Speeches by Obama versus speeches by Clinton. Blogs by men versus blogs by women. Song lyrics from the 80s versus lyrics from the 50s. The list of tag clouds on Many Eyes is a study in contrasts.

There’s no question that our users like visualizing the differences between related texts, but making comparisons by looking at one text a time is difficult. Today we’re launching a new version of our tag cloud, which we hope will allow for easier and clearer analyses.

If you want to compare two texts directly, you can merge them (see the instructions for details) and then see a special “interleaved” tag cloud, which will let you compare side by side the relative frequencies of the words in the two texts. You can see an example here:

which shows the US presidential State of the Union address from 2002 and 2003. The 2002 speech is in orange and the 2003 speech is in blue. You can see a number of differences directly: “Afghanistan” shrinks dramatically from 2002 to 2003, and “Saddam” seems to grow in proportion.

Give the new tag cloud a spin! We’re looking forward to seeing what comparisons our users draw next.

One of the joys of data analysis is the “aha” feeling when you figure out how different variables interact. Many of the visualizations on Many Eyes are good for inspecting the relationship between numerical variables–but until now there’s been a hole. How do you understand the relation between categories, things like political affiliations or occupations?

To fill this hole, the Matrix Chart was created by our talented intern Lee Byron. (You may already be familiar with his visualizations of music listening on last.fm.) This visualization, shown below, is a flexible and powerful way to make multivariate comparisons. It’s good for data with several non-numeric columns. For example, the matrix chart below shows data on members of the 17th Canadian Parliament, broken down by political party (y axis) and former occupation (x axis).

Here’s a second example, a visualization of NBA jersey colors. Not the most serious visualization on the site, but it does show off color customization, a first for a Many Eyes graphic.

You can read more about the technique here, and we’re collecting examples in a special topic hub, where we’ve put matrix visualizations of some existing data from the site. (One of the nice things about adding a new visualization is that it lets us revisit older data sets with a fresh pair of eyes.) You can leave comments or ideas for new features in the discussion area of the topic hub.

One last note for the technologists in the audience. This is our first experiment with Adobe Flash. We’re interested in your feedback on this as well–we’ll probably be conducting more experiments in the future.

New color slider

Our treemaps use color scales to portray numbers–which can be a tricky business. The treemaps make their best guess at the right scale, but sometimes it’s helpful set the color range by hand. Outlying values might “wash out” the rest of the scale, for instance, or for consistency you may want to set several visualizations to use the same color map.

Our color slider lets you make manual adjustments. The colored bar shows the current color map overlaid on the distribution of data values. The numbers indicate the extreme ends of the scale. (If there are values beyond these ends, they are “capped” at the extremes and a ‘+’ or ‘<' will appear next to the values.)
Interaction

Adjust the scale by clicking and dragging the brackets, or simply type directly in the labels next to the scale. Custom values are maintained when the user changes the dimension that is mapped to the color. To restore the initial color map, right-click the color scale.

Sorting

With few exceptions, Many Eyes visualizations have been faithfully showing information in the order it is found in the data set. For example, if a data set about GDP is alphabetically ordered by country name, then the bar for Afghanistan will appear at the beginning, the bar for Zimbabwe at the end. However, as Stephen Few has noted before, sometimes it’s useful to sort items by numerical value (e.g., sorting the bars from lowest-GDP country to highest).For this reason, we’ve added a “sorting widget” to the bar, line, and stack graphs so you can re-order by:

- Labels: alphabetical order (ascending or descending)
- Values: numerical order (ascending or descending)
- Data Order: follows the order of rows in the data set (same order or reversed)

The third sorting option - data order - is often helpful for time series. If your data set has columns that go “backwards” in time (recent to old), toggling the “data order” button will cause your visualization to show time going forward.We hope you enjoy the new features!

We’ve updated a number of the visualizations on the site, in response to user requests. Changes affect the bubble chart, the treemap as well as the stack graph. In this post I will discuss these and give a couple of example uses. Look for more feature additions in the next few weeks (hello, sorting!)

Bubble chart

The old bubble chart looked pretty, but pretty blue. That seemed like an opportunity to show another dimension. Wouldn’t it be nice if you could color each of the bubbles by a categorical attribute? The examples below show a before-and-after: what happens when you color a bubble chart of the US budget to show budget categories.

You can combine coloring and aggregation to create a variety of other views, including some that look like miniature pie charts. These are experimental, and we’re interested in your feedback–and look for a more detailed discussion in a future blog post.

