A framework of attribute extraction and dependable aspect term selection from reviews of hospital websites

Abstract

Online reviews found on hospital websites and external platforms constitute user-generated content where patients and their families share their firsthand encounters. As patients increasingly rely on online platforms to share their experiences, understanding the importance of their feedback is paramount for healthcare providers. The novelty of this research lies in the development of advanced frameworks that not only extract relevant information but also offer a more sophisticated and coherent analysis of the multifaceted aspects embedded in patient reviews. Hence, this work involves collecting data from various hospital websites, followed by data pre-processing to ensure accuracy and consistency. Subsequently, two distinct frameworks are proposed. The first framework aims to extract specific attributes (topics) mentioned in reviews, enhancing the granularity of information derived from the collected data. The second framework addresses the efficient extraction of aspect terms from pre-processed data, utilizing a coherence score-based approach called as modified latent dirichlet allocation term frequency-inverse document frequency (M-LDA TF-IDF). The M-LDA TF-IDF has achieved better a coherence score of 0.478 which is much better in comparison with other topic modelling approaches.