We introduce a novel approach to sentiment classification.
Our Reasoning Through Search (RTS) technique uses existing
labeled data, query formation strategies, and a case base
to estimate the sentiment of a text review. Unlike previous
systems, when classifying a document from a domain that
the system does not have explicit, in-domain training data,
our classification system leverages domain relatedness and
a case base of labeled reviews to perform the classification.
While the system does require labeled training data, it does
not rely on the guaranteed presence of in-domain labeled
training data.