edge densities over exact topological details. Experiment 2 benchmarks 16 computational measures against these human judgments, identifying Portrait divergence as the best-performing metric, though with only moderate alignment. Experiment 3 evaluates the potential of 3 state-of-the-art MLLMs (GPT-5, Gemini 2.5 Pro, Claude Sonnet 4.5) as perceptual proxies. The results demonstrate that MLLMs, particularly GPT-5, significantly outperform traditional measures in aligning with human graph similarity perception and provide interpretable rationales for their decisions, whereas Claude Sonnet 4.5 shows the best computational efficiency. Our findings suggest that MLLMs hold significant promise not only as effective, explainable proxies for human perception but also as intelligent guides that can uncover subtle nuances that might be overlooked by human analysts in visual analytics systems.
FarahaniRZMiandoabchiESzetoWY, et al. A review of urban transportation network design problems. Eur J Oper Res2013; 229(2): 281–302. https://doi.org/10.1016/j.ejor.2013.01.001
3.
GuzmanJDDeckroRFRobbinsMJ, et al. An analytical comparison of social network measures. IEEE Trans Comput Soc Syst2014; 1(1): 35–45. https://doi.org/10.1109/TCSS.2014.2307451
MoodyJMcFarlandDBender-demollS.Dynamic network visualization. Am J Sociol2005; 110(4): 1206–1241. https://doi.org/10.1086/421509
6.
BeckFBurchMDiehlS, et al. A taxonomy and survey of dynamic graph visualization. Comput Graph Forum2017; 36(1): 133–159. https://doi.org/10.1111/cgf.12791
7.
KimJLeeHNguyenDM, et al. DG comics: semi-automatically authoring graph comics for dynamic graphs. IEEE Trans Vis Comput Graph2025; 31(1): 973–983. https://doi.org/10.1109/TVCG.2024.3456340
JungSShinDJeonH, et al. MoNetExplorer: a visual analytics system for analyzing dynamic networks with temporal network motifs. IEEE Trans Vis Comput Graph2024; 30(10): 6725–6739. https://doi.org/10.1109/TVCG.2023.3337396
10.
JeonHQuadriGJLeeH, et al. Clams: a cluster ambiguity measure for estimating perceptual variability in visual clustering. IEEE Trans Vis Comput Graph2023; PP(1): 770–780. https://doi.org/10.1109/TVCG.2023.3327201
11.
TatuABakPBertiniE, et al. Visual quality metrics and human perception: an initial study on 2D projections of large multidimensional data. In: Proceedings of the international conference on advanced visual interfaces, Roma, Italy, 2010, pp. 49–56. ACM. ISBN 978-1-4503-0076-6. https://doi.org/10.1145/1842993.1843002.
12.
AlbuquerqueGEisemannMMagnorM.Perception-based visual quality measures. In: 2011 IEEE conference on visual analytics science and technology (VAST), pp. 13–20. https://doi.org/10.1109/VAST.2011.6102437.
13.
JardineNOndovBDElmqvistN, et al. The perceptual proxies of visual comparison. IEEE Trans Vis Comput Graph2020; 26(1): 1012–1021. https://doi.org/10.1109/TVCG.2019.2934786
QuadriGJRosenP.A survey of perception-based visualization studies by task. IEEE Trans Vis Comput Graph2022; 28(12): 5026–5048. https://doi.org/10.1109/TVCG.2021.3098240
16.
YoghourdjianVArchambaultDDiehlS, et al. Exploring the limits of complexity: a survey of empirical studies on graph visualisation. Vis Inform2018; 2(4): 264–282. https://doi.org/10.1016/j.visinf.2018.12.006
17.
Di BartolomeoSCrnovrsaninTSaffoD, et al. Evaluating graph layout algorithms: a systematic review of methods and best practices. Comput Graph Forum2024; 43: e15073. https://doi.org/10.1111/cgf.15073
18.
HermanIMelanconGMarshallMS.Graph visualization and navigation in information visualization: a survey. IEEE Trans Vis Comput Graph2000; 6(1): 24–43. https://doi.org/10.1109/2945.841119
19.
