From CS 294-10 Visualization Sp10

Jump to: navigation, search

Lecture on Feb 3, 2010




  • Visual information seeking: Tight coupling of dynamic query filters with starfield displays, Ahlberg & Shneiderman. (html)
  • Visual exploration of time-series data, Hochheiser & Schneiderman. (html) (pdf)
  • Postmortem of an example, Bertin (pdf)
  • The visual design and control of the trellis display. Becker, Cleveland and Shyu. (ps)

Optional Readings

  • Table lens, Rao and Card, (acm)
  • Human guided search: Survey and recent results, Klau, Lesch, Marks & Mitzenmacher. (pdf)
  • Design and evaluation of incremental data structures and algorithms for dynamic query interfaces. Tanin, Beigel & Schneiderman (citeseer)


Danielle Christianson - Feb 02, 2010 08:51:46 pm

My general reaction to all of these efforts, with the exception of Bertin, is that there seems to be a relatively high learning curve for use. In the ecological academic world, I have yet to encounter use of interactive data manipulation other than one custom built application by a student for a class project (granted my scope is likely limited). But it makes me wonder about the dissemination of interactive data visualization tools and why they are or are not being incorporated into various disciplines and academic/industry settings -- too high of learning curve? not easily customizable to different datasets? unawareness? And the demos all seemed relatively recent but the articles (not all have dates) seem pretty old (early 90s?) -- this surprised me.

I agree with Hochheiser & Shneiderman's inclusion of a kinetic data qualifying method (vs. querying by text or slider bars) -- I think it, although less precise, is very intuitive.

I though Fry's Zipcode visualization was really neat.

Jiamin Bai - Feb 03, 2010 02:31:44 am

What might seem to be common sense to us, (tight-coupling in many applications) is well explained and presented by Ahlberg & Shneiderman. What I thought was interesting was the idea of 'immediate and continuous display of results'. I think right now, it seems like the gold standard for continuous display of results is animating the transitions between states. We see the success of this with the iPhone (i can't help it), where transitions from apps, scrolling of phonebooks, browsing of music is done with meaningful animation.

Jeffrey Patzer - Feb 03, 2010 11:20:34 am

I think the most compelling idea behind Interaction with data and graphics is the idea that you can literally play with your data. As humans we love to interact with our world and the idea that you can "put your fingers in the pudding" so to speak, with your data, opens up a whole new realm of what graphics can do. When you have the chance to interact with the data, it allows for information that might have once been very difficult to discover, become quickly apparent. I really liked the idea of the movie finder. In fact, I am surprised that it has more capabilities that current movie finding websites. For instance, if I try to use netflix, they allow me to search with certain terms for movies, but the visual tool that allows the specification of data and instant visual result, does not exist. Imagine how much quicker my trip to blockbuster could be if I could walk in, use their visual search terminal to find a movie that suits my needs, give the movie name to the BB employee, they get my movie, and I can leave. The time spent there is much less and the effectiveness of my search much greater. What I am getting at is that while many search tools use text search, if a visual search were implemented, the effectiveness of the search would greatly increase.

Arpad Kovacs - Feb 03, 2010 12:02:34 pm

Although each of these papers felt like a "how to" guide for creating a certain type of visualization, what I found most useful were the direct manipulation design philosophies they have in common.

Ahlberg & Shneiderman's Dynamic Homefinder seems to be a trailblazer for modern scatterplot visualizations such as, which often seem to be just refinements of its successful formula of dynamic AND logic query filters, details on demand, and tight coupling.

It is amazing how closely Shneiderman & Hochheiser follow this formula, even though they are dealing with time series data. Timeboxes are simply another interactive mechanism for specifying range constraints in the time domain using AND logic, and timesearcher again reveals details on demand.

Finally, Bertin's scientific methodology of constructing visualizations also relies on the fact that each profile can be rearranged to provide a physical implementation of boolean filters. As in the other examples, this "mobility of image" lets the eye simplify and summarize data efficiently.

Aaron Hong - Feb 03, 2010 12:20:51 pm

From the Ahlberg & Shneiderman reading, they expound on the concepts they used in creating FilmFinder. Of the three main concepts they talked about, dynamic query filters, starfield display, tight coupling, only dynamic query filters and tight coupling have seen an increase in use on the internet. The internet being a wealth of data could see better and more interactive designs, many that are talked about in Ahlberg and Shneiderman's article. We are stuck large using lists, tables, and text as visualization when we have barely explored the 2d plane. Although this report was written awhile back, we have not seen such features in common use such as the starfield. Only in recent years have dynamic queries and tight coupling show up demonstrated in Firefox's 'awesome bar' and Google's auto-complete. Much more can be done since interactive websites are the norm these days, however many times we're using it for frivolous purposes.

