Sentiment Extraction by Leveraging Aspect-Opinion Association Structure

Sentiment extraction aims to extract and group aspect and opinion words from online reviews. Previous works usually extract aspect and opinion words by leveraging association between a single pair of aspect and opinion word, but the structure of aspect and opinion word clusters has not been fully exploited.
In this paper, we investigate the aspect-opinion association structure, and propose a “first clustering, then extracting” unsupervised model to leverage properties of the structure for sentiment extraction. For the clustering purpose, we formalise a novel concept syntactic distribution consistency as soft constraint in the framework of posterior regularization; for the extraction purpose, we extract aspect and opinion words based on cluster-cluster association. In comparison to traditional word-word association, we show that cluster-cluster association is a much stronger signal to distinguish aspect (opinion) words from non-aspect (non-opinion) words. Extensive experiments demonstrate the effectiveness of the proposed approach and the advantages against stateof-the-art baselines.