Lecture on Nov 30, 2010. (Slides)


  • Required

    • Voyagers and Voyeurs: Supporting Asynchronous Collaborative Information Visualization. Heer, Viégas, & Wattenberg. (html)

    • Designing for Social Data Analysis. Wattenberg & Kriss. (pdf)

  • Optional

    • Design Considerations for Collaborative Visual Analytics. Heer and Agrawala. (html)

    • Many Eyes: A Site for Visualization at Internet Scale. Viégas et al. (pdf)


mattbush wrote:

The four quadrants of the spectator interface taxonomy, referenced in Wattenberg & Kriss on p. 554, are an interesting concept I wish we spent more time on. It seems that most types of effective data visualization fall into the "expressive" quadrant, but there are many use cases for "magical", such as PowerPoint, and maybe even situations where suspenseful and secretive might be the best way to go, or where switching between these quadrants effectively is part of the interface.

msavva wrote:

I found the Wattenberg and Kriss paper to be a really insightful analysis of why NameVoyager turned out to be such a success story. It seems that the most important factor in this success was that many people could relate to the data and have fun or feel a sense of achievement while exploring it. This is connected to an interesting point made by the authors that you need to have a shared knowledge background with social implications to enable such highly successful collaborative social analysis. I imagine this is harder to get for data-sets in the "non-social" sciences, but then again, most interesting data-sets have implications at least for particular demographics concerned with what the data may be saying. One could design visualizations that would leverage these implications explicitly, such as for example, tailoring genome sequence visualizations to allow exploration of personal backgrounds and genomic history through hereditary lines. This is probably a factor in the success of personal genomics companies such as "23andMe". I also found the classification of users into game player roles to be simultaneously very amusing and effective at explaining behavior patterns. It would be cool if visualizations that tailor to these usage patterns resulted in many more data analysis tasks becoming hugely popular.

skairam wrote:

Regarding the question of how to design systems to facilitate social data analysis, it seems as if one of the major factors is balancing the benefits of incorporating multiple sources of knowledge and interpretation with the potential costs of "Groupthink" or losing diversity due to influence among the group.

In The Wisdom of Crowds, Surowiecki notes that a key element for successful analysis by crowds is maintaining independence of opinions. As Jeff noted in class, one successful strategy is to allow for individual exploration first and then to integrate these insights in a collaborative environment.

I wonder if there are particular design principles which could be utilized in social data analysis systems such as Sense.us or ManyEyes to most effectively balance interpretation and integration.

nandu wrote:

This is regarding the comparison of views of specific data situations versus interaction via text commenting or annotation. There is an interesting parallel that we can draw to web analytics here. To measure and analyze usage patterns on websites, two important metrics that are used often are page-views and clicks. There are far fewer clicks than views and usually clicking involves more interactive intent on the users behalf. Both are tracked carefully with certain kinds of web entities (like text ads) only paying out when there is interaction (a click) while other kinds of web entities (like branding banner ads) consider a wider reach, via a large number of views, more significant. Another point worth noting is here is that some people also measure and use the ratio between these two - the click-through-rate - as a measure characterizing the effectiveness of the web entity.

yanzhudu wrote:

It is interesting to see social data analysis happens around static data. This type of activities often requires user to bring in outside data source into discussion. It would be really useful if there is a way to integrate (hyperlink) datasets across different source. This allows users to build a more comprehensive connected dataset through discussion.

It would also be interesting to allow user to build up dataset in wikipedia style, and investigate effectiveness of this type of data collection.

nchen11 wrote:

I found it to be slightly amusing (but not all that surprising) that people tended to rate their own annotations as being more helpful than other people's annotations (p. 9 of Voyagers & Voyeurs).

I do think that scented information is quite valuable, since it combines both the statistics about the data itself and the way that others view the data. (As opposed to when the data is presented separately from queries about the data, such as with the English literature text frequency explorer that was shown in class. I don't remember what it was called, but Heer did mention how the creator also had a separate visualization showing the frequency of search queries on his site.)

If the ultimate goal is social collaboration, then data about other users could only serve to facilitate said collaboration, in my opinion.

Also, I wonder how many people were prompted to make queries in the NameVoyager after reading example comments in the Wattenberg & Kriss paper . . .

jbastien wrote:

As I said in class I'm very interested in social analysis of dynamically updating data and how the analysis feeds back into the dynamic data. The example I gave is benchmarking of software where the analysis points out shortcomings or bugs in the software, which then get fixed and in turn updates the data, driving more analysis.

