A Taxonomy of Data Visualization

A Taxonomy of Data Visualization

For some time at Visualizing, we’ve been working on a commonsense taxonomy of data visualization. This is still a work in progress, but we wanted to involve the wider community in the discussion.

There are already all-inclusive glossaries of specific techniques, and there are several academic approaches to classification (Bertin, Schneiderman, etc.). But we're looking for something in between: a general, top-level language to describe the forms used in visualization and information graphics. It should be useful to experts and non-experts alike, and so requires a balance between familiar words and ideas on the one hand and rigorous thinking on the other.
 
Three similar attempts at a commonsense taxonomy have been brought to our attention:
Influenced by these precursors and based on a survey of projects on Visualizing.org, we’ve come up with the first draft of our own taxonomy (illustrated with pieces by our users):
 
 
 

A few notes:

  • These categories are meant to describe a visualization’s primary method of comparing data. A map of the US with circles scaled to show the population in major cities is comparing quantities (Chart) but the primary comparison is geographic (Map). Some pieces, however, may fall into multiple categories equally.
  • You may question the inclusion of “infographic” in this taxonomy. Taking a cue from Tufte, we would argue that infographics are distinguished from other forms by the large proportion of “ink” they dedicate to explanatory or communicative goals rather than data comparison. Whereas a network graph, for instance, dedicates most of its visual material to representing interconnections, an infographic dedicates its material to branding, visual style, and illustrations. Infographics therefore deserve a separate category to mark this emphasis on communication over comparison.
What do you think?
Share your ideas or your own taxonomy of data visualization in the comments section or on the
Quora question we’ve posted

 

Add a Comment

Login or register to post comments

Comments

theresamarierhyne's picture

What about the use of wind vectors, glyphs, parallel coordinates and particle traces?

This taxonomy is 2D oriented. There are 3D structures from the field of scientific visualization such as volume rendering / isosurfaces that might be relevant.

Smiles... Theresa-Marie Rhyne
theresamarierhyne@gmail.com

Luca Masud's picture

I think that another interesting approach to visualization taxonomies is the one developed by Yuri Engelhardt in his thesis/book "The Language of Graphics, a framework for the analysis of syntax and meaning in maps, charts and diagrams" (unluckily it's not easy to find). He has worked a lot on a meta-taxonomy highlighting all the characteristic of the works of Richards, Holmes, Bertin, Bounford and others.

The interesting part of Engelhardt's taxonomy is that it is a combinatorial one: visualizations primary types get combined in secondary.
So for example a map+statistical chart = statistical map, a map+link diagram = path map, etc.

Moreover in some way the approach is similar to yours. Visualization types are not defined by the kind of data they represent but by the kind of "graphical objects" that forms them.

If you can give it a look, it's a really interesting read.
If you can't find it, let me know.

moebio's picture

I think 'taxonomy' implies that you have to have the nuts to accept that the intuition, common sense and familiarity will be abandoned and, in return you'll embrace a classification that's driven by inner structural, hard, abstract, sometimes hidden or not evident properties. That's basically what happened with life and genetics, right? The entire Linnaeus system had to be readapted because mere phenomenae observation ('the experience of the viewer' quoting Edward) stop being useful for life classification… at least for the new goals and challenges of life classification. And this classification shift brought immeasurable and core changes to science… that even what we understand by Life has deeply changed for ever.

Yes, I think the analogy is valid and we can associate genes with data structures (in fact a gene sequence IS a data structure!).

I think It's a problem of words and aims. If you want to build a visualization projects' explorer, you don't need a taxonomy (in the way I'm using the word), you'll rather use a classification that is "informative, not obfuscating". And that's exactly what has been done here http://www.visualizing.org/explore ; and Manuel' syntax is also a great example.

But a taxonomy it's also required, and I don't know any good attempt. But if a taxonomic system is not intuitive, not directly related with human experience, who could be useful? There are many answers for that question. I'll give two:

1. Answer to questions. Imagine you have a list of cities, a list of texts (the cities' histories), coordinates, and dates (cities foundation dates). A visualization taxonomy built up from data structures will give you a decisions tree of visualization methods that you can apply for this data set.

2. Evolution opportunities for the field. Having a taxonomy and a way to explore it (a map) will reveal which combinations of data doesn't have (yet) associated visualization methods.

Edward Lee's picture

Thanks for your comments everyone — very interesting points.