Although Hochheiser & Shneiderman's Timeboxes are useful for filtering data, which is simple and powerful. One thing they should have probably incorporated into their test was the use of other visualization methods (layering) on their 2d graph to help find patterns and points of interest. Some people may know where they want to draw their Timebox in the gray clumps, but some are looking for places to delve into. Starting with a useful graph (not just the gray clump), would be helpful for the users whether they use the Timeboxes or not.

DavidZats - Feb 03, 2010 01:07:28 pm

These papers covered a variety of topics. One was how interfaces should be designed so users can most effectively obtain the information they desire from large datasets. Examples of this included allowing users to quickly and easily modify queries and view the effects of these modifications. Another example of a useful feature was to allow outputs to be used as inputs. The topics also included the a description of the power of graphs to identify trends that would not be visible otherwise. Lastly, the readings introduced Trellis graphs, and showed how effective they can be at portraying data.

Jon Barron - Feb 03, 2010 01:08:55 pm

Ahlberg & Shneiderman:

Nothing mindblowing. The paper is well summed up by one of its sentences: "This starfield approach is a scatterplot with additional features to support selection and zooming". I do like the "tight coupling" / "display invariants" idea, as its something that is immediately recognizable as a good property for an interface, but lacked a name.

Hochheiser & Shneiderman:

This is actually a very clever and elegant idea! I'm not sure how useful it is to be able to phrase queries in such absolute terms... relative queries seem more useful ("give me a stock that doubles from t_1 to t_2, and then tanks from t_3 to t_4"), but that doesn't seems possible in this framework.

Becker, Cleveland, and Shyu:

I don't get it. Why don't these plots have labels?

Chetan - Feb 05, 2010 09:11:06 am

I especially liked how interaction is thought of in a general sense, not specifically tied to computers. The Bertin reading highlighted this fact -- here the user used slips of paper to 'interact' with the visualization.

With interaction in general, there seems to be a tradeoff between simplicity of design and power. Of course there's the holy grail of making everything simple and powerful, but that often times is not possible. It seems that some of the designs showcased here are indeed powerful, but not terribly simple or intuitive at first glance.

Stephen Chu - Feb 07, 2010 05:47:50 pm

Bertin writes, "Let us note that the information which preceded decision-making can always be written or imagined in the form of a single table of numbers..." I found this to be a useful rule to keep in the mind while we design our own visualizations. By following this rule, we are forced to understand and define the problems clearly (something that I forgot to do during my original redesign attempt of Assignment 2).

I also agree with Bertin's statement that modern graphics are not "'drawn' once and for all" but they are constructed and reconstructed over again. There are many, many different possible ways of creating the graphic and chances are your first time making the graphic will not be your best possible work. With the improvement of visualization tools, we increase the number of ways to quickly change our graphics, and I think this means we have to place more emphasis on revision. It is something I definitely need to spend more time on.

Jonyen - Feb 07, 2010 10:08:44 pm

Starfield displays are pretty nifty and the info visualizations are very helpful to see, such as the HomeFinder and ZipDecode visualizations. I think that one of the flaws of starfield displays is that they don't provide any further contextual data about the dots until you click on them to get more detail (provided that this functionality is implemented in that particular visualization). If some color or size information was provided, that could be helpful.

I liked the cellphones visualization too, which was pretty handy. Although I do have to question how they decide to categorize the cellphones into each particular specification (thinnest, lightest, etc). It would also be helpful to know how many cell phones are left when you are about to click on a particular tab, rather than having to click on a tab and then see nothing. It would help to save time to have information provided before the act of clicking or moving the mouse.

Akshay Kannan - Feb 08, 2010 01:25:54 pm

While many of the visualization software appeared to be quite different from the polished, cleaner tools that we see today, the aspect of interactivity adds a whole new dimension of information unavailable in a static printed form. The ability to filter interesting data points, or to only display certain points of interest switches the story-telling role from the visualization designer to the user. Rather than the designer having to pick a single story that they force upon all users, the user can choose to display only the relevant data that interests them and form their own conclusions. It will be interesting to see how we utilize our concepts from print visualizations towards designing interactive computer visualizations.

