Multilabel Aspect-Based Sentiment Analysis (M-ABSA) of Healthcare Services Based on Opinion Extraction and Powerset Labeling

Abstrak

Healthcare facility services are important in supporting public health and reducing health risks. Multilabel Aspect-based Sentiment Analysis (M-ABSA) can be used to analyze patient reviews from social media, allowing healthcare facility managers to identify areas that need improvement based on public perception, thereby supporting the continuous enhancement of healthcare service quality. This study compares two M-ABSA approaches: a transformation approach with powerset labeling and feature engineering techniques with opinion extraction. In addition, a traditional machine learning method (Random Forest) is compared with deep learning methods such as Long Short Term Memory (LSTM) and Indonesian Bidirectional Encoder Representative from Transformers (IndoBERT). The experimental results show that powerset labeling with random oversampling led to an accuracy of 0.79, while M-ABSA using opinion extraction achieved 0.81 with LSTM. These results indicate that the feature engineering approach using opinion extraction combined with deep learning, particularly LSTM, performs well in analyzing healthcare facility user review data.

Identifier

• DOI: https://doi.org/10.48084/etasr.16961
• OpenAlex: https://openalex.org/W7163748773