Collaborative Visual Analysis
From CS294-10 Visualization Fa08
Lecture on Nov 17, 2008
Readings
- Voyagers and Voyeurs: Supporting Asynchronous Collaborative Information Visualization. Heer et al.(html)
- Many Eyes: A Site for Visualization at Internet Scale.Viégas et al. (pdf)
Optional Readings
- Design Considerations for Collaborative Visual Analytics. Heer and Agrawala. (html)
Seth Horrigan - Nov 17, 2008 11:57:45 am
When introducing the doubly-linked discussion, Heer, et al. describe it as something where the user may "serendipitously hit upon a view that another person has already commented on." When evaluating the system with users, they found that although the feature was well-liked, "if there was any frustration with this feature, it was that users had to navigate to a precise location to see related comments." I would have anticipated this to be the dominant shortfall of the system. The idea of the doubly-linked comments is great, but finding the specific visualizations that elicited comments can be a difficult task relying on serendipity. I saw this in the video demonstration as well. As the authors suggest, I would like to see "social navigation cues" embedded into the visualization itself. Perhaps the "hot spots" widget they suggest, perhaps the "trackbacks" they suggest, or perhaps some new method of indicating modifications of the visualization that would lead to others' comments. Edit: scented widgets such as Jeff showed in class would probably be the most useful.
Also, I have been thoroughly impressed by Many Eyes. I have not added any data to the system, but I have enjoyed exploring others' data sets. Although the geographic visualizations do not necessarily use space and shape as efficiently to encode data (since national borders or landmass do not always correspond to actual grouping or concentration of specific demographics), I find them to be some of the most engaging to explore.
Scott Murray - Nov 17, 2008 03:01:25 pm
I appreciate how Heer, Viégas, and Wattenberg acknowledge how, currently, asychronous "collaborative" data exploration typically occurs over email, via which "writing about particular trends or views is difficult, involving awkward references to attached spreadsheets from the e-mail text." With sense.us, they have expanded the visualization's role to include that of being a social space or multi-user interaction space (as opposed to being for single users only). Something like sense.us, with the ability to use one's own data, could be invaluable for corporate environments, although it could be more interesting as an open platform like Many Eyes.
Ketrina Yim - Nov 17, 2008 08:14:06 pm
Have there been any attempts to make data analysis "chat rooms"? The collaborative environments presented during the lecture were more like wikis or discussion threads. In the analysis chat room, people could hold live discussions about data and its visualization online. Allowing people to gather in online groups in this fashion could encourage more conversation about a dataset. I suppose the annotation part could cause some complications, though.
On a side note, wordles and tag clouds seem to be very popular on Many Eyes. Is it just because most of the datasets available are speeches and documents, or do other factors drive this interest?
Simon Tan - Nov 17, 2008 09:34:26 pm
To follow from Scott's comment: I found it interesting that although it was stated that collaborative data exploration was often a difficult endeavour over e-mail, the medium (and other text-oriented mediums) was supported very well with the 'state-ful' URLs in sense.us. Managers can still conduct heated discussions with their groups over e-mail, except pointing with URLs instead of file attachments. It is just another (more bandwidth-efficient) way to get everyone on common ground.
(It's actually better because these links won't get lost in the flurry of replies and forwards of the original e-mail, as attachments tend to do.)
@Ketrina - I'm not sure if real-time chat would add much to the social aspect of the sense.us application, besides a sped-up feedback and discussion process. I thought about this in comparison to other social websites (e.g. Facebook), and they all seem to be highly trafficked without the need for truly 'live' updates.
(Although I suppose they are moving in that direction with Facebook Chat, but I feel that features like that are used more for urgent communication about an imminent event rather than enhancing the social site's other activities. I'm not sure if sense.us has that kind of need for realtime discussion.)
Perhaps real-time interaction would actually do more harm than good, since I notice that people's contributions in a social setting tend to be of a more rushed and less thought-out manner in real-time. (E.g. compare your average instant messaging conversation with the carefully written prose that is this comment post. ;-))
Maxwell Pretzlav - Nov 21, 2008 07:47:41 pm
I found both these papers and Jeff's presentation fascinating and uncomfortably relevant to my project—I wish I had read these before I started! I found the careful use of browser history and URLs in sense.us and Many Eyes particularly interesting. The ability to bookmark an interactive visualization's state, and email, post, etc the URL around seems incredibly useful. Now I think I'm going to have to add support for it in my Last.fm visualization; being an interactive online visualization, it seems irresponsible not to! For those interested, it appears there is already a well-respected actionscript library for tracking application state in URLs called SWFAddress, which I intend to try out.
