Page 166 - GHES-3-2

Global Health Economics and
Sustainability
Online health community reviews

two critical points as the emotional orientation threshold.
According to the program, when the sentiment value
is greater than 0.8, the review is considered a positive
review; when the sentiment value is less than or equal to
0.5, the review is considered a negative review; and when
the sentiment value is between 0.5 and 0.8, the review is
considered a neutral review. For instance, the review,
“Excellent medical skills, striving for perfection. Warm
attitude, kind words, patient guidance, careful diagnosis,
formulation of a good plan, instructions for precautions,
and finally, control of the disease always make us feel warm.
Thank you very much, Director Bai!” had a sentiment value
of 0.99, as calculated by SnowNLP; this review is regarded as
a positive review. In contrast, another review states: “Doctor
was impatient after talking with me for about 10 min. He did
not ask me the cause of my illness at all. He directly said that Figure 1. Distribution of sentiment value of reviews
I’d head to a rehabilitation institution to pay for my mental
problems and did not care about my condition. I can hardly decreases. However, in the “0 – 0.1” interval representing
understand! Should the psychiatrist not know the condition the most negative emotion, the number of reviews increased
in detail and then prescribe medicine? Finally, he threw the slightly compared to the intervals of “0.1 – 0.2,” “0.2 – 0.3,”
medical record book directly and signaled me to leave.” This and “0.3 – 0.4.” We can infer from this phenomenon that
review had a sentiment value of 0.26 and was thus regarded while most reviews in OHCs are positive, there are also a
as a negative review. Another review, “After asked me about significant number of extremely negative reviews.
my child’s condition, he directly prescribed medicine and Due to the difference in pathogenesis and prognosis of
did not communicate with my child; it seemed that he the selected diseases, patients adopt different approaches
was in a hurry,” had a sentiment value of 0.61, which was and seek medical treatment accordingly. To compare the
classified as a neutral review. similarities and differences of sentiments in the reviews
To verify the accuracy of the emotional orientation by patients with diabetes, leukemia, and depression, we
threshold, we randomly selected 100 reviews for the selected counted the number and percentage of reviews with different
diseases from the total number of reviews (leukemia: sentiments corresponding to the three diseases, respectively
11021; diabetes: 33274; and depression: 41130). For these (Table 1). Overall, the positive reviews of the three diseases
100 reviews, the results of program recognition were accounted for the highest proportion, whereas the negative
compared with those of manual recognition (Table A1). The reviews accounted for the lowest proportion, indicating
matching rate between program recognition and manual that most patients were satisfied with the doctors’ diagnosis
recognition is 87%, which indicates that the program can and treatment. Among the three diseases, leukemia was
correctly recognize sentiment for most reviews. For some associated with the highest proportion of positive reviews
reviews, a lower sentiment value is given by program and the lowest proportion of negative reviews, while diabetes
recognition for reviews with positive implications, leading was associated with the lowest proportion of positive
to a certain degree of mismatch with manual recognition. reviews and the highest proportion of negative reviews. This
In addition, it is revealed that positive reviews account for observation could be associated with the different prognoses
a large proportion, suggesting that the program can make of the two diseases. Diabetes is difficult to cure completely,
accurate judgments in identifying positive emotions. and the current treatment approach primarily relies on
Among the 85425 reviews, 80.3% were positive, 11.1% drug-assisted therapy. Patients with diabetes tend to leave
were neutral, and 8.6% were negative. To better understand more negative reviews, which may be related to the poor
the sentiment distribution law of different types of reviews, efficacy of drug treatment. In contrast, the treatment period
we divided the sentiment value range (0 – 1) into 10 equal for acute leukemia is significantly shorter and yields more
parts and separately carried out segmented statistical efficacious outcomes, making patients more inclined to give
percentages of the sentiment values of all reviews (Figure 1). positive reviews.
In general, the larger the sentiment value, the more reviews 3.2. Concern analysis
are distributed within the corresponding interval, resulting
in a higher proportion. As the sentiment value decreases, the After using the Jieba library for word segmentation, word
number of reviews distributed in the corresponding interval frequency statistics were carried out for all the reviews.

Volume 3 Issue 2 (2025) 158 https://doi.org/10.36922/ghes.7052