Prahalika Reddy - Feb 09, 2010 05:33:12 pm

A lot of the concepts in the readings and the lecture reminded me of the concepts we learned in CS 160. The user interface concepts definitely apply to the interactive visualization tools that we talked about and take them a step further because they return a visualization encoding complex data. For these tools, the user interface is the most important part since the user's interaction defines the tool.

I really liked the various interactive tools we saw in class, but some of the interfaces for them could easily have been improved. One of the tools I thought was very useful was the HomeFinder. I'm surprised that something like that was created so long ago, and yet today, finding houses or apartments is so hard. On the other hand, the cell phone finder didn't seem that useful at all because it wasn't very accurate. Some of the categories seem useless since they return no results, even when they are the only criteria, making me wonder why the category exists in the first place.

Shimul - Feb 09, 2010 08:57:37 pm

The article on tight coupling had some great examples that were also shown in class. The interactive tools are informative visualizations that go a long way in helping analyze data. I personally like the HomeFinder tool a lot. There is coupling in softwares like Office products and gmail which is indeed useful. If a picture speaks a thousand words, an interactive tool speaks a million! That being said, if there are too many features/sliders/buttons to play around with, things start to look rather intimidating. I am not sure of Photoshop is a befitting example in this context, but due to all the clutter of available functions/features, it has a steep learning curve.

The postmortem of an example was actually slightly confusing. The article walks the reader through the thinking process, but I am still confused about the image transformation and how the second image can be interpreted.

Subhransu Maji - Feb 09, 2010 09:19:37 pm

I find the tight coupling principle quite powerful. It makes a big difference on the usability of the tool. Its interesting to think about ways to achieve tight coupling when the updates are expensive. As an example, in Photoshop, a popular image editing software, adjusting the parameters of complex image filter may take a while to apply, violating the tight coupling principle. The application does something in between where they show the result in a small region of the image for quick feedback.

I had mixed feelings about "postmortem of an example". Since the basic idea is obvious, i.e. organizing the data, to make the patterns clear, it was a little boring to read. Nevertheless for nominal data organizing the data this way clearly shows the correlations among the variables quite well visually

I found the time boxes interface quite appealing. One way of improving this might be to color code each of these lines so the even when a bunch of lines are selected it is easy to make the distinctions between them.

Kerstin Keller - Feb 09, 2010 11:33:32 pm

Ahlberg and Schneiderman

I think that a lot of the topic that is covert in this paper - at least the concept - seems to be somewhat trivial today. Whether we use online shopping on Amazon, searching for flights on Expedia, using google maps, queries are made and result can be narrowed down and displayed ordered by different attributes. Still, websites can be confusing and some would do good in rethinking their display of given data.

Timothy Wheeler - Feb 09, 2010 11:48:37 pm

For relatively small data sets, such as Bertin's example, being able to dynamically interact with the data by moving things is a huge advantage. There is very little cost in modifying the original query or ordering of the data. However, for data sets that must be queried frequently, these low-cost dynamic queries may be too time consuming or not powerful enough. It may be useful to have a vocabulary of search operators that correspond to the dynamic widgets. Then, as the user dynamically modifies their query, a text box automatically updates to show the corresponding text-based query. The user can then type in the search box to precisely refine their query. One example of this behavior is Google's GMail -- If the user clicks the "Important" label hyperlink, the search box reads "label:important". This search box can immediately be modified to something like "label:important !label:finished in:trash".

Basically, the low-cost (but potentially imprecise) dynamic queries could be used to provide a starting point, and the more difficult text-based queries could be used to refine the search.

Mason Smith - Feb 10, 2010 01:12:23 am

Overall, I liked the A&S paper, but I didn't like how they glossed over the problem of OR queries. They suggested that users were satisfied with, or "preferred" making multiple queries, but the papers they cited regarding it didn't really discuss the problem (for instance, none of the systems studied allowed for OR queries).

Paul Ivanov - Feb 10, 2010 11:07:08 am

Immediately upon seeing the table at the beginning of the Bertin reading I was reminded of a completely analogous example often shown in classes about vision (both computer and biological). The lectures shows a big table with filled with a whole bunch of numbers, informs the class that the numbers are intensity values at different pixels in a picture of a person taken using a digital camera. The question follows: "Who's in the picture?" And it's next to impossible to tell (unless you've seen that slide before), but the next slide reveals the picture, with the same granularity that is formed by the converting the numbers back into intensity values, and there's Bill Clinton (or Einstein, etc). To someone studying vision, the point is "both slides contain the same information, but the numbers on the first slide is exactly the representation that a computer vision algorithm or visual system has access to: so how does it integrate all of that to be able to recognize objects, find patterns, etc. The point of the exercise is complementary to the one Bertin makes. Bertin says: you've got a visual system that's capable of doing all of this amazing stuff, so make use of it!