ArchambaultDPurchaseHPinaudB.Animation, small multiples, and the effect of mental map preservation in dynamic graphs. IEEE Trans Vis Comput Graph2011; 17(4): 539–552. https://doi.org/10.1109/TVCG.2010.78
Weller-FahyDJBorghettiBJSodemannAA.A survey of distance and similarity measures used within network intrusion anomaly detection. IEEE Commun Surv Tutor2015; 17(1): 70–91. https://doi.org/10.1109/COMST.2014.2336610
22.
SoundarajanSEliassi-RadTGallagherB.A guide to selecting a network similarity method. In: 2014 SIAM international conference on data mining, pp. 1037–1045. https://doi.org/10.1137/1.9781611973440.118.
JeonHLeeHKuoYH, et al. Unveiling high-dimensional backstage: a survey for reliable visual analytics with dimensionality reduction. In: Proceedings of the 2025 CHI conference on human factors in computing systems, CHI ’25, New York, USA, pp. 1–24. NY: Association for Computing Machinery. ISBN 9798400713941. https://doi.org/10.1145/3706598.3713551.
25.
ShneidermanB.Human-centered AI. Oxford University Press, 2022.
26.
KimNWAhnYMyersG, et al. How good is ChatGPT in giving advice on your visualization design?ACM Trans Comput-Hum Interact2025; 32(5): 1–33. https://doi.org/10.1145/3745768
27.
ChenNZhangYXuJ, et al. Viseval: a benchmark for data visualization in the era of large language models. IEEE Trans Vis Comput Graph2025; 31: 1301–1311. https://doi.org/10.1109/TVCG.2024.3456320
28.
ChoiJLeeJJoJ.Bavisitter: integrating design guidelines into large language models for visualization authoring. In: 2024 IEEE visualization and visual analytics (VIS), pp. 121–125. IEEE. https://doi.org/10.1109/VIS55277.2024.00032.
29.
RahmanzadehgerviPBoltonLTaesiriMR, et al. Vision language models are blind. In: Proceedings of the Asian Conference on Computer Vision, 2024, pp. 18–34. https://doi.org/10.1007/978-981-96-0917-8_17.
30.
BartolomeoSDSeveriGSchetingerV, et al. Ask and you shall receive (a graph drawing): testing ChatGPT’s potential to apply graph layout algorithms. arXiv preprint 2023. https://doi.org/10.48550/arXiv.2303.08819.2303.08819.
SchetingerVDi BartolomeoSEl-AssadyM, et al. Doom or deliciousness: challenges and opportunities for visualization in the age of generative models. Comput Graph Forum2023; 42(3): 423–435. https://doi.org/10.1111/cgf.14841
34.
HuangWEadesPHongSH.Measuring effectiveness of graph visualizations: a cognitive load perspective. Inf Vis2009; 8(3): 139–152. https://doi.org/10.1057/ivs.2009.10
35.
KypridemouEZitoMBertaminiM.The effect of graph layout on the perception of graph properties. EuroVis (Short Papers)2020. https://doi.org/10.2312/evs.20201039
36.
AbbasMMAupetitMSedlmairM, et al. ClustMe: a visual quality measure for ranking monochrome scatterplots based on cluster patterns. Comput Graph Forum2019; 38(3): 225–236. https://doi.org/10.1111/cgf.13684
37.
GogolouATsandilasTPalpanasT, et al. Comparing similarity perception in time series visualizations. IEEE Trans Vis Comput Graph2018; 25(1): 523–533. https://doi.org/10.1109/TVCG.2018.2865077
38.
Emmert-StreibFDehmerMShiY.Fifty years of graph matching, network alignment and network comparison. Inf Sci2016; 346-347: 180–197. https://doi.org/10.1016/j.ins.2016.01.074
39.
DonnatCHolmesS.Tracking network dynamics: a survey using graph distances. Ann Appl Stat2018; 12(2): 971.
KoutraDVogelsteinJTFaloutsosC. Deltacon: a principled massive-graph similarity function. In: Proceedings of the 2013 SIAM international conference on data mining, SIAM, pp. 162–170. https://doi.org/10.1137/1.9781611972832.18.
KerracherNKennedyJChalmersK.A task taxonomy for temporal graph visualisation. IEEE Trans Vis Comput Graph2015; 21(10): 1160–1172. https://doi.org/10.1109/TVCG.2015.2424889
49.
