C$^{3}$LPGCN:Integrating Contrastive Learning and Cooperative Learning with Prompt into Graph Convolutional Network for Aspect-based Sentiment Analysis

Anonymous

TL;DR: Utilizing contrastive learning and cooperative learning to introduce label information as explicit priori for Aspect-based Sentiment Analysis.

Abstract: Aspect-based Sentiment Analysis (ABSA) is a fine-grained task. Recently, using graph convolutional networks (GCNs) to model syntactic information has become a popular topic. In addition, there is a growing consensus to enhance sentence representation using contrastive learning. However, incorrect modeling of syntactic information may introduce additional noise. Meanwhile, as a method that implicitly incorporates labeling information as prior knowledge, contrastive learning does not make sufficient use of this prior. To alleviate these problems, we propose C$^{3}$LPGCN, which integrating Contrastive Learning and Cooperative Learning with Prompt into GCN. To tackle the first issue, we propose mask-aware aspect information filter, which combine prompt-tuned information with aspect information to filter the syntactic information. To further utilize the label information, we propose prompt-based contrastive learning and cooperative learning. We construct comparison samples containing labels so that the model pays more attention to aspect- and sentiment-related information during feature learning. Cooperative learning further extracts label information by aligning the representation and output distribution of input samples with true label samples. Extensive experiments on three datasets demonstrate that our method significantly improves the performance of the model compared to traditional contrastive learning methods. Moreover, our C$^{3}$LPGCN outperforms state-of-the-art methods.

Paper Type: long

Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining

Contribution Types: Model analysis & interpretability

Languages Studied: English