I think I understand the 2 types of questions in any data table (introduced by x, and by y), but it's not clear what's an example of a representation that "destroys any entry categories used in constructing the data table" - which the principle that any graphic should be able to answer both types of questions eliminates.

Zev Winkelman - Feb 10, 2010 04:52:43 pm

Some of the concepts and software discussed seem a bit dated, but it is nevertheless good to revisit the origins of designs that are still in use today.

I was particularly interested in the Hochheiser and Scheiderman "Timeboxes". I have had some experience in the field, but had not come across an interface like this.

I wonder if the design can be built to handle realtime data flows seen in today's marketplace.

Regardless, this technique of query formulation for time series data, whether financial or some other kind, is very interesting.

In the Bertin article in the section on "Definition the Problem" (3) is it still held that this is "purely a problem of imagination which no machine could solve. This first and fundamental stage in decision-making cannot be automated." ?

Boaz Avital - Feb 10, 2010 10:22:29 pm

The paper on FilmFinder strongly advocated graphing values on a scatter plot (points of light). For HomeFinder, this makes sense since the information lends itself to a 2-dimensional mapping. However, for all of FilmFinder's advantages, the argument in their paper that you can plot anything on a scatter seems iffy from a usability standpoint. Does the user really care to see on a timeline when each movie came out? Is it intuitive to see a scatter of movies organized by director vs. number of male actors? Scatters can be difficult for people to comprehend.

You can tell timeboxes are a good innovation because once you see how they work, you understand them immediately and you think "why didn't I think of that?" They're also something that can be extended easily to non-timeseries data, since they're basically just an SQL command on parameters. You can extend their functionality even further, if perhaps less intuitively, by making different colored timeboxes filter on different parameters. That seems like the logical step for breaking them out of 2 dimensions of information.

Priyanka Reddy - Feb 11, 2010 03:50:47 am

I think the concept of people's perception when viewing visualization is a really interesting topic, and makes the job of designing objectively good visualizations much harder. If so much of the message conveyed by a graph is determined by how the user sees it, how can the designer ensure that he gets the intended message scross? Some of the data provided about human perception and magnitude estimate provides designers with good suggestions, but ultimately the effects of human perception can not be removed completely.

Ebby Amirebrahimi - Feb 11, 2010 12:29:42 pm

I thought the Ahlberg & Shneiderman and Hochheiser & Schneiderman readings presenting interesting case studies for interactive visualizations. I thought the concept of dynamic queries was the most powerful and the most widely accepted innovation among those given in the readings.

I appreciated Bertin's article, which followed a very logical storyline of how graphics can help a hotel manager solve problems. I like how he made the method scientific and provided a formula for creating useful graphics.

Shimul Sachdeva - Feb 18, 2010 02:33:23 pm

Visual queries and idea of generalized selection are both powerful ways of visualizing and analyzing data. The query builder exercise was an interesting example, especially how it abstracts out the underlying database queries. A Ted talk was delivered recently where they used an interactive visualization to show the economic progress of countries and population growth over decades. It used the principles taught in class.

The discussions on the Haiti assignment submissions was quite interesting. Feedback on some of the images led to interesting viewpoints regarding good and bad visualizations. Although, some visualizations that looked pretty did not seem to show meaningful data and I wonder what is more essential - making it look good or going with conventional visualizations that depict more data.

Ryan Greenberg - Feb 28, 2010 04:11:01 pm

One point from the Ahlberg & Shneiderman reading that seems so obvious as to be overlooked is the power of details on demand. This is one of the biggest advantages of an interactive visualization: since you can arbitrarily show more information, you can give users details about a specific data point that they find peculiar or interesting. This lets users answer natural questions when they're looking at data like, "What's that outlier?" or "what exactly is this point that I've zoomed in on?". It lets users make potentially qualitative judgments about data points, and helps bridge the gap between exploration of quantitative data explicit in a dataset and accessing the user's understanding of a domain. More concretely, imagine in HomeFinder that a user finds a house in a desired location that is much cheaper than surrounding options. Using details on demand, which could bring up a photo of the house in question, the user could quickly determine that the low price is because the house is hideous.

I liked the Bertin example of the hotel assistant rearranging graphs on slips of paper because it emphasizes the real core of interactive vis: it isn't about a mouse or keyboard, but rather about providing mechanisms for the user to manipulate and regroup the data to further exploration.

[add comment]
Personal tools