Authors: Nelly Indriani Widiastuti; Bagus Perdana Yusuf; Kania Evita Dewi
DOI: 10.1109/INCITEST59455.2023.10396931
Abstract
Aspect-based sentiment Analysis (ABSA) has been contributing to a more nuanced understanding of sentiment-related issues and offering enhanced opportunities for sentiment analysis. The number of aspects in data sets are often unbalanced. This causes results that tend towards the majority class and ignore the minority class This research uses Bi-LSTM to classify the aspects and sentiments because Bi-LSTM is able to handle problems in LSTM. The primary objective of this research was to assess the efficacy of the Bidirectional Long Short-Term Memory method in the context of Aspect-Based Sentiment Analysis applied to reviews of Marketplace applications and common problems in machine learning in unbalanced datasets. The preprocessing of input data involved word embedding through Word2Vec Continuous Bag of Words (CBOW). Primarily, the training dataset encompassed 2250 instances, while the test dataset comprised 750. The evaluation enclosed a comparative analysis involving the proposed method (Bi-LSTM+CBOW+ROS), on Bi-LSTM+CBOW, Bi- LSTM+CBOW+RUS and Bi-LSTM+CBOW+SMOTE. The test results show that the proposed method is superior to all comparison methods. Bi-LSTM+CBOW+ROS produces the highest accuracy, namely the usability aspect of 92.40%, the maximum precision is 98.03%, the maximum recall is 100% and the maximum f1-score is 98.68%. The Usability aspect got the best results for all the methods tested. © 2023 IEEE.
Author keywords
Aspect-based sentiment analysis; Bidirectional Long Short-term Memory; Random Oversampling; Word2Vec CBOW
This article can be accessed at: https://www.scopus.com/pages/publications/85185201670
