Supporting Effective Common Ground Construction in Asynchronous Collaborative Visual Analytics Yang Chen * University of North Carolina at Charlotte Jamal Alsakran † Kent State University Scott Barlowe ‡ University of North Carolina at Charlotte Jing Yang § University of North Carolina at Charlotte Ye Zhao ¶ Kent State University ABSTRACT Asynchronous Collaborative Visual Analytics (ACVA) leverages group sensemaking by releasing the constraints on when, where, and who works collaboratively. A significant task to be addressed before ACVA can reach its full potential is effective common ground construction, namely the process in which users evaluate insights from individual work to develop a shared understanding of insights and collectively pool them. This is challenging due to the lack of instant communication and scale of collaboration in ACVA. We propose a novel visual analytics approach that automat- ically gathers, organizes, and summarizes insights to form common ground with reduced human effort. The rich set of visualization and interaction techniques provided in our approach allows users to ef- fectively and flexibly control the common ground construction and review, explore, and compare insights in detail. A working proto- type of the approach has been implemented. We have conducted a case study and a user study to demonstrate its effectiveness. Keywords: Visual analytics, asynchronous collaboration, insight, multidimensional visualization. Index Terms: H.5.3 [Group and Organization Interfaces]: Collab- orative computing—Web-based interaction; 1 I NTRODUCTION With the growth of public web-based visualization communities (e.g., Many Eyes [26] and sense.us [16]), people can collabora- tively analyze data outside the constraints of time and space. Asyn- chronous Collaborative Visual Analytics (ACVA) allows shared ac- cess to resources, such as expertise and datasets by releasing the constraints on when, where, and who works together. ACVA turns collaborative visual analytics into a social process where everyone can participate and thus makes it possible to analyze datasets of much larger scales [16]. In this paper, we focus on common ground construction in ACVA. In social and psychological research, common ground is defined as the shared understanding enabling communication be- tween conversational participants [11]. In collaborative visual an- alytics, common ground construction refers to the visual analytic process, in which users evaluate work that may have been created individually, to develop a shared understanding of data with a col- lection of insights and hypotheses [19]. Effective common ground construction may minimize the need to verbally confirm actions among collaborators, reduce the cost of collaborative effort [11]. * e-mail: ychen61@uncc.edu † e-mail: jalsakra@cs.kent.edu ‡ e-mail: sabarlow@uncc.edu § e-mail: jyang13@uncc.edu ¶ e-mail: zhao@cs.kent.edu In asynchronous settings, non-verbal cues for common ground is especially important since verbal communications between the col- laborators are usually difficult or even impossible. ACVA users face significant challenges in common ground con- struction. The scale of ACVA is usually larger than Collocated Collaborative Visual Analytics (CCVA). For example, Many Eyes received over 1463 registered users, 2100 datasets, and 450 users’ comments in its first two months of life [26]. Browsing and organiz- ing such a large amount of information from such diverse users and datasets are challenging tasks. To make it worse, there lacks instant communication among ACVA users. It is difficult for them to col- laboratively identify significant insights and capture relationships among insights through face to face discussion and direct manipu- lation as in CCVA[19]. Thus, there is an urgent need for effective and efficient visual analytics tools for ACVA common ground con- struction, especially for the following tasks: Task 1 Generating Overview: Common ground construction usually starts from forming an overview of the insights that have been recorded and gathered [19]. The overview presents the over- all structure, key aspects, and evolution of the insights to help users gauge the context and determine future direction [5]. Exist- ing ACVA systems provide limited capability with manual insight browsing, inspection, and organization, which hinders users’ effort on quickly forming a mental map of existing insights. New ap- proaches for overview generation that satisfy the following require- ments must be developed: • Collecting information effectively and efficiently: Rich se- mantic information about insights is needed for automated insight organization, retrieval, and association according to varying user interests. Collecting such information should not impose extra burden to users, i.e., their ongoing visual explo- ration process should not be disturbed or diverted. • Employing automatic insight analysis: Manual insight as- sociation and grouping are not realistic for a fast growing pool of ACVA insights. Automatic insight analysis techniques, such as automated insight correlation, clustering, and summa- rization, are direly needed to be developed and integrated into ACVA systems for fast and operative overview generation. • Supporting dynamic overview construction: ACVA users usually have diverse information needs. Dynamic overview construction should be supported in ACVA systems so that users can explore the insight space according to their specific needs. Moreover, the system should allow the users to dy- namically manipulate visualization results according to their changing interests and developing understanding. • Providing a rich set of views and interactions: Multiple co- ordinated views should be provided to allow users to examine insights from different aspects. For example, Temporal visu- alization helps users track and employ temporal evolution of insights, so that they can keep awareness of timing and pre- serve historical contents of insights [15, 19]. Furthermore,