Sora Kanosue

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2024-92

May 10, 2024

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-92.pdf

Domain experts often find themselves performing repetitive tasks restructuring large amounts of data. General-purpose data analysis tools like Tableau can help experts accomplish these tasks, but they have steep learning curves, and their flexibility and open-ended interaction model can be overwhelming for users. Task-specific interfaces can circumvent this issue by guiding users through each phase of a data transformation task, but authoring bespoke tools for each task is time-consuming and difficult. In this paper, we present HiLT, a domain-specific language embedded in Python which facilitates the creation of task-specific data transformation GUIs. Programs written in HiLT generate human-in-the-loop data transformation GUIs which walk users through the process of a given data transformation task. We conducted a formative user study with 17 participants who were tasked with constructing data transformation interfaces using HiLT and other existing frameworks in order to explore HiLT's learnability and usability relative to other tools in the same space. Our findings from the formative study suggest directions for future tools in this space.

Advisors: Sarah Chasins

---

BibTeX citation:

@mastersthesis{Kanosue:EECS-2024-92,
    Author= {Kanosue, Sora},
    Title= {HiLT: A Framework for Generating Human-in-the-Loop Data Transformation GUIs},
    School= {EECS Department, University of California, Berkeley},
    Year= {2024},
    Month= {May},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-92.html},
    Number= {UCB/EECS-2024-92},
    Abstract= {Domain experts often find themselves performing repetitive tasks restructuring large amounts of data. General-purpose data analysis tools like Tableau can help experts accomplish these tasks, but they have steep learning curves, and their flexibility and open-ended interaction model can be overwhelming for users. Task-specific interfaces can circumvent this issue by guiding users through each phase of a data transformation task, but authoring bespoke tools for each task is time-consuming and difficult. In this paper, we present HiLT, a domain-specific language embedded in Python which facilitates the creation of task-specific data transformation GUIs. Programs written in HiLT generate human-in-the-loop data transformation GUIs which walk users through the process of a given data transformation task. We conducted a formative user study with 17 participants who were tasked with constructing data transformation interfaces using HiLT and other existing frameworks in order to explore HiLT's learnability and usability relative to other tools in the same space. Our findings from the formative study suggest directions for future tools in this space.},
}

EndNote citation:

%0 Thesis
%A Kanosue, Sora 
%T HiLT: A Framework for Generating Human-in-the-Loop Data Transformation GUIs
%I EECS Department, University of California, Berkeley
%D 2024
%8 May 10
%@ UCB/EECS-2024-92
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-92.html
%F Kanosue:EECS-2024-92