A Multi-Granularity Opinion Summarization Method

Abstract

Existing opinion mining (OM) is limited to applications on commercial reviews, with aspect and sentiment of the opinions in a coarse-grained form. In this paper, we further explore the definition of OM by extending the concepts of aspect and sentiment, and propose an opinion summarization method based on Multi-granularity Clustering and BERT (Jacob et al., 2018), i.e., MCB for emergent online discussion record in keeping with the further definition. A supporting Chinese corpus, ZH45 comprising 45 groups of discussion, and assorted metrics are also proposed. Experiments based on ZH45 and the metrics demonstrate that MCB produces succinct and insightful opinion summaries.

Tasks

Reproductions