Emotional differences based on comments on doctor-patient disputes with varying levels of severity

Abstract

BACKGROUND

The risks associated with negative doctor-patient relationships have seriously hindered the healthy development of medical and healthcare and aroused widespread concern in society. The number of public comments on doctor-patient relationship risk events reflects the degree to which the public pays attention to such events.

AIM

To explore public emotional differences, the intensity of comments, and the positions represented at different levels of doctor-patient disputes.

METHODS

Thirty incidents of doctor-patient disputes were collected from Weibo and TikTok, and 3655 related comments were extracted. The number of comment sentiment words was extracted, and the comment sentiment value was calculated. The Kruskal-Wallis H test was used to compare differences between each variable group at different levels of incidence. Spearman’s correlation analysis was used to examine associations between variables. Regression analysis was used to explore factors influencing scores of comments on incidents.

RESULTS

The study results showed that public comments on media reports of doctor-patient disputes at all levels are mainly dominated by “good” and “disgust” emotional states. There was a significant difference in the comment scores and the number of partial emotion words between comments on varying levels of severity of doctor-patient disputes. The comment score was positively correlated with the number of emotion words related to positive, good, and happy) and negatively correlated with the number of emotion words related to negative, anger, disgust, fear, and sadness.

CONCLUSION

The number of emotion words related to negative, anger, disgust, fear, and sadness directly influences comment scores, and the severity of the incident level indirectly influences comment scores.

Key Words: Doctor-patient relationship; Doctor-patient dispute; Comments; Emotional differences; Weibo; TikTok

Core Tip: This study applies sentiment analysis methods to specific instances of doctor-patient disputes and explores differences in sentiment among different levels of incidents. Sentiment analysis using a combination of manual and machine methods compensates for the lack of using a single method to some extent. This study selected emotion as an entry point to explore the factors influencing comment scores on doctor-patient dispute incidents.

INTRODUCTION

There has been steady progress in the extent of medical and health system reforms. Consequently, terms such as “doctor-patient relationship” have become a hot topic of discussion among the public, and phrases such as “medical accidents” and “violent injury to doctors” have appeared in the media, thereby triggering ongoing concern and public discussion.

The Chinese government has always attached great importance to the establishment of harmonious doctor-patient relationships, as evidenced by the implementation of measures such as mediation mechanisms and dispute prevention mechanisms. However, doctor-patient relationships in China have not fundamentally improved, and the number of doctor-patient judicial cases has been increasing annually. With the rapid development and wide application of internet-based self-media technology, platforms such as Weibo and TikTok have become important channels through which the Chinese public can freely express their opinions[1]. Individuals use network platforms as information exchange channels; specifically, they use text, pictures, video, and other diversified text forms to express their views and vent their emotions regarding hotspot reports on doctor-patient disputes, thereby leading to diverse expressions of emotional output. As doctor-patient disputes continue to ferment in relevant social media reports and comments from the public, healthcare incidents that are closely related to people’s lives and health have gradually become a hot topic of public opinion online [2]. The emotional impact of public discourse in the online evolution of doctor-patient disputes is increasing, thus generating a variety of significant social impacts and subconsciously guiding and changing the public’s emotions and perceptions of both doctors and patients during incidents.

Emotional differences among the public in doctor-patient dispute incidents are an important factor influencing cognitive differences in doctor-patient relationships. Thus, the study of emotional differences in the evaluation of doctor-patient disputes provides new research topics and ideas for new research directions with respect to the rational resolution of doctor-patient disputes as well as proper guidance for the development of healthy doctor-patient relationships. From the perspective of emotional transmission, research on real comments on social media platforms, the public’s emotional expression, the nature of the output content, the process of the evolution of emotional dynamics, and the triggers of emotion generation in doctor-patient disputes can help to deepen our understanding of doctor-patient disputes and to establish a reasonable mechanism for mitigating the risk of medical malpractice.

