BART-based contrastive and retrospective network for aspect-category-opinion-sentiment quadruple extraction

• Published in: International journal of machine learning and cybernetics Vol. 14; no. 9; pp. 3243 - 3255
• Main Authors: Xiong, Haoliang, Yan, Zehao, Wu, Chuhan, Lu, Guojun, Pang, Shiguan, Xue, Yun, Cai, Qianhua
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01-09-2023; Springer Nature B.V
• Subjects: Artificial Intelligence; Complex Systems; Computational Intelligence; Control; Data mining; Engineering; Mechatronics; Original Article; Pattern Recognition; Reading comprehension; Robotics; Sentiment analysis; Systems Biology; Triplets; Aspect-category-opinion-sentiment quadruple extraction; Implicit sentiment analysis; Contrastive learning
• ISSN: 1868-8071; 1868-808X
• DOI: 10.1007/s13042-023-01831-8

Aspect-category-opinion-sentiment (ACOS) quadruple extraction is a fine-grained sentiment analysis task to extract full sentiment information, which aims to extract all the ACOS quads in a given sentence. ACOS contains four types of quads: explicit aspect and explicit opinion, implicit aspect and explicit opinion, explicit aspect and implicit opinion, and implicit aspect and implicit opinion. Current studies generally apply the two-stage methods to ACOS studies. However, there are two main limitations. One is the error propagation while the other is the ignorance of diversity among different types of quads. In this work, we propose a BART-based Contrastive and Retrospective Network (BART-CRN), which tackles ACOS extraction as a sequence generation task. Specifically, a machine reading comprehension based (MRC-based) supervised contrastive and retrospective learning module is developed, which aims to learn the associations among all types of quads and determines the context-related generative quads through an end-to-end way. Experimental results on two ACOS datasets reveal that our model outperforms the baseline methods and achieves advanced performances.