The word “visualization” currently covers at least two activities with entirely different goals.

One is drawing measured data so people can spot patterns — enterprise dashboards, monitoring panels, and business reports all fall here.

The other is presenting knowledge that already exists so people can build understanding — algorithm animations, human anatomy, mathematical functions, and cosmic structure fall here.

The first is data visualization. The second is knowledge visualization. This isn’t a distinction we invented: knowledge visualization has been a research area with its own literature since the 2000s (the framework proposed by Eppler & Burkhard; Chinese-language journals covered it as early as 2005).

The two are routinely discussed together, yet their constraints, evaluation criteria, and failure modes are almost entirely different. Conflating them has a predictable cost: applying one set of standards to the other, and satisfying neither.

1. The Core Differences

Dimension	Data Visualization	Knowledge Visualization
Object	Data: measured, dimensioned values	Knowledge: concepts, structures, mechanisms, processes
Source	Databases, APIs, real-time streams, surveys	Textbooks, domain knowledge, first principles
Truth constraint	Hard: every visual element must map exactly to a value	Soft: schematic, simplified, even exaggerated is acceptable
Primary goal	Insight — find patterns, anomalies, trends	Understanding — build a correct mental model
Typical question	“Why did conversion drop this month?”	“Why is quicksort O(n log n)?”
Purpose of interaction	Filter, drill down, compare	Experiment, predict, construct
Failure mode	Misleading — the graphic distorts the data	Misunderstanding — the motion is smooth but the mechanism is wrong
Evaluation criteria	Accuracy, information density, readability, reproducibility	Whether a correct mental model formed, and whether it transfers
What remains without the graphic	A table; almost no information is lost	A paragraph of description; substantial information is lost
May it be fictional?	No	Yes — schematic cells and simplified gears are both accepted
Time dimension	Usually a real time series	Usually “sequence of a process,” not real time

The “may it be fictional?” row separates the two most cleanly.

In data visualization, coloring a 529-case region and a 51,986-case region the same shade is an error. In knowledge visualization, using a schematic cell or exaggerating gear thickness so the meshing is visible is a legitimate teaching device.

This isn’t a difference in standards — it’s a difference in constraints arising from different goals.

2. Data Visualization: Data as the Object

The core question is “what does this data say?” By use case, it splits into several families:

① Monitoring and operations — Grafana, Kibana, Prometheus + Grafana. Characterized by dense time series, frequent refresh, and alert-driven use.

② Business intelligence — Tableau, Power BI, Alibaba Cloud Quick BI, Apache Superset, DataEase, FineBI. Characterized by metric governance, multidimensional drill-down, and permission control.

③ Data dashboards (大屏) — Alibaba Cloud DataV, FanRuan FineVis, Baidu Sugar, Tencent RayData. Built for briefings and site visits; emphasizes visual impact and full-screen layout.

④ Charting libraries (the layer beneath) — Apache ECharts, D3.js, Highcharts, Plotly, Chart.js, AntV G2. These are the tools used to build the three families above; they don’t produce content themselves.

⑤ Statistical and scientific plotting — matplotlib, ggplot2, seaborn, Vega-Lite. For analysts and researchers.

⑥ Geospatial — deck.gl, Mapbox GL, AntV L7. For data carrying coordinates.

⑦ Data journalism — The Pudding, Our World in Data, the Financial Times Visual Vocabulary, and in China NetEase Data Blog and The Paper’s data desk. For public communication.

⑧ Exploratory analysis — Observable, Jupyter, RStudio. For analysts computing and looking simultaneously.

Where the hard constraints show up:

• Axes must not be truncated (bar charts start at zero); otherwise ratios are visually inflated
• Classification thresholds must not be arbitrary; otherwise objects of wildly different magnitude get the same color
• Circles and bubbles must be encoded by area; encoding by radius squares the values
• 3D is usually harmful unless the data itself is three-dimensional; perspective distorts readings

These aren’t aesthetic preferences — they’re accuracy requirements. Authoritative references include Google’s official Visualization Traps documentation and the UK Office for National Statistics’ chart guidance.

3. Knowledge Visualization: Understanding as the Object

The core question is “how does this work?” By discipline:

① Computing and algorithms

• VisuAlgo — algorithm visualization covering sorting, graph theory, trees, geometry, and dynamic programming
• Data Structure Visualizations (USFCA) — David Galles’ classic collection; pure JavaScript, extremely lightweight
• Algorithm Visualizer — an open-source platform that generates visualizations from code
• Python Tutor — steps through code showing stack frames, heap objects, and references
• Red Blob Games — interactive explanations of A* pathfinding, hexagonal grids, and more
• Linggan Visual · Data Structure Visualization (ds.linggankeshi.com) — renders data structure operations as SVG animations, covering linear lists, strings, arrays, generalized lists, trees (including threaded binary trees, Huffman trees, red-black trees, B+ trees), graphs (four storage structures), union-find, searching, and seven internal sorts — 48 topics in total, with randomized code quizzes

② Mathematics

• GeoGebra — a dynamic environment for geometry, algebra, and calculus
• Desmos — the benchmark for interactive function graphing
• Mathigon — described as an “interactive mathematics textbook”
• Explained Visually — minimalist interactive treatments of eigenvectors, PCA, and Markov chains
• Seeing Theory — an interactive probability and statistics course (Chinese edition available)
• 3Blue1Brown / Manim — the ceiling of mathematical animation, on a purpose-built engine

