Better Document-level Sentiment Analysis from RST Discourse Parsing

TL;DR

This paper demonstrates that utilizing RST discourse parsing, including reweighting discourse units and recursive neural networks, significantly enhances document-level sentiment analysis accuracy.

Contribution

It introduces a novel approach combining RST discourse parsing with neural networks to improve sentiment analysis at the document level.

Findings

Reweighting discourse units improves lexicon-based sentiment accuracy.

Recursive neural networks over RST structures outperform classification methods.

Discourse structure integration yields substantial performance gains.

Abstract

Discourse structure is the hidden link between surface features and document-level properties, such as sentiment polarity. We show that the discourse analyses produced by Rhetorical Structure Theory (RST) parsers can improve document-level sentiment analysis, via composition of local information up the discourse tree. First, we show that reweighting discourse units according to their position in a dependency representation of the rhetorical structure can yield substantial improvements on lexicon-based sentiment analysis. Next, we present a recursive neural network over the RST structure, which offers significant improvements over classification-based methods.