A second approach, suggested by many users, would be to show commentary related, though not directly attached to, the current view. Requested features include showing comments from other views that contain links to the current view (“trackbacks”), and related commentary on “nearby” or “similar” views. The latter could help alleviate cross talk. Along these lines, there are appealing possibilities for generalizing the notion of view indexing, for example, suggesting conversations on views deemed semantically similar to the current view. This would require an index of visualization state providing not just equality comparisons, but distance measures. Such a retrieval model might be used to provide additional benefits, such as general searchability and data-aware auto-complete mechanisms.

Users have also suggested using visitation data or explicit ratings of “interestingness” to suggest views of potential interest. Others suggested supporting keyword tagging of comments²² and mining usage data. For example, both manual and automated tagging of questions or other action items could be used to help direct collaborative effort.

The scope of comment visibility is a larger issue that affects all discussion models. What happens when the amount of discussion becomes untenably large, or users don’t want their activity exposed to everyone? The ability to form groups and limit comment visibility to group members is one means requested by users to support privacy and make discussion-following both more relevant and tractable.

Although individual usage varied substantially, most lab study users ( 87.5%) did use the bookmark trails, which proved essential for comments that included multiple views. Multiple users remarked on the usefulness of the bookmark trails and wanted to more easily share trails as first class objects. At times, users were frustrated when following multiple links in a comment, as the original comment would disappear when a new view was loaded, requiring use of the back button to perform “ hub-and-spoke” browsing. In response, users suggested adding a dedicated “presentation” mode to facilitate tours and storytelling.

Finally, the graphical annotations saw significant usage, despite mixed reactions from users. Though they were used for pointing, many users did not find them necessary for dis-ambiguation. We expect that the value of annotations varies significantly depending on the type of visualization being referenced. Regardless, annotations were used regularly for pointing and sometimes for socializing.

If the free-form annotations prove helpful, a second challenge would be to extend them to cover dynamic or evolving data sets. The decoupled nature of geometric annotations can prove problematic when the underlying data changes. Similar problems have been investigated in the context of document annotation. ⁵ More recent work¹⁹ has explored “data-aware” annotations that translate user selections into declarative queries over the underlying data, allowing annotations to be applied to time-varying data and different visual encodings.

6. 2. communities and data

Since the original sense.us experiment, there have been several new examples of systems that support conversation around data. Web sites such as Swivel.com have provided social-network-style platforms for conversation around data, along with basic charting capabilities. Tableau Software launched its Tableau Server product, which (much like Spotfire’s DecisionSite Posters) allows users to collaborate asynchronously around intranet-based visualizations. Little has been published about usage of these systems, however.

One new system where results have been reported is the Many Eyes Web site. ²⁶ In contrast to sense.us, or tools like Tableau or Spotfire, Many Eyes is freely available on the public internet and allows users to upload their own data. Unlike data-oriented sites like Swivel, Many Eyes lets users apply more than a dozen interactive visualization techniques. They may then have discussions around visualizations, though annotation capabilities are more basic than in sense.us. The experiences on the site²⁶ lend support to the idea that visualization can catalyze discussion. While these discussions can be analytical, they also can be purely social, partisan, or game-like. In addition, the move from a closed setting to the public internet has made clear that these discussions can be highly distributed, ¹³ with a significant proportion of collaboration occurring (via hyperlinks) off the site. Designing for this type of multisite conversation suggests a whole new set of challenges.

7. conclusion

In this article, we investigated mechanisms supporting asynchronous collaboration around interactive information visualization, seeking to more tightly tie the perceptual and cognitive benefits of visualization to social processes of sensemaking. To do so, we implemented a collaborative data visualization site, sense.us. We then observed usage of the site in order to better understand the social dynamics surrounding collective use of visualizations as well as the efficacy of the particular features.

The features of the site—doubly linked discussions, bookmark trails, geometric annotations, and comment listings—were all exploited by users. The doubly linked discussions successfully enabled users to fluidly transfer attention between visualization and commentary and we suggested ways to further improve this type of discussion. Bookmark trails and geometric annotations were also well used, enabling tours through multiple views and pointing to items of interest, respectively. Finally, users played the roles of both voyager and voyeur, alternating between data-driven exploration directly within the visualization and social navigation through comment listings and user profiles to discover new views of interest.

Overall, we believe these results show the value of focusing on the social aspects of visual analysis. Our user studies indicate that combining conversation and visual data analysis can help people explore a data set both broadly and deeply. From a design perspective, there lies a promising opportunity for exploring new widgets and modes of interaction aimed at enhancing collaboration.