Emotion is a generic term for a range of subjective cognitive experiences. This term describes a person’s attitudinal experience of objective things and the corresponding behavioral responses. Emotion is generally considered a psychological activity that is mediated by the individual’s wants and needs. Furthermore, it is an internal subjective experience that is always accompanied by some kind of external manifestation that is interpreted as a behavioral trait[3]. Public comments on incidents involving doctor-patient disputes are external manifestations of emotions, as these comments involve making subjective emotional evaluations of the doctor and patient in objective incidents of doctor-patient disputes through perspectives of individuals in the public.

As the outcomes of a doctor-patient dispute progress, the position and emotional state of the public’s evaluative discourse change accordingly. Recent research on doctor-patient disputes has focused on factors influencing doctor-patient disputes and countermeasures but has not examined changes in public emotions and positions caused by news reports on doctor-patient disputes. Therefore, we take emotions as the entry point in this study, and explore differences in emotions at different levels of doctor-patient dispute incidents after grading doctor-patient dispute incidents according to their severity. From the perspective of the emotional state of public opinion, we propose rationalized suggestions for establishing a harmonious doctor-patient relationship and publishing truthful and reliable news reports on doctor-patient disputes.

Current research status: Domestic and foreign scholars have conducted numerous studies of topics related to doctor-patient disputes, online public opinion, and textual sentiment analysis. Recent research on doctor-patient disputes has mainly focused on the analysis of the causes of doctor-patient disputes and countermeasures and has relied on the basic framework of the doctor’s image, doctor-patient communication, social media coverage and public opinion control. Domestic and foreign scholars mainly attribute doctor-patient disputes to a lack of in-depth communication between doctors and patients[4], a lack of trust between doctors and patients[5], information asymmetry[6], and an uneven distribution of healthcare resources[7]. However, few researchers have focused on the moderating role of emotions in incidents of doctor-patient dispute. Therefore, in this study, we explored the influence of emotion on doctor-patient dispute incidents, examined the differences in public emotion at different levels of doctor-patient dispute incidents and investigated the influence of the public’s emotional intensity and the position they represent in their comments on doctor-patient disputes.

Wei et al[8] proposed that in an online virtual environment, both the truthfulness and symmetry of information can lead to cognitive bias in the public. Official media reports of doctor-patient disputes are more authentic and can affect the public’s cognitive judgement and emotional attitudes towards the behavior of doctors and patients in these disputes. Social perspectives develop through social interaction[9]. The spread of numerous perspectives is accomplished through interactions between subjects in society[10,11]. The emotions expressed by the public affect the width and breadth of the dissemination of doctor-patient disputes and the emotions generated by doctor-patient disputes and the positive emotional output advocated by the mainstream of society have a mutually interfering and antagonistic effect. Negative emotions can even become the cause of the user’s negative behaviors, thereby affecting the development of incidents. The opinions of public figures can also have a direct impact on the development of an incident. The better their arguments are and the more moderate their attitudes are, the more likely they are to steer public opinion[8]. Doctor-patient relationships in online public opinion refer to the sum of all public emotions, attitudes, opinions, and subsequent impacts on hot healthcare incidents that occur on the internet[12]. Doctor-patient relationships in online public opinion systems have the characteristics of general online public opinion, such as negative topics, distorted information, and difficulty in control[13,14]. The wrongdoing of the responsible side in doctor-patient disputes based on news reports can easily affect public emotions and opinions.

