Title: Guidance Communication in Mixed-Initiative Visual Analytics

Date: Wednesday, December 7, 2022

Time: 9:00 AM - 11:00 AM EST

Location (in-person + virtual): TSRB 334 (VIS Lab) and Zoom (https://gatech.zoom.us/j/98577182875)

Arpit Narechania (http://narechania.com)

Computer Science Ph.D. Student

School of Interactive Computing

Georgia Institute of Technology

Committee

Dr. Alex Endert (advisor; School of Interactive Computing, Georgia Institute of Technology)

Dr. John Stasko (School of Interactive Computing, Georgia Institute of Technology)

Dr. Duen Horng (Polo) Chau (School of Interactive Computing, Georgia Institute of Technology)

Dr. Clio Andris (School of City and Regional Planning, Georgia Institute of Technology)

Dr. Sham Navathe (School of Computer Science, Georgia Institute of Technology)

Dr. Mennatallah El-Assady (Department of Computer Science, ETH Zurich)

Abstract

Visual Analytics systems have enabled effective understanding, reasoning and decision making on large and complex datasets across a wide variety of domains. These systems combine automated analysis techniques with interactive visualizations, building upon machines’ superior computational and humans’ dominant perceptual and analytic skills. However, these automated actions may be premature or flawed as the machine’s (or system’s) interpretation of the human's (or user’s) intent might be incomplete or incorrect, necessitating them both to engage in an efficient dialog to mutually minimize their “knowledge gap” while also ensuring proper analytic progress.

To actively resolve this “knowledge gap,” a popular approach is to enrich user interfaces with guidance to help the user and/or the system reach their respective goals, the same way we seek guidance from our parents, teachers, or software assistants (e.g., Clippy). The goal of this thesis, thus, is to design guidance-enriched interfaces with a strong focus on facilitating visual communication of appropriate and timely guidance between the user and the system. Towards this goal, we developed multiple visual analytics systems that provide guidance during different stages of a data preparation and analysis pipeline: (i) Debug-It-Yourself (DIY) provides step-by-step explanations for debugging natural language-to-SQL question-answering scenarios; (ii) Lumos utilizes users’ interaction logs to increase awareness of biased analytic behaviors during visual data exploration; (iii) DataPilot utilizes data quality and usage information to help users select effective subsets from large datasets during data preparation; and (iv) DataCockpit, similarly, helps users navigate and discover knowledge from data lakes.

We propose to formalize how mixed-initiative user interfaces, wherein both the user and the system can take analytic initiatives on each other’s behalf, can visually communicate guidance by (i) creating a design space of visual guidance communication (e.g., guidance-enriched control panel widgets or ‘Guide Me’ panels); and (ii) designing a formal model that represents how guidance can be communicated directly through visualization transformations (e.g., notations for zooming into a particular chart region). By implementing aspects of this design space and formal model, we propose to develop a new guidance-enriched visual analytics system prototype to study how user interfaces can adapt themselves in response to different types and amounts of guidance; e.g., depending on their “knowledge gap,” the system can either provide none, one, or multiple options for the user to choose from. The user in response can provide feedback to appropriately steer the system. By making these contributions open-source, this thesis ultimately hopes to enable visualization developers to design their own guidance-enriched user interfaces.