Compound Aspect Extraction by Augmentation and Constituency Lattice

Xiaojun Quan,Zhengcheng Min,Kun Li,Yunyi Yang

DOI: https://doi.org/10.1109/taffc.2022.3161683

IF: 13.99

2022-01-01

IEEE Transactions on Affective Computing

Abstract:Aspects are opinion targets to extract in aspect-based sentiment analysis. While existing methods can already produce satisfactory extraction results, they suffer when faced with compound aspect terms, typically phrase-level aspect terms that have inner structure and occur infrequently in the training set. This issue can be mainly attributed to the scarcity of training examples targeting compound aspect terms and by the neglect of the syntactic structure of a sentence in the modeling process. In this article, we aim to cope with compound aspect extraction by a two-stage hybrid approach. First, we introduce a conditional generation method for data augmentation in a masked sequence-to-sequence framework, which is controllable to preserve original aspects while generating a new sentence. Second, we propose a constituency lattice structure that is induced from the constituency-based parse tree of a sentence. Experimental results on two review datasets show that this approach can greatly improve the effect of compound aspect extraction.