With the development of the internet, text sentiment analysis has gradually become a popular research topic in natural language processing. Sentiment detection is closely related to sentiment analysis, which uses natural language processing[15] and involves identifying the positive, negative, and neutral nature of a text. Sentiment analysis involves seven basic emotions, disgust, fear, happiness, anger, guilt, sadness and shame[16,17]. Extracting sentiment information from comment texts requires sentiment analysis techniques, which involves sentiment classification. Currently, sentiment classification is either machine learning-based [18] or lexicon-based[19]. Pang et al[18] were early researchers working on machine learning-based sentiment analysis of text. They applied plain Bayesian, maximum entropy, and support-vector machine algorithms to analyze the sentiment of film reviews. Their results showed that the support-vector algorithm performed better than the other approaches. In a deep learning model, Su et al[20] used sentence-level analysis to identify the overall sentiment polarity conveyed in a given sentence. Zhang et al[21] proposed a method for classifying emotions based on emotion-specific word vectors. They constructed a heterogeneous network to obtain the vectors and then after obtaining them, they trained a long short-term memory network for emotion classification. In their approach, Zhang et al[22] extended the sentiment lexicon of the Dalian University of Technology by adding related domain-name lexicons, such as the network lexicon, emoji lexicon, and others. Finally, they computed the sentiment extremes of the Weibo comment information by transforming the uniform weights of its sentiment classification. Zhu et al[23] developed and used a method based on semantic rules and weighted expressions to classify the sentiment of Chinese microblog comments. In a detailed empirical study on sentiment classification, Liu et al[24] used various multilabel classification methods, including the Dalian University of Technology Sentiment Lexicon, National Taiwan University Sentiment Lexicon, and HowNet lexicon. They concluded that the Dalian University of Technology Sentiment Lexicon was the best among the 2014 different sentiment lexicons. In this study, we used the Dalian University of Technology Sentiment lexicon and applied both manual and machine classification methods to perform text sentiment analysis. The use of both machine and manual analysis minimizes the drawbacks of isolating single emotion words by machine analysis. Sentiment value calculation can be carried out by using a machine to extract emotion words and manual annotation to review and comment.

Many domestic and foreign scholars have conducted studies on the cause of doctor-patient disputes and analyzed textual sentiments, but few have examined the emotional tendencies reflected by public evaluation of reports of doctor-patient disputes. In this study, we aimed to explore the difference of emotions associated with the severity of doctor-patient dispute incidents. We wanted to investigate the impact of the intensity of emotions and the positions represented by public comments about these incidents. Based on the results of existing studies, we propose three hypotheses.

Hypothesis 1: Comment scores for doctor-patient dispute incidents are related to the severity of the incident (i.e. the severity of the damage caused by the behavior of the responsible at-fault side). Hypothesis 2: There is a positive correlation between the severity of the incident and the number of words related to sadness (i.e. the tendency of the public to feel pity and sadness for the victim) in the public comments. Hypothesis 3: The intensity of negative emotions brought about by the negative-emotion words in public comments has a direct effect on the incident comment score and influence of the opinion. This empirical study used a sentiment classification method based on sentiment lexicon analysis of Weibo and TikTok comments. The intersentence and organizational structure rules are shown in Figure 1.

Figure 1 Conceptual model of the sentiment classification method for Weibo and TikTok comments.

MATERIALS AND METHODS

Sample and data collection

We collected and collated 30 doctor-patient dispute incident reports from social media platforms (e.g., Weibo and TikTok). The incidents were manually graded by severity. Level I incidents included verbal conflicts between doctors and patients (e.g., a woman was treated rudely and had an argument with hospital staff after discussing cost issues). Level 2 incidents included physical conflict between doctor and patient (e.g., Shanghai Sixth Hospital notified medical staff of physical conflict). Level 3 incidents included doctor-patient disputes leading to injuries (e.g., in Wuhan, a man with a knife hacked a doctor and was detained). Level 4 incidents included doctor-patient disputes leading to fatalities (e.g., in Hefei, a 27-year-old female patient died in the hospital due to medication). There were 12 level 1, 4 level 2, 9 level 3, and 5 level 4 cases.

