A Composite Sentiment Summarizer Score for Patient Reviews: Extending RoBERTa

Recently, sentiment analysis is applied in the healthcare domain to improve patient’s experiences through collecting and analyzing patient reviews. With the ascent of artificial intelligence, the Robustly Optimised Bidirectional Encoder Representations from Transformer Pretraining Approach (RoBERTa) has shown exceptional performance in the task of Natural Language Processing, including sentiment analysis. The result of sentiment analysis using RoBERTa generates three sentiment scores, representing the degree of negative, neutral, and positive sentiment in the analyzed text. Nevertheless, these sentiment scores are presented separately, making it difficult to gain a comprehensive understanding of the mixed sentiments expressed in the review text. In this paper, the concept of composite sentiment summarizer is introduced by combining the positive, negative, and neutral scores into an overall sentiment score, a more complete analysis of the mixed sentiments can be obtained from the unstructured reviews. This composite sentiment summarizer score provides a more balanced and precise representation of the sentiment expressed in the text. This paper investigated four proposed approaches by extending capabilities of RoBERTa with the composite sentiment summarizer score. The results demonstrated the potential of using the ranking approach to calculate a composite sentiment summarizer score for sentiment analysis and highlighted the importance of considering a more nuanced approach to sentiment analysis with a wider range of sentiment intensities.