Span-based relational graph transformer network for aspect–opinion pair extraction

Aspect extraction and opinion extraction are two fundamental subtasks in aspect-based sentiment analysis. Many methods extract aspect terms or opinion terms but ignore the relationships between them. However, such relationships are crucial for downstream tasks, such as sentiment classification and c...

Full description

Saved in:
Bibliographic Details
Published in:Knowledge and information systems Vol. 64; no. 5; pp. 1305 - 1322
Main Authors: Li, You, Wang, Chaoqiang, Lin, Yuming, Lin, Yongdong, Chang, Liang
Format: Journal Article
Language:English
Published: London Springer London 01.05.2022
Springer Nature B.V
Subjects:
Classifiers
Computer Science
Data mining
Data Mining and Knowledge Discovery
Database Management
Information Storage and Retrieval
Information Systems and Communication Service
Information Systems Applications (incl.Internet)
IT in Business
Recommender systems
Regular Paper
Sentiment analysis
Transformers
Sentiment analysis
Relational graph convolution
Aspect–opinion pair extraction
ISSN:0219-1377, 0219-3116
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Aspect extraction and opinion extraction are two fundamental subtasks in aspect-based sentiment analysis. Many methods extract aspect terms or opinion terms but ignore the relationships between them. However, such relationships are crucial for downstream tasks, such as sentiment classification and commodity recommendation. Recently, methods have been proposed to extract both terms jointly; however, they fail to extract them as pairs. In this paper, we explore the aspect–opinion pair extraction task that aims to extract aspect and opinion terms in pairs. To carry out this task, we propose a span-based relational graph transformer network that consists of a span generator, a span classifier, and a relation detector. The span generator enumerates all possible spans to generate the candidates for aspect or opinion terms and filters non-aspects or non-opinions terms, while the relation classifier extracts aspect–opinion pairs. We propose a relational graph convolutional network to capture the dependent relationships between aspect and opinion terms. Extensive experiments show that the proposed model achieves the state-of-the-art performance using four benchmark datasets.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0219-1377
0219-3116
DOI:10.1007/s10115-022-01675-8