Manual text preprocessing

Emoticons, pictures, web links, and the symbol @, which is used to mention someone, frequently appear in the text of Weibo and TikTok comments. These elements not only enrich the text but also pose some difficulties to the study. Consequently, The text was preprocessed for the analysis, and the symbol “@username” was used in Weibo and TikTok comments to tell someone something or to receive someone’s attention: (1) Preprocessing had no substantial impact on sentiment analysis, so it was filtered; (2) As the comments collected for the analysis corresponded to the topic, instances of “#Theme#” were ignored and filtered directly; (3) Web links, animations, videos, pictures, and emoticons were filtered; and (4) To avoid loss of comments, traditional Chinese was converted to simplified Chinese and English was converted to Chinese.

We collected and organized comments related to relevant incidents based on graded doctor-patient disputes. We collected a total of 3655 comments after manual text preprocessing. We collected and organized comments related to relevant incidents based on graded doctor-patient disputes. Examples of the collected comments are shown in Table 1.

Table 1 Examples of collected comments on incidents of doctor-patient disputes.

Serial number	Comments
1	Doctor-patient conflicts are not without cause
2	Not all people are good, and not all doctors are good
3	There are problems on both sides
4	Do the right thing. If the doctor doesn’t do a good job, complain to him
5	I think the doctor’s attitude is very bad now
6	Medical ethics are important
7	Where there is a cause, there is an effect. In particular, some patients are already suffering from illnesses, coupled with the irresponsibility of doctors, so it is only right that some doctors have been killed because of doctor-patient disputes
8	Are patients the only ones to blame for doctor-patient conflicts?

Construction of a lexicon of relevant emotions

Some of the more mature emotion lexicons studied in China are HowNet, the NTUSD of National Taiwan University, and the Dalian University of Technology Chinese Emotion Vocabulary Ontology Database. In this study, we used the Dalian University of Technology Chinese emotion lexical ontology database as the basic emotion lexicon. Additionally, the negative lexicon was taken into consideration. The Chinese emotion vocabulary ontology database of Dalian University of Technology has 27467 words. The lexical categories in the emotion vocabulary ontology are divided into seven categories: Noun, verb, adjective, adverb, network neologism, idiom, prepositional phrase, and prepositional phrase. The final vocabulary ontology divides emotions into seven categories and twenty-one subcategories: “happy”, “good”, “angry”, “sadness”, “fear”, “disgust”, and “surprise”. The sentiment intensity was classified into five classes, with 1 being the lowest and 9 being the highest. Each word corresponds to a polarity under each category of emotion, where 0 represents neutrality, 1 represents positive, 2 represents negative, and 3 represents both positive and negative. The polarity value of 2, representing pejorative, was changed to -1 in this study to facilitate subsequent calculations. Words with both positive and negative properties were manually judged to be positive or negative based on the context of each comment. We modified the polarity and intensity of the comments to better calculate the sentiment value by changing the polarity*intensity to weights. The formula for the emotion word weights of the vocabulary was:

where, s(w) denotes the emotional word weight of the lexicon, v(w) denotes the emotional intensity of the lexicon, and p(w) denotes the emotional polarity of the lexicon. An example of the basic sentiment lexicon is shown in Table 2. In addition, we used a negative lexicon to further improve the accuracy of commenting on sentiment judgements. The negation lexicon included a total of 71 words. Negative words have the opposite meaning of the original sentence, so the weight of the negative words was set to -1. An example of the negation lexicon is shown in Table 3.

Table 2 Examples of basic emotion lexicons.

Word	Lexical category	Intensity	Polarity	Weight
Dependency	Verb	1	0	0
Careful	Adjective	3	1	3
Eccentric	Adjective	1	-1	-1
Guffaw	Verb	5	-1	-5

Table 3 Negative lexicon examples.

Word	Weight	Quantity
Didn’t, no, not, couldn’t, rarely, never, terminated, missing, never	-1	71

Extracting the number of emotion words in text using the Python algorithm

Pandas were used to read the data, and the data were imported into the Dalian University of Technology Chinese emotion vocabulary ontology database using “word”, “lexical category”, “lexical number”, “lexical number”, “emotion classification”, “intensity”, and “polarity”. The list of emotion words was collated according to the seven major emotion divisions, and the number of emotion words in each of the collected comments was statistically counted based on the seven emotion divisions of “happy”, “good”, “angry”, “sadness”, “fear”, “disgust”, and “surprise”.

