2016_acl_doclabel.tex

\documentclass[11pt]{article}
\usepackage{style/acl2016}
\usepackage{titlesec}
\usepackage{graphicx}
\usepackage{amsmath}
%\usepackage{subfig}
\usepackage{times}
\usepackage{url}
\usepackage{latexsym}
\usepackage{breakurl}
\usepackage{subcaption}

\makeatletter
\newcommand{\@BIBLABEL}{\@emptybiblabel}
\newcommand{\@emptybiblabel}[1]{}
\makeatother
\def\imagebox#1#2{\vtop to #1{\null\hbox{#2}\vfill}}

\usepackage[bookmarks=false,
pdfpagelabels=false,
hyperfootnotes=false,
hyperindex=false,
pageanchor=false,
]{hyperref}

\makeatletter
\let\saved@hyper@linkurl\hyper@linkurl

\let\saved@hyper@link@\hyper@link@
\AtBeginDocument{
  
  
  \NoHyper
  
  
  
}



\aclfinalcopy 






\newif\ifcomment\commentfalse
\input{style/preamble}
\newcommand{\name}[0]{\abr{alto}}


\addtolength\titlebox{.5in}

\newcommand{\sect}[1]{\label{sec:#1} \input{sections/#1}}

\newcommand\BibTeX{B{\sc ib}\TeX}
\title{ALTO: Active Learning with Topic Overviews for Speeding Label Induction and Document Labeling}
\author{Forough Poursabzi-Sangdeh \\
              Computer Science \\
	     University of Colorado\\
	    \smalltt{forough.poursabzisangdeh@colorado.edu} 
	\And
	Jordan Boyd-Graber \\
              Computer Science \\
	     University of Colorado\\
	    \smalltt{Jordan.Boyd.Graber@colorado.edu}\protect
	 \AND
	 Leah Findlater\\
         iSchool and \abr{umiacs} \\
	 University of Maryland\\
	\smalltt{leahkf@umd.edu}
	\And
	Kevin Seppi \\
         Computer Science \\
	 Brigham Young University\\
	\smalltt{kseppi@cs.byu.edu}
	}



\begin{document}

\maketitle

\begin{abstract}
  Effective text classification requires experts to annotate data with
  labels; these training data are time-consuming and expensive to
  obtain. If you know what labels you want, active learning can reduce
  the number of labeled documents needed.  However, establishing the label
  set remains difficult. Annotators often lack the global knowledge
  needed to induce a label set.  We introduce \name{}: Active Learning
  with Topic Overviews, an interactive system to help humans annotate
  documents: topic models provide a global overview of what labels to
  create and active learning directs them to the right documents to
  label.  Our user study with forty annotators shows that while active
  learning by itself is best in extremely resource limited conditions,
  topic models (even by themselves) lead to better label sets, and
  \name{}'s combination is best overall.
\end{abstract}

\section{Introduction}
\label{sec:introduction} \input{2016_acl_doclabel/sections/introduction}

\section{\name{}: Active Learning with Topic Overviews}
\label{sec:ALTO} \input{2016_acl_doclabel/sections/alto}

\section{Study Conditions}
\label{sec:conditions} \input{2016_acl_doclabel/sections/conditions}

\section{Data and Evaluation Metrics}
\label{sec:data_metrics} \input{2016_acl_doclabel/sections/data_metrics}

\section{Synthetic Experiments}
\label{sec:synthetic_exp}\input{2016_acl_doclabel/sections/synthetic_exp}

\section{User Study}
\label{sec:user_exp_results}\input{2016_acl_doclabel/sections/user_exp_results}

\section{Related Work}
\label{sec:related_work}\input{2016_acl_doclabel/sections/related_work}

\section{Conclusion and Future Work}
\label{sec:conclusion}\input{2016_acl_doclabel/sections/conclusion}

%\clearpage

\section*{Acknowledgments}
We thank the anonymous reviewers, Philip Resnik, and Burr Settles for
their insightful comments. We also thank Nikolaos Aletras for
providing the automatic topic labeling code.  Boyd-Graber and
Poursabzi-Sangdeh's contribution is supported by \abr{nsf} Grant
\abr{ncse}-1422492; Findlater, Seppi, and Boyd-Graber's contribution
is supported by collaborative \abr{nsf} Grant \abr{iis}-1409287
(\abr{umd}) and \abr{iis}-1409739 (\abr{byu}).  Any opinions,
findings, results, or recommendations expressed here are of the
authors and do not necessarily reflect the view of the sponsor.

\bibliographystyle{style/acl2016}
\bibliography{bib/forough,bib/journal-full,bib/jbg}




\end{document}