ZhouHLaiPSunZ, et al. AdaMotif: graph simplification via adaptive motif design. IEEE Trans Vis Comput Graph2025; 31: 688–698. https://doi.org/10.1109/TVCG.2024.3456321
50.
CrnovrsaninTShilpikaSChandrasegaranS, et al. Staged animation strategies for online dynamic networks. IEEE Trans Vis Comput Graph2021; 27(2): 539–549. https://doi.org/10.1109/TVCG.2020.3030385
51.
AmarREaganJStaskoJ.Low-level components of analytic activity in information visualization. In: IEEE symposium on information visualization, INFOVIS 2005, pp. 111–117. IEEE. https://doi.org/10.1109/INFVIS.2005.1532136.
52.
GhoniemMFeketeJDCastagliolaP. A comparison of the readability of graphs using node-link and matrix-based representations. In: IEEE symposium on information visualization, 2004, pp. 17–24. IEEE. https://doi.org/10.1109/INFVIS.2004.1.
53.
GuoHHuangJLaidlawDH.Representing uncertainty in graph edges: an evaluation of paired visual variables. IEEE Trans Vis Comput Graph2015; 21(10): 1173–1186. https://doi.org/10.1109/TVCG.2015.2424872
54.
ChangCBachBDwyerT, et al. Evaluating perceptually complementary views for network exploration tasks. In: Proceedings of the 2017 CHI conference on human factors in computing systems, pp. 1397–1407. https://doi.org/10.1145/3025453.3026024.
55.
BurchMHuangWWakefieldM, et al. The state of the art in empirical user evaluation of graph visualizations. IEEE Access2021; 9: 4173–4198. https://doi.org/10.1109/ACCESS.2020.3047616
56.
von LandesbergerT.Investigating graph similarity perception: a preliminary study and methodological challenges. In: International conference on information visualization theory and applications, Vol. 4, 2017, pp. 241–250. scitepress.org; https://doi.org/10.5220/0006137202410250.
57.
MooneyGJPurchaseHCWybrowM, et al. The multi-dimensional landscape of graph drawing metrics. In: 2024 IEEE 17th Pacific visualization conference (PacificVis), pp. 122–131. IEEE. https://doi.org/10.1109/PacificVis60374.2024.00022.
LeeSChangMParkS, et al. Assessing graphical perception of image embedding models using channel effectiveness. In: 2024 IEEE visualization and visual analytics (VIS), pp. 226–230. IEEE.
60.
QuadriGJRosenP.Modeling the influence of visual density on cluster perception in scatterplots using topology. IEEE Trans Vis Comput Graph2021; 27(2): 1829–1839. https://doi.org/10.1109/TVCG.2020.3030365
61.
XiaJZhangYSongJ, et al. Revisiting dimensionality reduction techniques for visual cluster analysis: an empirical study. IEEE Trans Vis Comput Graph2022; 28(1): 529–539. https://doi.org/10.1109/TVCG.2021.3114694
62.
BaeSSCaveKGorgC, et al. Bridging network science and vision science: mapping perceptual mechanisms to network visualization tasks. IEEE Trans Vis Comput Graph2025; 31: 7137–7152. https://doi.org/10.1109/TVCG.2025.3541571
BasoleRCMajorTBasoleRC, et al. Generative AI for visualization: opportunities and challenges. IEEE Comput Graph Appl2024; 44(2): 55–64. https://doi.org/10.1109/MCG.2024.3362168
ParkSSongYLeeS, et al. Leveraging multimodal LLM for inspirational user interface search. In: Proceedings of the 2025 CHI conference on human factors in computing systems, CHI ’25, New York, USA, pp. 1–22. NY: Association for Computing Machinery. ISBN 9798400713941. https://doi.org/10.1145/3706598.3714213.
67.
BendeckAStaskoJ.An empirical evaluation of the GPT-4 multimodal language model on visualization literacy tasks. IEEE Trans Vis Comput Graph2025; 31: 1105–1115. https://doi.org/10.1109/TVCG.2024.3456155
68.
