A two-stage framework for designing visual analytics systems to augment organizational analytical processes.

摘要

A perennially interesting research topic in the field of visual analytics is how to effectively develop systems that support organizational knowledge worker's decision-making and reasoning processes. The primary objective of a visual analytic system is to facilitate analytical reasoning and discovery of insights through interactive visual interfaces. It also enables the transfer of capability and expertise from where it resides to where it is needed---across individuals, and organizations as necessary.;The problem is, however, most domain analytical practices generally vary from organizations to organizations. This leads to the diversified design of visual analytics systems in incorporating domain analytical processes, making it difficult to generalize the success from one domain to another. Exacerbating this problem is the dearth of general models of analytical workflows available to enable such timely and effective designs.;To alleviate these problems, this dissertation presents a two-stage framework for informing the design of a visual analytics system. This two-stage design framework builds upon and extends current practices pertaining to analytical workflow and focuses, in particular, on investigating its effect on the design of visual analytics systems for organizational environments. It aims to empower organizations with more systematic and purposeful information analyses through modeling the domain users' reasoning processes.;The first stage in this framework is an Observation and Designing stage, in which a visual analytic system is designed and implemented to abstract and encapsulate general organizational analytical processes, through extensive collaboration with domain users. The second stage is the User-centric Refinement stage, which aims at interactively enriching and refining the already encapsulated domain analysis process based on understanding user's intentions through analyzing their task behavior. To implement this framework in the process of designing a visual analytics system, this dissertation proposes four general design recommendations that, when followed, empower such systems to bring the users closer to the center of their analytical processes.;This dissertation makes three primary contributions: first, it presents a general characterization of the analytical workflow in organizational environments. This characterization fills in the blank of the current lack of such an analytical model and further represents a set of domain analytical tasks that are commonly applicable to various organizations. Secondly, this dissertation describes a two-stage framework for facilitating the domain users' workflows through integrating their analytical models into interactive visual analytics systems. Finally, this dissertation presents recommendations and suggestions on enriching and refining domain analysis through capturing and analyzing knowledge workers' analysis processes.;To exemplify the generalizability of these design recommendations, this dissertation presents three visual analytics systems that are developed following the proposed recommendations, including Taste for Xerox Corporation, OpsVis for Microsoft, and IRSV for the U.S. Department of Transportation. All of these systems are deployed to domain knowledge workers and are adopted for their analytical practices. Extensive empirical evaluations are further conducted to demonstrate efficacy of these systems in facilitating domain analytical processes.