Sentiment values are manually calculated for the extracted comment data

The comment data was analyzed using the sentiment lexicon and Python was to extract the number of emotion words. We subsequently calculated the sentiment value for each text comment manually. S(w) denotes the weight of the emotion word, O denotes the sentiment value in a single sentence, and S denotes the sentiment value of all the sentences in a comment combined. An example of a specific situation in a comment is shown below. Because the sentiment value of each comment depends on the weight of the emotion words, we propose that the sentiment value of a single sentence be calculated as:

where K is the number of emotion words in the sentence.

When an emotion word followed a negative word, the emotion word was interpreted to express the opposite meaning, and the sentiment value of the comment was related to the number of negative words. The sentiment value changed when the number of negatives was odd, and the polarity of the emotion words was reversed. When the number of negatives was even, the polarity of the emotion words and the sentiment value were in agreement. Therefore, when there were negatives, the equation for the emotion words was:

where n is the number of negatives in the sentence. In summary, if the above possible scenarios were present in a comment, the formula for calculating the comment sentiment value was:

Statistical analysis

The Python language was used to extract the number of emotion words in the comments of the collected incidents based on the sentiment lexicon. The sentiment score of the comments was calculated by manually reviewing the number of emotion words in the extracted comments. SPSS 26.0 statistical software (IBM Corp., Armonk, NY, United States) was used to explore whether there were differences in the number of words and comment scores for each emotion across the four different levels of incidence using the Kruskal-Wallis H test. Correlation analysis was performed to examine the association between incident levels, the number of words in each emotion, and comment scores. Multiple stepwise regression analysis were used to explore the factors influencing the comment scores. Differences were considered to be statistically significant at P < 0.05.

RESULTS

Python was used to count the number of emotion words in each comment that was collected. Positive = happy + good + surprise, and negative = anger + sadness + fear + disgust. The results are shown in Table 4. The number of emotion words were extracted from 3655 comments related to 30 doctor-patient disputes. The 1608 comments for level-1 incidents (verbal conflicts) contained a total of 58 emotion words. These words were distributed across seven categories: “anger” (3.6%), “disgust” (68.7%), “fear” (8.0%), “sadness” (9.9%), “surprise” (1.1%), “good” (54.6%), and “happy” (9.6%), with 1104, 128, 159, 17, 878, and 155 words in each category, respectively. The 486 comments for level-2 incidents (physical conflicts) contained 11 emotion words in the anger, 238 in the disgust, 13 in the fear, 39 in the sadness, 1 in the surprise, 228 in the good, and 31 in happiness categories. The corresponding proportions were 2.3%, 49.0%, 2.7%, 8.0%, 0.2%, 46.9% and 6.4%. The number of emotion words in the 945 comments about level 3 incidents (injuries) was 24 for anger, 494 for disgust, 44 for fear, 130 for sadness, 11 for surprise, 481 for good, and 111 for happy, corresponding to 2.5%, 52.3%, 4.7%, 13.8%, 1.2%, 50.9%, and 11.7%, respectively. The number of emotion words in the 616 comments about level 4 incidents (fatalities) was 21 was for anger, 423 for disgust, 47 for fear, 125 for sadness, 3 for surprise, 427 for good, and 82 for happiness, corresponding to 3.4%, 68.7%, 7.6%, 20.3%, 0.5%, 69.3%, and 13.3%. Therefore, it was concluded that there were more “disgust” and “good” emotion words than other emotion words in all doctor-patient disputes.

Table 4 Example of partial results of Python extracting the number of emotion words corresponding to the matrix data.

