•  
  •  
 

Abstract

Sentiment analysis is an important field in Natural Language Processing (NLP) that focuses on processing consumer opinions to gain useful insights. The information generated from sentiment analysis can be used as a basis for business decision-making, service quality evaluation, and the formulation of more effective marketing strategies. In the local context, Bangkalan Batik, as one of Madura's distinctive cultural products, has high economic value and cultural identity. However, consumer reviews available online, for example through Google Maps, are still rarely utilized optimally by MSMEs as a source of strategic information. Therefore, this study was conducted to develop a sentiment classification model capable of overcoming the limitations of small data and unbalanced class distribution. The research dataset consisted of 1,000 Bangkalan Batik consumer reviews categorized into three sentiment classes: positive, neutral, and negative. After the preprocessing stage, text representation was performed using TF-IDF. Next, three basic algorithms, Naïve Bayes, Logistic Regression, and Support Vector Machine, were combined through the Soft Voting approach. To improve performance, Logistic Regression meta-classification was used as an additional layer of stack-based Meta-Learning. In addition, the Synthetic Minority Oversampling Technique (SMOTE) was applied to overcome class imbalance so that the model was more sensitive to minority opinions. The results of experiments with stratified k-fold cross-validation show that the proposed model performs better than both single and conventional ensemble models. The developed hybrid model achieves 87% accuracy and an F1 score of 88%, and shows a significant improvement in remembering minority classes. This research contributes to the development of ensemble-based text classification methods for small and imbalanced datasets.

DOI

https://doi.org/10.17977/um018v8i12025p81-91

First Page

81

Last Page

91

Download

Share

COinS