# External-knowledge enhanced dual encoder and contrastive learning for aspect sentiment triplet extraction

## TL;DR

This paper introduces a new method for extracting sentiment triplets from text using dual encoders and contrastive learning to improve accuracy.

## Contribution

The novel approach combines dual encoders with contrastive learning to better capture sentence semantics and word meanings for ASTE.

## Key findings

- The dual encoder setup improves semantic and syntactic information extraction.
- The boundary-aware contrastive learning module enhances boundary tag recognition.
- Experiments show the method outperforms existing approaches on ASTE datasets.

## Abstract

Aspect Sentiment Triplet Extraction (ASTE) is an emerging subtask of Aspect-Based Sentiment Analysis (ABSA), aiming to extract aspect terms, opinion terms, and the corresponding sentiment polarity from sentences. Many existing ASTE methods neglect to mine the deeper semantics of the sentence as well as ignore the intrinsic meanings of individual words. In order to address these limitations, this paper proposes a novel approach for ASTE. Firstly, dual encoders are used to extract the semantic and syntactic information of the sentence, the semantic encoder uses BERT and Graph Convolutional Networks (GCNs) to extract the semantic information, and the syntactic encoder employs a Bi-directional Long and Short-Term Memory (Bi-LSTM) network and GCNs to extract the syntactic information. Secondly, a feature fusion module is designed to fuse the information from the dual encoders. Finally, to enhance the ability of the model to recognize boundary tags, we design a boundary-aware contrastive learning module. Experimental results on ASTE-Data-V1 and ASTE-Data-V2 demonstrate the effectiveness of our proposed method.