Natural Language Interfaces for Data Analytics
Name
Wellens-qwellens-meng-eecs-2021-thesis.pdf
Description
Thesis PDF
Size
3 MB
Format
Adobe PDF
Checksum (MD5)
7bb7c802958eda414eb392663d86562a
Author(s)
Wellens, Quentin
Advisor(s)
Kraska, Tim
Date Issued
June 2021
Publisher
Massachusetts Institute of Technology
Abstract
As more processes become data-driven, anyone should be able to gather insights into databases without needing to develop complex computer skills typically required for data analytics software. We propose to design new paradigms in which users rely on their own natural language to analyze and visualize data. To that end, we develop three different approaches (unsupervised, rule-based, and supervised) to infer formal specifications from natural language utterances. Contrary to most other work, we developed these approaches in a low-resource environment using synthetically generated training sets, rather than expensive and labor-intensive expert annotations or crowd-sourced examples. Finally, we conducted a study to compare our proposed paradigm to drag-and-drop mechanisms. Not only does our best-performing model, Alcurve, achieve an 86.3% test accuracy on real user input, it also enables users to be 30% more productive when solving analytical tasks, which further highlights the important improvements in usability language-based interfaces can provide.
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Terms of Use
In Copyright - Educational Use Permitted
Copyright MIT
Persistent DSpace Link
