From CS294-10 Visualization Sp11
Lecture on Feb 16, 2011
- Generalized selection via interactive query relaxation. Heer, Agrawala & Willett. (html)
- The visual design and control of the trellis display. Becker, Cleveland and Shyu. (ps)
- Exploration of the Brain's White Matter Pathways with Dynamic Queries. Akers, Sherbondy, Mackenzie, Dougherty, Wandell. Visualization 2004. (html)
- Fry's zipdecode
- Wattenberg's NameVoyager
- LA Homicides. Heer's example of generalized selection.
Julian Limon - Feb 16, 2011 01:17:06 pm
I found the selection and query relaxation techniques presented in Heer et al. to be really interesting and enlightening. They provide users with the ability to dynamically select different characteristics of the data set and to see their selection in multiple visualizations. The query relaxation techniques allow the users to expand their selection based on a semantic model, which can be hierarchical or flat. However, I was a bit disappointed with the results of their study. I was hoping that this techniques would present a significant improvement in the ability to interact with data. Although the results do show a certain improvement, it was not as dramatic as I had expected. I wonder whether the learning curve for this new kinds of interactions plays a very important role. Participants were only given certain tasks and had limited time playing with the techniques. It would be interesting to see if the selection and relaxation provide better results once the users have been using them for a longer period. That kind of analysis could inform new designs--a designer could know if it is worth including these techniques depending on the kind and length of the tasks that users will perform.
I was also curious about the query language. It seems that the language was used internally to generate new queries depending on the selection, but was never exposed to the users. It would be interesting to see if users would understand the system better if they understood the new queries that were generated when they click or select something. This probably won't work for all types of users, but depending on the specific audience of the visualization one might assume a certain familiarity with SQL-like sequences.
Dan - Feb 16, 2011 02:21:22 pm
The generalized selection via interactive query relaxation was really cool. I like how you can convert graphical and interactive data into semantic queries for pulling useful information from datasets. I like the graphical communication methods. The interface also provided a lot more data by the fact that it was interactive.
Exploration of the Brain's White Matter Pathways with Dynamic Queries was also very interesting. Combining MRIs and visualization to visualize neural pathways is groundbreaking. The images produced were also of very high quality. This was impressive work. It seems that the interactivity of the software they used was definitely a part of making this work so impressive, though. Perhaps the raw data by itself was not very interesting but with a few tweaks they could highlight patterns and find interesting forms of visualization.
Matthew Can - Feb 16, 2011 10:33:17 pm
Julian raised a good point about the learning curve for the new interactions presented by Heer et al. For me, the biggest problem with the generalized selection was related to this, but I would call it a discovery problem rather than the learning curve. The first time I played with Heer’s LA Homicides example, I only used simple selection interaction. It was only after reading the paper that I learned how to make the system execute relaxed queries.
One of the strengths of this system is its implementation, the translation of selection interactions into an underlying query language. This makes the system modular and helps support collaborative visualization. For example, this allows the system to link annotations to queries, so that the queries can be replayed when the annotations are shared. It also makes the system robust to changes in the underlying visual representation of the data. That is, the visual display can change and the queries still work.
I would have liked to see more in the paper about the tool for authoring semantic hierarchies. I think this is an important part of the system, because without it only attribute-based relaxations are possible.
Michael Hsueh - Feb 16, 2011 11:04:14 pm
I am impressed overall by the work on visual dynamic queries. The system for exploring white matter pathways by Akers et al. allows researchers to simplify a large amount of physiological data into manageable displays while seamlessly retaining the integrity of the full body of data. The brain renderings appear to be of high quality, and the interface seems intuitive and powerful in the hands of trained specialists. We've seen the VOI technique in other software and I have no doubt that its potential applications are far reaching.
The work by Heer et al. is also great. I agree that tools for specifying semantic hierarchies and determining how they interface with query generation modules are a key step towards generalizing query relaxation. The experiments/results of the project are encouraging. Of course, the full potential of dynamic/relaxation query techniques are simply yet to be discovered as new optimizations, applications, and user proficiency related to these techniques come about. We encounter dynamic query / relaxation applications in our day to day interactions with electronic retailers (e.g. Amazon.com) and their product search and categorization features. Granted, they count less as data visualizations, but the basic concepts of rapid, dynamic query refinement wrapped up in a pretty front end are there.