ChoeKLeeCLeeS, et al. Enhancing data literacy on-demand: LLMs as guides for novices in chart interpretation. IEEE Trans Vis Comput Graph2025; 31(9): 4712–4727. https://doi.org/10.1109/TVCG.2024.3413195
FilipovVCenedaDArchambaultD, et al. Timelighting: guided exploration of 2D temporal network projections. IEEE Trans Vis Comput Graph2024; 31(3): 1932–1944. https://doi.org/10.1109/TVCG.2024.3514858
TsitsulinAMottinDKarrasP, et al. NetLSD: hearing the shape of a graph. In: Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, ACM, 2018, pp. 2347–2356. https://doi.org/10.1145/3219819.3219991.
73.
ErdősPRényiA.On the existence of a factor of degree one of a connected random graph. Acta Math Acad Sci Hung1966; 17(3-4): 359–368. https://doi.org/10.1007/BF01894879
DoğrusözUMaddenBMaddenP.Circular layout in the graph layout toolkit. In: NorthS (ed.) International symposium on graph drawing. Springer, 1997, pp. 92–100.
81.
McInnesLHealyJMelvilleJ.Umap: uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:180203426. 2018. https://doi.org/10.48550/arXiv.1802.03426.
PandeyAVKrauseJFelixC, et al. Towards understanding human similarity perception in the analysis of large sets of scatter plots. In: Proceedings of the 2016 CHI conference on human factors in computing systems, ACM, pp. 3659–3669.
LuMZengXLanirJ, et al. Sticky links: encoding quantitative data of graph edges. IEEE Trans Vis Comput Graph2024; 30(6): 2968–2980. https://doi.org/10.1109/TVCG.2024.3388562
89.
NobreCZhuKMörthE, et al. Reading between the pixels: investigating the barriers to visualization literacy. In: Proceedings of the 2024 CHI conference on human factors in computing systems, pp. 1–17. https://doi.org/10.1145/3613904.3642760.
90.
XiongCVan WeeldenLFranconeriS. The curse of knowledge in visual data communication. IEEE Trans Vis Comput Graph2020; 26(10): 3051–3062. https://doi.org/10.1109/TVCG.2019.2917689
91.
BallwegKPohlMWallnerG, et al. Visual similarity perception of directed acyclic graphs: a study on influencing factors and similarity judgment strategies. J Graph Algorithms Appl2018; 22(3): 519–553. https://doi.org/10.7155/jgaa.00467
92.
BridgemanSTamassiaR.A user study in similarity measures for graph drawing. J Graph Algorithms Appl2002; 6: 225–254. https://doi.org/10.7155/jgaa.00053
93.
TsengCQuadriGJWangZ, et al. Measuring categorical perception in color-coded scatterplots. In: Proceedings of the 2023 CHI conference on human factors in computing systems, pp. 1–14. https://doi.org/10.1145/3544548.3581416.
94.
BerlingerioMKoutraDEliassi-RadT, et al. Netsimile: a scalable approach to size-independent network similarity. arXiv preprint arXiv:12092684, 2012. https://doi.org/10.48550/arXiv.1209.2684.
RozemberczkiBSarkarR.Characteristic functions on graphs: birds of a feather, from statistical descriptors to parametric models. In: 29th ACM international conference on information & knowledge management, 2020, pp. 1325–1334. https://doi.org/10.1145/3340531.3411866.
HeimannM.Regal: representation learning-based graph alignment. In: ACM international conference on information and knowledge management, 2018, pp. 117–126. https://doi.org/10.1145/3269206.3271788.
100.
HermannsJ.Grasp: graph alignment through spectral signatures. In: Web and big data: 5th international joint conference, 2021, pp. 44–52. Springer. https://doi.org/10.1007/978-3-030-85896-4_4.
101.
BorgwardtKMKriegelHP.Shortest-path kernels on graphs. In: Fifth IEEE international conference on data mining (ICDM’05), p.8. IEEE. https://doi.org/10.1109/ICDM.2005.132.
SinghRXuJBergerB.Global alignment of multiple protein interaction networks with application to functional orthology detection. Proc Natl Acad Sci2008; 105(35): 12763–12768. https://doi.org/10.1073/pnas.0806627105
105.
LandisJRKochGG.An application of hierarchical kappa-type statistics in the assessment of majority agreement among multiple observers. Biometrics1977; 33: 363–374. https://doi.org/10.2307/2529786
GhoshDSchmidtLSuY, et al. Why are visually-grounded language models bad at image classification?Adv Neural Inf Process Syst2024; 37: 51727–51753. https://doi.org/10.52202/079017-1639