③ Physics and chemistry

• PhET — interactive simulations from CU Boulder covering physics, chemistry, biology, math, and earth science (Chinese edition available)
• Paul Falstad’s applets — circuits, waves, quantum mechanics, and fluid simulations
• Molecular viewers — MolView, ChemDoodle and similar, turning formulas into rotatable 3D models

④ Biology and medicine

• Anatly (anatly.com) — interactive 3D human anatomy with muscle origins and insertions, nerve supply, actions, and quizzes
• Cell Forge (cellforge.org) — turns microscope photos into interactive 3D cell models, aimed at AP Bio, IB HL, A-Level, MCAT, and USMLE
• Protein structure viewers — 3D structure browsing over PDB data
• Brain atlases — such as the Allen Brain Atlas, turning neuroanatomical data into explorable maps

⑤ Astronomy and earth science

• Super Universe (super-universe.app) — a 3D map of the universe covering the Solar System, stars, exoplanets, the Milky Way, and the cosmic web
• NASA Eyes — official Solar System and mission visualization
• Google Earth — spatial understanding at planetary scale
• Windy — dynamic visualization of meteorological fields

⑥ Networks and relational structure

• Cytoscape — an open-source platform for complex network analysis and visualization; the de facto standard in biological networks
• Gephi — general-purpose network analysis and layout
• Neo4j Bloom — a visual exploration interface for graph databases
• Kumu — a mapping tool for relationships and systems thinking
• Obsidian’s graph view — visualizing the link structure between notes

⑦ Concepts and thinking structures

• CmapTools — the classic concept-mapping tool
• Nicky Case’s Explorable Explanations — interactive treatments of game theory, trust, and systems thinking

⑧ Engineering and mechanics

• Bartosz Ciechanowski — long-form interactive explanations of gears, combustion engines, cameras, and GPS; every article can be taken apart and dragged

Where the soft constraints show up:

• Gears may be drawn thicker than reality so the meshing is legible
• Organelle proportions may be distorted to highlight a structure
• An animation may slow a process down a thousandfold to reveal its order
• As long as the mechanism is correct, the representation is free to make trade-offs

But soft constraints do not mean no constraints. The most common failure in knowledge visualization happens exactly here: the animation looks great, but the mechanism is wrong — showing a tree rotation upside down, or making recursion appear to execute breadth-first.

4. The Blurry Middle

There’s no clean dividing line. Several characteristic overlaps:

Network graphs — The same Cytoscape instance drawing social-network users is data visualization (nodes are measured entities); drawing gene-function associations is knowledge visualization (nodes are concepts). The test is whether a node is a “measured object” or a “concept.”

Digital twins — Primarily real-time data (data visualization), but containing physical models and rules (a knowledge attribute). These systems typically have to satisfy both sets of standards at once.

Geographic visualization — A choropleth is data visualization; a geological cross-section or a plate-tectonics schematic is knowledge visualization. Same medium, different object.

Explanatory graphics in data journalism — The main chart is data visualization; the accompanying mechanism diagram is knowledge visualization. They routinely appear in the same article.

5. A Workable Test

When you’re unsure which category a graphic belongs to, ask three questions:

Question 1: Is there a “true value” that must be matched?

Yes, and changing the number would be an error → data visualization No, and numbers may be adjusted to make things clear → knowledge visualization

Question 2: Remove the graphic — how much information survives?

Almost none (the data and the conclusion live in the graphic) → data visualization Quite a lot (the core is a conceptual description; the graphic merely assists) → knowledge visualization

Question 3: When it fails, is it “misleading” or “misunderstood”?

The graphic distorts the data and readers draw a wrong conclusion → data visualization The visuals are smooth but readers build a wrong model of the mechanism → knowledge visualization

6. Why the Distinction Matters

One — the evaluation criteria differ and cannot be swapped.

Demanding “information density” from knowledge visualization produces dense graphics nobody can read. Applying a “schematic is fine” mindset to data visualization produces misleading work. The two kinds of “good” are entirely different kinds of good.

Two — the required team skills differ.

Data visualization rests on data engineering, metric governance, and chart standards. Knowledge visualization rests on domain expertise, interaction design, and animation implementation. Few people are strong in both.

Three — the target market and acquisition path differ.

Buyers of data visualization are enterprises (BI departments, data platforms, operations centers): long decision chains, large budgets, emphasis on compliance and integration. Buyers of knowledge visualization are educational institutions and individual learners: short decision chains, lower unit prices, but stronger virality and repeat use.

Four — and this is the easiest trap: using the wrong tool.

Building algorithm teaching on Grafana means fighting its data model. Building anatomy teaching on a dashboard tool means landing in “looks good but can’t be explored.” Tools are designed for goals.

Conclusion

Data visualization answers “what does the data say?” Knowledge visualization answers “how does this work?”

The ceiling for the first is accuracy — not misleading is the minimum bar. The ceiling for the second is transfer of insight — growing the same model in someone else’s head that you have in yours.

Their highest achievements look nothing alike: one lets you spot an anomaly across a hundred thousand rows at a glance; the other lets you genuinely understand quicksort in ten minutes.

---

All sites cited here were verified as reachable at the time of writing. For the academic grounding of knowledge visualization, see the Eppler & Burkhard framework and the Chinese-language literature introducing the field from 2005.