本研究將統整目標情感分析最新的做法，基於各深度學習模型方法的不同來探討應用於中文資料集上的適應性，並進行比較與分析，提供目標情感分析使用於中文資料集上的準則。實驗結果顯示，基於BERT模型的LCF-ATEPC、R-GAT和AEN-BERT在中文文本的效果表現皆優於未使用BERT的模型，其中又以AEN-BERT為最佳模型方法。;With the development of the Internet, people play the role of information sharers by sharing reviews on goods and services through the Internet. For consumers, reviews can be used to understand the strengths and weaknesses of different goods and services. Therefore, we can use sentiment analysis to find the hidden value from reviews.
Majority of previous studies based on aspect sentiment analysis were conducted on English datasets. In this study, we focus on the latest and highly discussed deep learning models of aspect sentiment analysis to explore adaptability on Chinese language. So as to provide the marketing department in the industry and it will help marketing department with online reviews to understand actual opinions from customers.
This study will integrate the latest methods of aspect sentiment analysis. Based on each deep learning model approach to explore adaptability of Chinese text, compare and analyze them. Provide the guidelines for use of aspect sentiment analysis on Chinese text. Our experimental results show that the BERT-based models, LCF-ATEPC, R-GAT and AEN-BERT perform better in Chinese text than the non-BERT based models. Moreover, the AEN-BERT is the best overall performing model in our experiments.