@myndworkz and @moebio: you both make great arguments for having a defining purpose or rationale when creating a taxonomy. Starting out, we looked at a handful of academic efforts, all of which begin with rigorous formal structures to describe the data and the representation. The problem was that these taxonomies began with the form of the data rather than the experience of the viewer, and the resulting classification systems were pretty divorced from our casual understanding of visualizations. We knew we wanted a way to categorize visualizations by their form for our new Explore page, and it was a major requirement that the groupings feel intuitive and friendly to the non-expert. Therefore, the proposed categories originate from the viewer's experience of the work: the experience of a map is mainly one of geographic relationships, the experience of a network is mainly one of interconnections, and so on. Perhaps calling this a "taxonomy" implies too rigid a categorization, but the labels have already proven useful for exploring the archive of projects. What other user-centric categories would you propose?

myndworkz's picture

Hi, I think you need to step back and look again at this. Two things occur to me straightaway, a taxonomy must be informative, not obfuscating. A classification system should have a set of defining principles, a rationale by which any candidate may be judged for inclusion or exclusion. These principles may be revised as part of the process and to handle ( and identify) special cases. It would surely be better to determine the function of the taxonomy - the 'what is it for? - so that we can determine functional categories for the various visualisations. As, by and large, we are all trying to tell a story with these tools, perhaps the defining characteristics are to be cached in literary terms? Epics, poems, novellas, pamphlets, thrillers and romance. Perhaps not, but any principle of classification, if it is to be explanatory, must involve categories which hold all their members in exclusion to those of other categories. The current offering does not fulfill that criterion, and so is not explanatory or enlightening...it provides no added value, which, after all, is the point of data analysis and visualisation (allowing for aesthetic value as well of course).
The second thing that struck me was what a perfect project to flex the very tools that we celebrate on this and similar sites. Surely this is crying out for a network of relationships involving many variables. The taxonomy would arise from the analysis of what it is we do with these different types of visualisations, why we choose one over another, and which are more successful, or, in social networking terms 'liked'. My suggestion is that instead of imposing a taxonomy we gather the data to see whether there are patterns to be found, and then test any categorising hypotheses against that data. Who knows what we may discover?

Manuel Lima's picture

Santiago, thanks for pointing out the concept of Anastomosis, this is the hybrid metaphor I've been looking for a long time. On the taxonomy: It will always be tricky to have an all-encompassing framework able to accommodate all possible types and variations. It might be easier to have a group of smaller, more contained/specific taxonomies.

BrianBBrian's picture

moebio +1

moebio's picture

This classification mixes purposes or styles (Infographic), patterns (Flow), types of data structures (Hierarchy) and data structures (Network), as in Borges tale:

"…animals are divided into (a) those that belong to the Emperor, (b) embalmed ones, (c) those that are trained, (d) suckling pigs, (e) mermaids, (f) fabulous ones, (g) stray dogs, (h) those that are included in this classification, (i) those that tremble as if they were mad, (j) innumerable ones, (k) those drawn with a verfy fine camel's hair brush, (l) others, (m) those that have just broken a flower vase, (n) those that resemble flies from a distance." (Borges, Other Inquisitions)

I don't think that's necessarily a problem, in fact yours seems to be the most common sense approach. But it has some issues. For instance, there are complex intersections between categories: A Tree is a Network, a Flow pattern might be used to visualize a Network (it is the case of the project from where the thumbnail was taken), an Infographic might be based on charts, a matrix is suitable to depict Networks, etc… When it takes to choose the proper classification for a project it could be hard, or impossible, to select a single category, and afterwards they are more suitable to be used as tags. So, they might conform a interesting group of important categories, but not a classification or a taxonomy. I insist, I don`t believe this is a problem, because actually I don't really believe there is a good taxonomic classification for visualization methods.

My proposal (and that's exactly what I did in Impure) is to classify data structures (that's a mathematical task), and then attach visualization methods to the data structures they use. Some are attached to a single data structure (a treemap to a Tree), but some are attached to more than one structure since they are able to express different structures (a matrix to a Network and to a NumberTable), and finally others are attached to more than one structure because they combine them: (GeoPolygon + NumberList >>> Map with circles).

The result of this process is not a hierarchical classification, a taxonomy, but at least the visualization methods lay upon a 'skeleton' that is (the Tree of data structures). Thus, the resulting structure is not exactly a pure Tree but a Tree whose branches anastomose http://en.wikipedia.org/wiki/Anastomosis