基於 BERT-DAOA 的意見目標情感分析

Translated title of the contribution: Aspect-Based Sentiment Analysis Based on BERT-DAOA

Chen Yu Chen, Chia Hui Chang

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Social media networks provide rich and diverse information, making opinion analysis and network volume analysis a new method to investigate and understand the market. Sentiment analysis aims to determine the emotional category in a given text. Since there might be several targets being commented on in the text, Aspect-base sentiment analysis (ABSA) has been proposed to explore the sentiment categories of a target in different aspects. In this paper, we explore the idea of ABSA for sentiment analysis of singers on social networks. We utilize BERT as the embedding layer method for characters and words to explore the relationship between a given sentence and a mentioned target. We consider two attention mechanisms to enhance the performance. The experimental results show that adding the attention layer on top of BERT outperforms the basic BERT-CLS model.

Translated title of the contributionAspect-Based Sentiment Analysis Based on BERT-DAOA
Original languageChinese (Traditional)
Title of host publicationROCLING 2020 - 32nd Conference on Computational Linguistics and Speech Processing
EditorsJenq-Haur Wang, Ying-Hui Lai, Lung-Hao Lee, Kuan-Yu Chen, Hung-Yi Lee, Chi-Chun Lee, Syu-Siang Wang, Hen-Hsen Huang, Chuan-Ming Liu
PublisherThe Association for Computational Linguistics and Chinese Language Processing (ACLCLP)
Pages398-412
Number of pages15
ISBN (Electronic)9789869576932
StatePublished - 2020
Event32nd Conference on Computational Linguistics and Speech Processing, ROCLING 2020 - Taipei, Taiwan
Duration: 24 Sep 202026 Sep 2020

Publication series

NameROCLING 2020 - 32nd Conference on Computational Linguistics and Speech Processing

Conference

Conference32nd Conference on Computational Linguistics and Speech Processing, ROCLING 2020
Country/TerritoryTaiwan
CityTaipei
Period24/09/2026/09/20

Fingerprint

Dive into the research topics of 'Aspect-Based Sentiment Analysis Based on BERT-DAOA'. Together they form a unique fingerprint.

Cite this