Jeffrey pointed out that this field was still very green and that the evolving nature of the data was a problem, especially in mapping analysis properly to the corresponding data.

abhatta1 wrote:

As Jeff mentioned sometimes the number of views and the number of comments often help users in browsing through topics. Other factors which might influence such decisions are the recent comments (i.e. the timeline), the number of commenting users on a topic, the number of annotations (changes to the visualization) made by commenting users etc. In fact, I felt that vizualisation of the annotation history might be an interesting topic in itself as it will help users to make effective choices in selecting interesting topics

wulabs wrote:

The namevoyager tool was a great example of a tool used within a means of both entertainment and utility. In today's world there are many such social tools that can be found on iPhones or in the form of Facebook applications. With so many interconnected people in an 'always wired' phase, it is no wonder the new 'web 3.0' will be built on top of this social platform. So far nobody has really tried to enhance visualization on the cloud but this will probably be more easily done once HTML5 has become more stable

acravens wrote:

The examples in these papers and discussed in class focus primarily on "asynchronous collaboration." For web 2.0/web 3.0 type applications used by virtual communities who probably will never meet, this makes sense. However, the applications I'm most interested in are the ones where there is a combination of using visualization when people are together in the same room (e.g. periodic meetings) and asynchronous collaboration or individual analysis (e.g. preparing for meetings, reviewing group aggregations or proposals or decisions, etc.). I've been thinking since yesterday's lecture about how you might change the design of a visualization application when it's intended to be used in both of theses settings. For instance, it seems like it would be useful to have more ability to review comments and for an editor/group leader to clear out cycles of analysis. (This might relate to the dynamic analysis, @jbastien). Any other thoughts?

amirg wrote:

I think using information scent as a guide to let users find views that are likely to have comments is a great idea. It gives great clues into where you are likely to find interesting results without really impeding individual exploration of the data set. This way a user can see what others have seen and discussed, while also knowing what is completely new.

Also, an interesting link to more info by Jakob Nielsen about information foraging and scent and why they are important for design: http://www.useit.com/alertbox/20030630.html

On a totally different note, with respect to the question posed in class about what factors enable viable collaborations, I think sense.us nailed a lot of these, but they really depend on the type of collaboration you are trying to enable (whether it's different place vs same place and different time vs same time). One of the big ones in my opinion is making sure that two people in different places at different times see the same thing, so they are not working from different views of the data. I think another important aspect of facilitating collaboration across the web is trust of other users both in the sense of providing accurate information and providing relevant information. I think the notion of trust could be expanded for sense.us. For example, trust is huge on Wikipedia, where every statement must be backed by a source, but of course this also makes the collaboration slightly more burdensome, so there is a big tradeoff here.

andreaz wrote:

I think that another strategy to promote high-quality contributions from users is to risk their reputation. Sense.us accomplished this by attaching the full name of the user to the contribution and testing within a closed facility. The full name of the user attaches their reputation to the contribution, so users are more likely to behave appropriately. In contrast, Swivel allows users to hide behind a handle of their choice, and this anonymity gives them full freedom to behave however they please without risking their real life reputation. Testing within the closed facility of IBM forced users to interact with peers familiar with them, and this familiarity further increased the user's risk of ruining their reputation with a bad-quality contribution.

I also think that emphasizing the "road-less-traveled" is an incredibly useful feature for allowing the user to make new discoveries within the data. As mentioned in lecture, sense.us accomplishes this through the information scents--indirectly though, since a term lacking a scent cues the user that the term is uncharted territory. Wattenberg employs this feature in BookVoyager in a more direct manner through the "color by history" feature, which explicitly draws attention to the unvisited information by graying out the previously visited series.

estrat wrote:

I wonder what sorts of datesets/visualizations lend themselves to collaborative analysis. I see tons of visualizations every day but most of them are not collaborative. Is that because they don't need to be or because it's too much work? Is the idea simply to "crowdsource" analysis of large datasets? I tend to be critical of recent attempts to socialize everything, but maybe it makes sense for this.

jsnation wrote:

I really like some of the collaborative ideas implemented in sense.us like the use of tying comments to states, and allowing the discussions to reference the visual state and add to it with markers and other items to aid in the discussion. I think that for certain visualizations, having a real-time system for collaboration could be really useful. Kind of how google docs is for shared editing, where you can see where and what different people are doing. When we were talking in class about the 'scented' UI page, I was thinking that it could also be cool to scent based on how many people are currently viewing the page, so you can jump into an active discussion on something. Also I think more ways to interact with the commentary vs. the visualization, by searching comments or having a feed of the most active or most recent ones could be good.

hyatt4 wrote:

I am intrigued by @andreaz's comments regarding the influence of collaboration because of knowing someone's personal data. This is certainly helpful in ensuring people play nice, but I could see a potential downside if people are intimidated or would not like to honestly express a view because they feel it may make them look foolish or may not be appreciated by a higher up who can fire them (worst case). That being said, I don't think letting everyone post in an anarchy type fashion is the way to go either (e.g. recall the "pot pot pot" collaborations). The first paper pointed out the voyeur aspect and impact, and I think there should remain some degree of secrecy/anonymity for individuals but with some way to weed out or vote out less productive contributors. I think some sort of peer ranking system would be helpful in this regard where a person could eventually lose their privilege of posting, or perhaps even be exposed if that was not enough (e.g. repeat offenders).

adh15 wrote:

One variation on collaboration that would be interesting to explore would be to create a (possibly limited) channel for real-time communication between simultaneous visitors to a web-based visualization for the purpose of facilitating their collaborative exploration of the data. Also, it could be interesting to add game-play elements to the interface and to group the visitors into teams that can earn points or achievements by providing interesting insight into the data or by discovering the answer to a pre-defined question. Such pre-defined questions might even spur further inquiry and be a way of exposing a more general audience to various techniques for data analysis.

