Smarter Decisions through Interactive Visual Analytics
The value of visualization as a means of understanding large amounts of data is well accepted in today’s business world. Faced with ever increasing amounts of information, the average business user needs to incorporate insights gleaned from such information into his/her daily decision process. However, existing analytical tools provide little support to the user who has extensive understanding of a given data domain but has limited skills when it comes to effectively visualizing that data. Moreover, these tools offer little support for collaboration in the decision making process, or for automatically tracking that process in order to increase transparency and provide a "corporate memory" of such decisions.
Smarter Decisions is an interactive and intuitive Web-based tool for visual analytics that helps everyday business users derive insights from large data sets. It increases the consumability of visual analytics by enabling users who are not visualization or computer experts to upload, access, and visually interact with both structured (e.g. relational database, spreadsheets) and unstructured (e.g., paragraphs of text, blogs, articles) information. It also helps businesses reach smarter decisions by automatically capturing trails of the analytic process. Such trails can be bookmarked, shared, and restored to understand the provenance of such decisions, as well as to facilitate collaborative decision making and expertise transfer.
Smarter Decision's main user interface has four main areas. The query panel(a) allows the user to dynamically build ad-hoc queries using select-lists to retrieve the data. The visualization canvas(b) displays the automatically instantiated visualization of the retrieved data. Alternate visualization choices are provided in the visual recommendation panel (c). Users can switch to any of these alternates simply by clicking on them. The history panel(d-j) contains the current analytic trail, which consists of the sequence of user actions during the user's current investigational thread, is displayed at the bottom of the interface. Each step (e) in the trail consists of a semantic user action defined in the trail model (e.g. Query, Filter), and the visualization displayed as a result of the action. Hovering the mouse over a step shows a small thumbnail of the associated visualization for the step (d). Clicking on a step reveals information about the action performed during the step in the form of parameter name-value pairs (g), and a menu of the operations that can be performed in this step (f). The single-step undo button (h) is used to remove the last step. The snapshot button (i) allows the user to export the current visualization to an image that can be embedded in reports and presentations. The bookmark button (j) enables the user to save the sequence of actions included in the current trail. A unique URL is assigned to each bookmarked trail, which is used for sharing and restoring the trail.
Dynamic visual recommendation (Visual Recommender)
Choosing the most effective visualization for a particular task and a particular kind of data requires a significant level of visual literacy that is often beyond the typical user. Moreover, users may not know exactly what the data "looks like" ahead of time, and may select inappropriate visualizations. Furthermore, after choosing a particular visualization type, the user still needs to map the data to the visualization parameters (e.g., axes, colors, etc.). The complexity of this task often leads to the selection of sub-optimal visualizations, hence reducing the task accuracy and increasing its completion time. To address this problem, we have developed a rule-based Visual Recommender that assists users by automatically showing the data into the "best" visualization given the structure of the data. We have also developed a behavior-driven Visual Recommender that attempts to detect the user's task (based on an analysis of the steps taken) and recommends a visualization if the current one is sub-optimal.
Automatic capture of semantic actions (Analytic Trails)
Decision making is a complex, data-driven process that involves the exploration of several lines of inquiry, along which users gather nuggets of useful information. These individual insights are then combined together to reach a final decision. Capturing the provenance of such a decision is difficult. Requiring users to manually document this process is onerous and distracting, as well as prone to errors and omission. Smarter Decisions transparently captures the meaningful steps of the analytic process (e.g. filter, query, change-view) as the user works. These steps are stored in what we call an Analytic Trail. Because trails can be stored and shared, they allow teams to collaborate in decision making, even if team members are separated by time and distance. Trails also create a "corporate memory" of the decisions that were made, so that they can be rolled back at any time to view the data, the person, and the discussions associated with a given decision. Non-editable Analytic Trails support governance and compliance in decision making (who made the decision and why) whereas editable Analytic Trails (decision templates) create a model that supports the use and sharing of best practices. Enterprises can create a library of Trails, thus increasing both the "stickiness" of the tool and the value to the enterprise.