MITEXTEXPLORER: Linked brushing and mutual information for exploratory text data analysis

Abstract

In this paper I describe a preliminary experimental system, MITEXTEXPLORER, for textual linked brushing, which allows an analyst to interactively explore statistical relationships between (1) terms, and (2) document metadata (covariates). An analyst can graphically select documents embedded in a temporal, spatial, or other continuous space, and the tool reports terms with strong statistical associations for the region. The user can then drill down to specific term and term groupings, viewing further associations, and see how terms are used in context. The goal is to rapidly compare language usage across interesting document covariates. I illustrate examples of using the tool on several datasets: geo-located Twitter messages, presidential State of the Union addresses, the ACL Anthology, and the King James Bible.

Extracted Key Phrases

6 Figures and Tables

Cite this paper

@inproceedings{OConnor2014MITEXTEXPLORERLB, title={MITEXTEXPLORER: Linked brushing and mutual information for exploratory text data analysis}, author={Brendan T. O’Connor}, year={2014} }