Serial number	Comments	Length	Positive	Negative	Anger	Disgust	Fear	Sadness	Surprise	Good	Happy
1	It ends 80% of the time with an apology, the nurse showing under-standing, and ending up at home for the rest of the year	14	1	0	0	0	0	0	0	1	0
2	If they can’t be punished severely; similar things will keep happening	10	0	0	0	0	0	0	0	0	0
3	Looking at the injuries, it feels like they may not be able to return home next New Year’s Eve, either	10	0	1	0	1	0	0	0	0	0
4	Why? I wouldn’t accept an apology if I were you	10	1	0	0	0	0	0	0	1	0

Results of manual sentiment value calculation on extracted comment data

Based on the equations for the sentiment value of comments in different situations, we manually assigned sentiment values to the collected comments on doctor-patient dispute incidents. In the process of manually calculating the assigned values, we corrected the shortcomings of the machine algorithm and manually changed or eliminated emotion words that were analyzed incorrectly. For example, the word “Caineng” in the Chinese expression is used as an adjective meaning “knowledge and ability” and as an adverb indicating that one is able to do something. Therefore, after the machine recognized it, we manually verified and corrected it based on the context of the comment. The final scores of the comments for the different levels of incidents were follows: The total score of the 1608 comments about level 1 incidents (verbal conflicts) was -2651. The score of the 486 comments about level 2 incidents (physical conflicts) was -333. The score of the 959 comments about level 3 incidents (injuries) was -1061, and the score of the 616 comments about level 4 incidents (fatalities) was -642. The results of the manual assignment of comment score data are shown in Table 5.

Table 5 Example of partial results for the corresponding matrix data after commenting on the manual assignment.

Serial number	Comments	Length	Positive	Negative	Anger	Disgust	Fear	Sadness	Surprise	Good	Happy	Comment score
1	Excessive people are inexplicable	5	0	2	0	2	0	0	0	0	0	-8
2	The point is, you can’t leave an emergency room unattended	7	0	0	0	0	0	0	0	0	0	0
3	What is an emergency? What is an urgent case? The family’s in a hurry, the doctor doesn’t panic	12	1	1	0	1	0	0	0	0	1	-3
4	A few doctors have some really bad attitudes	6	0	1	0	1	0	0	0	0	0	-3
5	Doctor, you have to know what an emergency is. Every second counts, understand?	14	1	0	0	0	0	0	0	1	0	5

SPSS 26.0 was used for analysis of the comment data because they were found to be nonnormally distributed (P < 0.05). The nonparametric Kruskal-Wallis H test was used to compare differences between groups of variables. The results of this analysis are shown in Table 6.

Table 6 Kruskal-Wallis H test.

Variable	Comparison of incident levels		SE	P value
Positive	Group differences	14.822		0.002
	2 vs 1	113.738	48.270	0.018
	2 vs 4	-177.112	56.576	0.002
	3 vs 1	83.643	38.222	0.029
	3 vs 4	-147.016	48.289	0.002
Negative	Group differences	35.397		0
	2 vs 1	218.660	50.200	0
	2 vs 4	-281.295	58.838	0
	3 vs 1	140.349	39.750	0
	3 vs 4	-202.985	50.219	0
Anger	Group differences	1.779		0.619
Disgust	Group differences	27.043		0
	2 vs 4	-165.943	56.332	0.003
	2 vs 1	182.248	48.062	0
	3 vs 4	-143.057	48.080	0.003
	3 vs 1	159.362	38.057	0
Fear	Group differences	21.502		0
	2 vs 4	-78.824	25.924	0.002
	2 vs 1	84.332	22.118	0
	3 vs 4	-51.862	22.127	0.019
	3 vs 1	57.370	17.514	0.001
Sadness	Group differences	44.747		0
	2 vs 3	-81.824	31.089	0.008
	2 vs 4	-187.128	33.791	0
	1 vs 3	-55.240	22.829	0.016
	1 vs 4	-160.543	26.391	0
	3 vs 4	-105.304	28.841	0
Surprise	Group differences	5.539		0.136
Good	Group differences	14.396		0.002
	3 vs 1	98.017	37.044	0.008
	3 vs 4	-147.035	46.800	0.002
	2 vs 1	95.630	46.782	0.041
	2 vs 4	-144.648	54.831	0.008
Happy	Group differences	9.647		0.022
	2 vs 3	-57.709	28.480	0.043
	2 vs 4	-94.158	30.956	0.002
	1 vs 4	-49.162	24.176	0.042
Comment score	Group differences	14.206		0.003
	1 vs 3	-81.270	40.833	0.047
	1 vs 4	-104.633	47.204	0.027
	1 vs 2	-177.166	51.567	0.001