mariasan wrote:

I would love to see applications like the job voyager and its likes to move into a space where users could generate content for articles, reports, or maybe even research publications. These tools are great for exploring data and lends themselves well for discovery, it would be interesting to think about how to take the communication part to the next level. When applied to a "serious" data set, I see no reason why generated visualizations wouldn't be good enough to publish.

iofir wrote:

I think that there are essentially three aspects that make collaborative exploration possible. beyond those three main aspects there are other improvements that are convenient but not essential. the main three are: conversation, referencing and illustrating. (these also apply for collaboration in person, not only online) In the online case, conversation comes in the form of text messages and comments. communication is the MOST important part of collaboration. referencing comes in the form of links to other views or perspectives. and illustrating comes in the form of arrows, highlighting and circling different parts of the data.

there are a lot of other ways to improve these aspects, but all three are regularly used in collaborative conversations. (whether in person or online.)

jtamayo wrote:

Seems interesting to think what features of a visualization are helpful for individual exploration but break when using a visualization in a social context, and viceversa.

Animations and interactivity, for example, don't seem to add much in a conversation. Labels and annotations, on the other hand, are almost mandatory in a social context; when exploring a visualization individually they're rarely useful.

asindhu wrote:

Most of the lecture focused on collaboration at the level of "sensemaking." I found the one slide showing the many levels in which people can contribute to a visualization to be very interesting because it highlights that collaboration and shared contributions can happen not just at the highest level of abstraction (called "visual analytics" in the slide) but also in areas like actually contributing raw data or organizing the data, designing the visual encodings, etc. While models of contribution seem very well hashed out at the top level with such things as comments and forums, it's much less obvious how one might facilitate contributions at lower levels. I would be very interested to see if anyone has done any work in building tools to help people collaborate in managing raw data, designing visualizations, or other low-level tasks.

rakasaka wrote:

The design space of web-based collaboration provides much to be excited about and even further more to be afraid of, and I think sense.us and manyeyes have done remarkably well to provide some guidance while allowing individual exploration to occur. I really believe that data visualization comes alive when it takes place alongside storytellling because the story-teller doesn't necessarily become the authority of the interpretation, merely the mouthpiece of it. In doing so the listener feels equally empowered to challenge that interpretation with one of their own. Lastly, I am somewhat disappointed that sense.us never quite made it to the general public - I can only imagine the discussions it would have spurred!

sholbert wrote:

Regardless of the validity of commentary on a visualization, I feel that people are more likely to believe any kind of collaborative comments. The visualization provides a platform for proof that anyone can build off of with their comments, and it's easy to agree with theories for causes of trends, but these correlations might just be associative. If the majority of comments support some causation trend, people are more likely to just believe any theories, as valid or speculative as they might be.

arievans wrote:

What I found most compelling about the lecture was the example of the voyager visualization with the option to save an image stream to convey a story. Along with the annotations, this is an extremely powerful way to convey and discuss data with the aid of very specific views. Users are also granted the freedom to go to any view and modify it as well.

I think I am most impressed because this setup seems to get at what I think is the most difficult part about collaboration, period. It's the whole "I want to see what you are seeing" problem. I find myself using screen sharing programs more often with time because I find that merely looking at the same info alleviates so many other communication problems that arise otherwise. With the voyager example, we get exactly the same thing, so it is no wonder that the user base has taken so well to utilizing it. Ultimately I'd like to see this concept applied elsewhere on the web: tweak data, come up with a view sequence, and share it with comments and annotations.

selassid wrote:

I thought it was interesting that none of the case studies we looked at really let users collaborate on the visual encoding itself to the extent that I would have expected. Sure simple things like zooming and selection are crucial, but it seemed like there would be a lot to gain from allowing collaboration on the encoding instead of just the view. This seems like it would be a hard problem to effectively tackle while still allowing disparate users to collaborate without confusion. This almost requires the difficult question of how do you let laypeople easily program complex visualizations to be a solved problem, too. This area of the field is obviously pretty fledgling. It's interesting that it's the sort of area where you can't really do too much laboratory research beforehand because of it's social nature.

ankitak wrote:

I found the paper by Wattenberg to be a very interesting exploration of what made NameVoyager a success. In particular, I was highly intrigued by Bartle's categorization of the denizens of online multiplayer environments. Moreover, Wattenberg provided a good study for the design considerations for social data analysis. However, his hypotheses can work only in cases where the data under consideration is not the user's personal data which they might not be interested in sharing with others.

esegel wrote:

Working in finance, it is very common to send around a chart via email and have a long email chain of people commenting on the data. Do products exist that facilitate this interaction?

sklesser wrote:

I'm really intrigued by the idea of crowd-sourcing such as NASA ClickWorkers, I think there may be interesting opportunities for less menial tasks given the right tools. With proper collaboration tools amateur groups can come together to solve problems difficult for computers or too time-consuming for experts. A good example of this is the foldit protein folding game. Once users get to a high enough level they are given very difficult proteins to solve and can talk to each other in a live chat window to discuss it.

Leave a comment