Siamak Faridani 02:03, 17 February 2011 (CST)
I am wondering if similar to databases we can harness the power of relational algebra for visual queries. For example in the "Generalized selection via interactive query relaxation" paper, let's say we want to not just select datapoints but also perform some sort of function on the set. I can imagine averaging, or summation commands is easy to perform but what if I want to select all the points that are neat china and have a population higher than 5 million and then average their population group by the language they speak? we can express this easily in relational algebra but how can we convert the relational algebra to visual queries?
Krishna - Feb 17, 2011 01:43:14 pm
All the visualization systems enable users to find only exact matches for their queries. I believe, compared to systems that would rank data points based on a query, these systems would require users to go through a greater number of exploratory steps to find their information. Because of the exact match criteria, the filter boxes and cubes may need to be changed and repositioned multiple times as the query gets refined, whereas there is a reasonable probability that the user might find the intended information even if it only fuzzily matches his queries. However, as discussed in class, it is not entirely clear how such a system could be implemented and qualitatively analyzed.
Sally Ahn - Feb 17, 2011 03:59:24 pm
The generalized selection tool is definitely one of the most complete and user-friendly visualization tools I've come across. What I like about this tool is how they simplify this the query relaxation process for users by allowing them to specify the attributes of a an existing data point as the domain of relaxation. As the authors of the paper point out, not many interfaces allow users to combine different forms of reference (e.g. detailed and deictic), and they try to overcome this limitation to allow for more "fluid" interaction. I also appreciate the details of the software, such as the smooth animation of points from one coordinate system to the next, which I think helps the user process the changes that are happening between the transitions and reduces the chances of change blindness that might occur if the transitions were immediate. There may be certain data sets where the patterns of change in transforming the data from one set of axes to another could lead to important insights.
Karl He - Feb 17, 2011 04:40:32 pm
The generalized selection paper presents a great use of interactive visualization. The ability to easily select subsets of data is pretty useful by itself, but it truly shines with the ability to maintain the selection while changing the type of visualization to something like a bar chart. The system also gives numerous other ways for a user to modify his selection or the view on his selection. The abilities granted to the user allow him to ask questions that the original creator of the visualization might not even have taken into account, such as "the number of 15-20 year old murderers by month". Like the author stated, its more like querying the system than just looking at a static visualization. This seems to be one of the better uses of interactive visualizations.
Saung Li - Feb 17, 2011 05:52:17 pm
The visualizations showing the brain's white matter pathways packs a ton of data into an aesthetically pleasing graphic. Using box-selections to spot pathways can help neurologists understand the brain better, and I can see such interactive visualizations be applied to other areas. For example, seeing where a blood cell travels might be an interesting visualization. As mentioned before, it would be great if we could somehow build an interactive tool that takes in general data of some type and automatically generate visualizations of them, instead of having to build these tools from scratch every time.
The generalized selection paper is quite useful as it shows us what we can do to narrow down what we want to see, given a complex overall visualization. Once again we see a close relationship between visualization and databases so that queries can be done efficiently and the interaction is smooth. I hope to see more tools like this one, as they are user-friendly, compelling, and visually pleasing. Perhaps such tools could have brief descriptions somewhere alongside them explaining the important things user can do so that they don't have to play around too much to learn how to use them.
Thomas Schluchter - Feb 17, 2011 11:44:42 pm
When I read about dynamic queries, I wondered about the relationship between the entire visualization and individual data points. Dynamic queries are great because they allow a user to watch what happens to the entire data set when they manipulate a certain parameter -- that way, the user can answer fairly complex questions easily (or discover new questions to ask). Dynamic queries help understand the data as a whole. But what if the goal is to learn something about and individual data point? As the display of data changes with each continuous manipulation of the interface, the visibility of a data point is a) somewhat unpredictable and b) difficult to recover.
Has there been any work on maintaining a history of changes made to a VIS interface in order to recover a combination of parameters to show a specific data point?