@Seth regarding comment-finding: I found it notable that Many Eyes took a different approach towards comments than sense.us (possibly because of the shortcomings you noticed?)—comments on Many Eyes are all attached to one main visualization and the comments themselves encode the visualization state at the time of the comment, rather than being attached to a specific visualization state. I think this is a good compromise as it allows a discussion to explore different states and details of a complex visualization without getting fragmented and spread out.
Michael So - Nov 22, 2008 07:17:31 pm
I had not really thought about the social aspects of visualizations. Thinking about it, visualizations are meant to be seen and analyzed by people, so having a forum on discussing visualizations seems like a natural thing to have. And something like Many Eyes where it is on the internet and accessible to anyone online is nice because pretty much anybody can post a comment on some visualization, or create a visualization (given that they have to register first). And because anybody can access the site, it is nice that Many Eyes is engineered so that you don't have to be a specialist or some expert to understand how to participate in the web site.
And at the end of the Many Eyes reading, I found it at interesting idea at the conclusion where Many Eyes could be used as a platform for testing of new visualization techniques. It would be really beneficial to researchers because they have a place where they can have potentially thousands of participants in their studies. And because Many Eyes supports social features, people can post comments and you can have immediate feedback.
Chris - Nov 23, 2008 06:59:20 pm
A common feature to the collaborative systems presented (as well as other, common systems such as Wikipedia) is that they all seem to have a linear revision history. I would be curious to see what the difference in results would be were a system to make an effort to encourage "branching" revisions rather than linear revisions. That is to say, when adding a comment, annotation, or altering a visualization, the user would create a new leaf, and subsequent edits could be done to that leaf or to any other node in the revision history tree. Enabling navigation of such a history would be much more involved process (rather than simply having people view "top of tree," they would have to choose among many options), but such a system would not be without its benefits. Two benefits which come to mind are
- divergent branches allow for different avenues of inquiry to be pursued separately, rather than having to compete for attention on a "top of tree"
- evaluation of "interesting" features would be easier to gauge, in that they would be the bifurcation points between branches of differing popularity (unlike in a linear system where one cannot what particular user-generated content on a page draws other users in).
Does anyone know of any systems of this nature which have been created?
Calvin Ardi - Nov 24, 2008 02:31:27 am
Out of curiosity, does anyone know which paper Jeff is referring to on the slide regarding Merging Analysis Structures (Brennan et al '06)? I was hoping a reference list was floating around somewhere, but didn't find any. Update: Jeff sent me a link to the reference list (added above or here).
Both papers (Heer et al, and Viégas et al) presented interesting systems and approaches to facilitating collaboration at the network level. The case studies done in Heer (sense.us) provided particular insight as to how a collaborative system could be used at the corporate level (which would probably vary on the Internet scale; as experienced on Wikipedia, rampant vandalism could potentially occur on these systems, diminishing the value of the information gained unless a system of administrators and moderators were put in to place). The annotation system (graphical and textual) seemed rather useful and easy enough to use without requiring advanced technical knowledge; I'm curious, however, if there was any support for citations or perhaps a more precise way of annotation (for example, "tagging" a specific year or date, in the case of the time-series visualization of the US labor force [figure 1]). The authors also mentioned that this system would be useful for collaborative analysis of graphs/charts (e.g., sales figures for a presentation) but it didn't seem like a case study was done in this manner. It's one thing to get users acquainted with a system, but perhaps a more rigorous case study would involve this visualization system in conjunction with the work being done.
The Many Eyes (Viégas et al) system felt more developed and solid in contrast; it was pretty neat to explore the website and look at the different visualizations created. It lacks the annotation features found in sense.us, hence not necessarily facilitating collaboration (rather exploration of data set visualizations, data sets themselves, and commentary). Just browsing a few visualizations from the home page, it would be neat to see Many Eyes become a bit more authoritative if there were ways to establish authenticity or authority of certain data sources example: arctic food chain. I certainly liked the "Learn More" series, educating users on the different types of visualizations and when to use each; I think it would be more helpful if users were perhaps allowed to collaborate on these explanations like a wiki (or a commentary system) or if more resources could be linked to. I particularly like how the network diagrams were done (example: mashability index, indicating how often an artist was "mashed" or remixed in other songs). Clicking on an artist, for example, highlights all the nodes connected to it (however, this was a bit difficult to see, especially in more dense data sets; perhaps a visibility layer could be toggled when a user clicks on a node).
