William Darling, Guillaume Bouchard
Will appear on ACM Transactions on Knowledge Discovery from Data.
To reach the ideals that a true democracy can achieve, citizens must be able to participate in the political
decision making process as easily as possible. Soliciting opinions on the Internet is a start, hut even if the
people heed the call, it is then left to the policy makers to make sense of the data that has been left, which
will likely include redundancy. off-topic complaints, spam, and other noise. To deal with this problem, we
present a system that intelligently summarizes citizens’ opinions in the political context and presents them
to policy makers in a structured way such that the electorate can be understood in terms of their problems
and their sentiment with respect to proposed and completed political decisions. We present a high-quality
novel fully automatic summarization system by harnessing the “wisdom of the crowd”; while we make use of
iupervmed machine learning algorithms, our training data is created in real-time by calling crowd-sourcing
platforms that interactively build up our classifiers. In this paper, we describe our system as a four step
process: finding opinion bearing comments, learning the topics that they cover, classifying the sentiment that
is reflected, and selecting representative sentences for the outputted structured summary We present our
approach along with quantitative and qualitative experimental results and example summaries on a notice
and comment dataset for the USDA’s proposed National Organic Program regulations. Our results show
that this method efficiently creates interpretable and useftil summaries that will be helpful in improving political participation.
Report number: