SimCPD: a Simple Framework for Contrastive Prompts of Target-Aspect-sentiment Joint Detection
Cai Ke,Qingyu Xiong,Chao Wu,Hualing Yi,Min Gao,Jie Chen
DOI: https://doi.org/10.1007/s00521-023-08529-6
2023-01-01
Neural Computing and Applications
Abstract:Target-aspect-sentiment detection (TASD) task is a vital task for aspect-based sentiment analysis, which aims to detect (target, aspect, sentiment) triples from sentences. However, most of the existing methods do not generate high-quality textual representations and thus produce poor performances on the TASD task. Moreover, how to deal with insufficient high-quality labeled data is also a great challenge. In this paper, we design DPDA, a Dual Prompts-based Data Augmentation method, that can generate diverse sentences with a question prompt view and an answer prompt view for each input. Then, we propose SimCPD, a Simple Framework for Contrastive Prompts of Target-Aspect-Sentiment Joint Detection, that maximizes the agreement between the two prompt views to obtain high-quality textual representations. Experiments on SemEval-2015 and SemEval-2016 demonstrate that the proposed DPDA can generate high-quality semantic information and alleviate the problem of sparsely labeled data. Moreover, the proposed SimCPD can detect (target, aspect, sentiment) triples more efficiently than other methods. Furthermore, it also achieves better performance on five TASD subtasks.