Using FastText-BERT to Extract Semantic Relations and Improve Sentiment Analysis of Persian Healthcare Service Reviews
Introduction: The analysis of patients’ opinions is considered a valuable indicator for assessing the quality of healthcare services. The increasing volume of textual reviews about healthcare has made these reviews a critical factor in other patients’ decision-making processes when selecting medical...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | fas |
| Published: |
Shiraz University of Medical Sciences
2025-01-01
|
| Series: | مجله علوم پزشکی صدرا |
| Subjects: | |
| Online Access: | https://smsj.sums.ac.ir/article_50773_7744ec54d51130074e8f1dca0bb3e230.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Introduction: The analysis of patients’ opinions is considered a valuable indicator for assessing the quality of healthcare services. The increasing volume of textual reviews about healthcare has made these reviews a critical factor in other patients’ decision-making processes when selecting medical services. Consequently, researchers aimed to extract valuable insights, classify sentiments, and identify patient needs and behavioral patterns through sentiment analysis, thereby developing appropriate strategies to enhance patient satisfaction. However, patient reviews often contain a significant amount of specialized terminology, and existing sentiment analysis tools are typically trained on general-domain data. Therefore, to analyze these reviews accurately, it is essential to employ models and their combinations in a way that ensures reliable and valid results.Methods: To improve the efficiency and accuracy of sentiment analysis for Persian healthcare reviews, this study utilized the FastText-BERT hybrid embedding model for semantic relation extraction and the CNN-BiLSTM model for sentence-level sentiment classification.Results: The proposed framework achieved an accuracy of 86% and an F1-score of 84.99%.Conclusion: The results demonstrated that combining embedding models leverages the strengths of both approaches, enabling the identification of specialized and out-of-domain expressions and the extraction of semantic relationships between them. This combination significantly enhances the efficiency and accuracy of sentiment analysis. |
|---|---|
| ISSN: | 2322-4339 |