Abstract

Data-intensive models have become critical to understanding the world. In order to reuse or combine datasets to support modeling, scientists must select, understand, and align them manually, a laborious process that requires understanding different domains and formats. To assist the modeling process, we present an unsupervised approach that identifies units in source data and provides a corresponding semantic representation. Then, we provide a method that enables scientists to perform data transformations, such as unit conversions, which are commonly necessary in modeling world systems. Our preliminary results demonstrate that our method can be used to automatically capture and transform units over spreadsheets achieving an F1-score of 0.48 in unit detection and parsing, and an accuracy of 62% in the semantic representation and transformation.

Date

May 14, 2019

Authors

Basel Shbita, Arunkumar Rajendran, Jay Pujara, Craig A Knoblock

Journal

Modeling the World’s Systems

Pages

7