Witton Chou - Nov 24, 2008 04:22:30 am
I really appreciate the features that are available in "Voyagers and Voyeurs: Supporting Asynchronous Collaborative Information Visualization". The ability to save certain views and be able to draw and label what one's comments are refering to is a great time saver and excellent tool for facilitating discussions. There are countless times when I've found myself trying to use words to explain something or just merely what I'm looking at on a visualization and find myself spending more time pointing out what I'm looking at rather than discussing it. I also find the address bar's reflection of the search parameters is very useful. One of the biggest problems with flash is navigation within the contents of the flash object because the user often has to follow the same sequence of actions to get to the same place whenever he visits the page. Being able to take a url and go directly to what is important really helps sharing visualizations and referring back to it quickly. Many of the features implemented here really help promote the discussion of the data set rather than instructions on how to get to a particular view.
Yuta Morimoto
In many cases, data analysis is needed to take long time. I think most casual way to take data analysis is conducted with certain application such like Tableau and Spotfire. However,until we get a significant visualization, we need to take a sequence of trial and error processes. I think collaborative visual analysis helps me lessen such trial and error processes, since other people's viewpoints are useful to come up with a hypothesis on some data. I think a useful and easy collaborative site allows me to analyze data and interact other perspectives. I believe such accessible way can facilitate people including me to obtain and comment on visualization, even if many people are Voyeurs other than Voyagers.
David Poll - Nov 24, 2008 12:51:03 pm
I'm quite intrigued by this problem space. While the presentation focused mostly on data-driven scenarios, I think it's interesting to look at it from the viewpoint of interaction or transaction-driven scenarios. Imagine, for example, an interior decorator working with a customer, and the series of interactions they may have. How might collaborate visual analysis help here? The designer could post photos for approval, the customer could mark them up, and so on -- eventually coming to some sort of agreement on content. It seems to me that such scenarios exist all over the place in "real-world" applications, and it's just a matter of time before we start seeing a whole slew of new paradigms for collaboration due to the flexibility these applications require.
Sarah Van Wart - Nov 24, 2008 11:55:18 pm
I found sense.us to be a really compelling application for a collaborative interactive visualization. Though the quantity of data collected from the census is massive, sense.us provides the average person a way to explore and interact with it. Hence a massive information repository of important information is exposed to the general population in a simple way. I also felt that the choice of dataset was perfect for collaboration. To collectively explore a dataset in which every American is taken into account in some way (though I guess that could be a contentious statement) seems to encourage people to share their personal thoughts and analysis as a member of the whole. I also really feel like this area of exploration has so much potential. We really learn so much by discussing ideas and facts with one another, and providing more tools to facilitate this kind of discourse could make human analysis that much more powerful.
It was interesting that at the conclusion of the sense.us pilot test at IBM, employees had opted to use their more familiar communications infrastructure to discuss particular subsets of the data (referencing links to sense.us) rather than using the embedded commenting and markup tools provided within sense.us. It sounds like collaborative applications like sense.us might have to come up with some creative ways to entice users to share ideas within the system so that everyone can benefit from collective analysis.
Dmason - Nov 24, 2008 01:58:05 pm
There are a number of groups out there that are also interested in this topic. To name a few:
- COVIS at Northwestern http://www.covis.northwestern.edu/
- VisAD at Wisconsin http://www.ssec.wisc.edu/~billh/visad.html
- SCoVE at Geogria Tech http://www.gtri.gatech.edu/casestudy/scove
and interestingly enough, this topic has many connections to interactive visualization techniques. I personally find the breadth of this field surprising.
Matt Gedigian - Nov 24, 2008 02:34:41 pm
In collaborative environments, people hone in on specific topic and add metadata. It's a little discouraging that this work often cannot be repurposed. For this reason, I really enjoyed the data-aware annotations. This enables people to do things like interactively querying a map and then add information to the data that matches the query. The benefit of attaching the metadata to the underlying model is that it can be used in other contexts. Minimally, in a variant of the same visualization, but also in general data-mining ways.
@Seth The limitation of the doubly-linked discussion implementation appeared in Jeff's in-class demonstration too. While looking at one view, he observed that a trend for a subpopulation was difficult to see. He then switched to a view that isolated that subpopulation and added a comment. While this was the proper view on which to center the discussion, it meant that other people would have to come to that page. It would be very useful to have a visual indication on the original page, whether the phenomena was difficult to observe. In their discussion of social navigation, they describe ways of leading people to particular views. It seems that these same techniques could be used to steer people towards subviews with discussions.
HeatherDolan - Nov 25, 2008 11:54:08 pm
I really enjoyed Jeff's talk and paper, Voyagers and Voyeurs. It seems really useful to have different people spotting patterns in data and sharing their diverse findings. While this is possible now, sense.us provides a consistent space in which to perform analysis and ensure people are seeing the same (or different) views of the data depending on what the user wants to draw attention to. I also think this environment encourages exploration while visualizations passed around by email (especially considering size, filters etc.) or other means may not.
Thanks for the additional links Douglas.