1: Level 1 incidents (verbal conflicts); 2: Level 2 incidents (physical conflicts); 3: Level 3 incidents (injuries); 4: Level 4 incidents (fatalities).

As shown in Table 6, the variables positive, negative, disgust, fear, sadness, good, and happy, as well as the comment scores at the four levels of incidents, all show a significant difference between the groups of the levels (P < 0.05). This means that there was a significant difference between at least 2 groups of incident comments at the four different levels of incident comments. For the positive variable, differences observed between level 2 incidents and level 1 incidents, between level 2 incidents and level 4 incidents, between level 3 incidents and level 1 incidents, and between level 3 incidents and level 4 incidents were significant. For the negative variable, significant differences were observed between level 2 incidents and level 1 incidents, between level 2 incidents and level 4 incidents, between level 3 incidents and level 1 incidents, and between level 3 incidents and level 4 incidents. For the disgust variable, significant differences were observed between level 2 and level 4 incidents, level 2 and level 1 incidents, level 3 and level 4 incidents, and between level 4 and level 1 incidents. For the fear variable, significant differences were observed between level 2 and level 4 incidents, between level 2 and level 1 incidents, between level 3 and level 4 incidents, and between level 3 and level 1 incidents. For the sadness variable, significant differences were observed between level 2 and level 3 incidents, between level 2 and level 4 incidents, between level 1 and level 3 incidents, between level 1 and level 4 incidents, and between level 3 and level 4 incidents. For the good variable, significant differences were observed between level 3 and level 1 incidents, between level 3 and level 4 incidents, between level 2 and level 1 incidents, and between level 2 and level 4 incidents. For the happy variable, significant differences were observed between level 2 and level 3 incidents, between level 2 and level 4 incidents, and between level 1 and level 4 incidents. For comment scores, significant differences were observed between level 1 and level 3 incidents, between level 1 and level 4 incidents, and between level 1 and level 2 incidents.

Because the data were continuous and nonnormally distributed, we performed Spearman’s correlation analysis to examine correlations. A binary variable correlation matrix was constructed to show the associations between incident levels, the number of words for each sentiment, and comment scores (Table 7). The results of the correlation analysis revealed that incident level was negatively correlated with the number of emotion words for disgust and positively correlated with both the number of words for sadness and comment scores. The positive variable was positively correlated with negative disgust, sadness, surprise, good, and happy emotion word counts as well as comment scores. The negative variable was positively correlated with the number of emotion words for anger, fear, sadness, good, and happiness and negatively correlated with comment scores. Comment scores were positively correlated with incident level and the number of emotion words for positive, good, and happy and negatively correlated with the number of emotion words for negative, anger, disgust, fear, and sadness.

Table 7 Spearman’s correlation analyses between incident levels, number of words for each sentiment and comment scores.

