Page 168 - GHES-3-2

Global Health Economics and
Sustainability
Online health community reviews

A B

C D

Figure 2. Word cloud for patient reviews: (A) All reviews; (B) positive reviews; (C) negative reviews; and (D) neutral reviews

should possess. The terms “check,” “doctor,” “consultation,” continuously evolve (Cui et al., 2023; Fu et al., 2023;
and “convenient” also appear more frequently in the Elbattah et al., 2021; Nandwani & Verma, 2021). Many
negative reviews, possibly due to unpleasant or inconvenient scholars have employed topic models, such as Latent
experiences that patients encounter during the registration Dirichlet Allocation (LDA), to mine discussion topics
process or while seeing a doctor. This suggests that patients and latent themes among users in health communities.
are concerned not only with the basic diagnosis and They have also optimized model structures or introduced
treatment process but also with factors such as hospital algorithms, such as random forests, to enhance the
environment and treatment procedures, which are less precision and applicability of topic identification (Bi et al.,
directly related to the doctors themselves. 2020; Chen et al., 2024).
Finally, for neutral reviews, terms like “responsible” and In addition, sentiment analysis and emotion recognition
“satisfied” indicate that patients generally approve of the have also garnered significant attention. Numerous studies
doctor’s diagnosis and treatment process. Conversely, some have identified positive or negative emotional tendencies
terms, such as “patience,” “attitude,” “explain,” “revisit,” and of users during the process of seeking medical advice
“reply,” suggest that patients may not be entirely satisfied and medication by constructing sentiment dictionaries
with their communication with the doctors. Although or training classification models, revealing emotional
patients are mostly satisfied with the doctors’ diagnostic fluctuations and their impacts at different times and on
and treatment services, the frequent appearance of the different topics (Han et al., 2018; Liu & Kong, 2021; Luo
term “hope” also implies that patients are attentive to et al., 2020; Necaise & Amon, 2024; Rustam et al., 2021).
areas for improvement in various aspects. Ultimately, the To better extract and understand the users’ potential needs
combination of both positive and negative emotions results and behavioral patterns from vast amounts of unstructured
in neutral evaluations. text with higher accuracy in semantic understanding and
topic clustering, some studies have adopted methods that
4. Discussion integrate deep learning with topic modeling (Shen et al.,
2021). Furthermore, research from the perspective of
4.1. Related studies
social support and emotional connection has revealed the
Text analysis, a research hotspot in the field of natural driving role of negative emotions in users’ information
language processing, has seen its methods and topics needs, as well as the importance of positive support for

Volume 3 Issue 2 (2025) 160 https://doi.org/10.36922/ghes.7052