Treemap

The treemap previously required you to upload a numerical column with your hierarchical data. This had two disadvantages: In many cases, you might be simply interested in the number of items in each category. In other cases the treemap’s scaling can create tiny, hard to spot rectangles. The treemap below shows an overview of a backup log, but because the size is proportional to the number of bytes backed up, it contains a large number of tiny rectangles.

If we want to get an overview of the number of jobs run, instead of the number of bytes backed up we can choose ‘no selection’ for the size column, and the treemap will assign equal sizes to each log entry. We also let you color by categories, which makes the overall structure more clear. The image below shows the backup data using constant size and colored by hierarchy level (click for live version).

Stack Graph

You may have seen the reordering widget at the top of the treemap. Now we’ve incorporated that same widget in the stack graph for categories. In the sample below I reused the budget data to show the increases in planned US government spending. Rearrange the widget yourself to see the effect.

We encourage you to try out some of these improvements and see for yourself, so, stop reading this blog post and go play with some data!

Question: What do Pride and Prejudice, Swinburne’s poetry, and Green Eggs and Ham have in common?

Answer: Users of Many Eyes have uploaded word frequency data for each one.

One goal for Many Eyes was to find out just what kinds of data people wanted to visualize. In the past few weeks we’ve noticed a strong, unexpected trend: our users are extremely interested in looking at “unstructured” data sets, from Victorian poetry to the silliest Seuss, from a Ph.D dissertation to tags on del.icio.us. In response to this demand we’re happy to announce a new technique: an interactive tag cloud.

A tag cloud is a collection of words where the font size for each word corresponds to a numeric value, such as frequency. These purely verbal pictures are at least a decade old but have become something of a Web 2.0 emblem, visualizing the descriptive “tags” users attach to objects. Our tag cloud–created by ace intern Guilherme Boettcher–follows the standard model but adds a couple of features that make it more than an emblem.

You can do an instant, interactive search to narrow down the set of terms. As with all our visualizations, you can highlight sets of items to point out discoveries to other people. You can use a tag cloud on tabular data (a column of words and a column of numbers) but we’ve also revamped our data model to allow you to upload freeform text. That means we can do word frequency calculations for you, including removing common words across several languages and an option for finding two-word phrases. We also can show you the context of any word or phrase when you point to it with your mouse.
This technique still qualifies as experimental—and we encourage you to experiment! One excellent source of public domain books is Project Gutenberg. We’re curious to see what you come up with. (You can read this introduction for detailed instructions as well as a discussion of some of the pluses and minuses of the cloud technique.) And now that Many Eyes allows you to upload freeform text, look for more “unstructured” visualizations in the future.

Information visualization (infovis) has always been something that is done by experts for experts. Think scientists in white lab coats pouring over visualizations of complex data sets in order to further human knowledge. You might also be reminded of economists and financial gurus forever hypnotized by the glow of monitors showing esoteric visualizations of streaming financial data.

But in the last few years there have been a couple of occasions when infovis has dabbed into the public arena to help spark debate and insight in exciting ways. Take, for example, the series of maps of the US that circulated on the Internet and in the media showing “red” and “blue” states during the last presidential election:

As this page from the University of Michigan shows, the first map that circulated showed each state colored in either red or blue depending on whether a majority of voters voted Republican or Democratic. It was a fine representation of the country but one that gave the superficial impression that the “red states” dominated the picture, since they covered far more area than the blue ones. Quickly afterwards, another map came along showing voting patterns in counties instead of states and coloring each county based on percentage of votes in order to represent results more accurately. Finally, a third map distorted the size of counties based on population count (a technique also known as a cartogram), which better represented the high concentration of people in big cities.

In short, the election maps got progressively more sophisticated as people tried to understand voting results. They also illustrated the fact that there are multiple ways of telling the same story. The maps became an essential part of a national debate on politics, a divided country, and what it means to represent complex data.

We believe this kind of collective sensemaking is not an isolated phenomenon, but rather, an exciting example of social data analysis around visualization. This is precisely our intent in having created Many Eyes: to enable people to collectively reason about the trends and patterns they see on the vivid representations of data called visualizations.

We invite you to visit Many Eyes, play with the visualizations, upload data, share your perspective, and get conversations started. Already we are seeing users discuss a plethora of topics ranging from McDonalds nutrition data, to bioinformatics, to the bible.

What’s in your data set?