Variable	Incident level	Positive	Negative	Anger	Disgust	Fear	Sadness	Surprise	Good	Happy	Comment score
Incident level	1.000
Positive	0	1.000
Negative	-0.010	0.101b	1.000
Anger	0	0.028	0.197b	1.000
Disgust	-0.0340a	0.104b	0.842b	0.058b	1.000
Fear	-0.027	0.025	0.262b	-0.020	0.048b	1.000
Sadness	0.086b	0.047b	0.366b	0.046b	0.072b	0.023	1.000
Surprise	-0.008	0.122b	0.015	0.020	0.021	-0.023	-0.003	1.000
Good	-0.007	0.905b	0.094b	0.026	0.094b	0.027	0.047b	0.039a	1.000
Happy	0.027	0.390b	0.069b	0.026	0.073b	0.009	0.042a	0.036a	0.092b	1.000
Comment score	0.036b	0.223b	-0.470b	-0.070b	-0.488b	-0.064b	-0.099b	0.014	0.228b	0.041b	1.000

^aP < 0.05 (two-tailed).

^bP < 0.05 (two-tailed).

Multiple stepwise regression analysis were conducted using the comment score as the dependent variable and the incident level and the number of words for each sentiment as independent variables. The results are shown in Table 8. The number of emotion words for negative, anger, disgust, fear, and sadness directly influenced the comment score, and the number of emotion words for incident, positive, surprise, good, and happy indirectly influenced the comment score.

Table 8 Linear regression analysis.

Variable	Step 1		Step 2
	β	SE	β	SE
Incident level	0.020	0.059
Positive	0.452	2.372
Negative	-1.084c	0.648	-1.009c	0.730
Anger	0.136c	0.743	0.131c	0.838
Disgust	0.304b	0.644	0.302b	0.725
Fear	0.193c	0.667	0.189c	0.750
Sadness	0.295c	0.632	0.300c	0.713
Surprise	-0.035	2.475
Good	-0.017	2.371
Happy	-0.094	2.366
Adjusted R2	0.460		0.309
F	312.274c		327.572c

^bP < 0.01.

^cP < 0.001.

DISCUSSION

This study examined the emotional words in the comments on media reports of doctor-patient disputes at various levels of severity and found that “good” and “disgust” were the most common emotions. Specifically, words that indicated ‘disgust’ accounted for greatest proportion of emotional words, followed by ‘good’ emotional words, and emotional words that indicated other meanings accounted for a smaller proportion of emotional words. The results show that in comments on doctor-patient disputes, the public evaluated the good and bad aspects of both sides of the incident based on the content of the report and commented on the incident with a certain tendency according to the specific situation. For the emotional words that express “disgust” in the comments of media reports, the public’s evaluation of the object of concern was mainly focused on medical institutions (hospitals), the main fault side of doctor-patient disputes. Comments on the relevant content mainly focused on the following points: Criticism of. the negligence and malpractice of the medical profession; questioning or denouncing the regularity and safety of hospitals and the legitimacy of diagnostic and therapeutic means; condemning the violence of patients against the medical profession; harshly reproaching and resenting those who injured medical practitioners; demanding that the relevant public prosecutors and law enforcement agencies be severely punished; accusing the healthcare system of being inadequate and criticizing the shortcomings of the healthcare system. For the emotional words that expressed “good” in the comments on the media coverage of the incident, the public’s evaluation was more focused on solidarity with the injured side in the specific doctor-patient dispute incident. The main content included a positive emotional tendency to call for respect for the facts, respect for both doctors and patients, and praise of the outcome of the incident. Our findings are consistent with those of Zhao et al[25].

In this study, for the variables positive, negative, disgust, fear, sadness, good and happy, as well as the comment scores for the four levels of incidents, the difference between the groups was significant (P < 0.05); that is, there was a significant difference between at least 2 groups of incident comments at the four different levels of incident comments. Based on the results of the Spearman correlation analysis, it can be concluded that the comment score is positively correlated with the number of words of emotion, such as incident level, positive, good and happy, and negatively correlated with the number of emotion words, such as negative, anger, disgust, fear and sadness. Positive, good, and happy all denote positively inclined (positive) emotional significance, and negative, anger, disgust, fear, and sadness all denote negatively inclined (derogatory) emotional significance, indicating that the higher the comment score, the stronger the positive inclination of the comment and the weaker the negative inclination. The incident level was negatively correlated with the number of words of disgust and positively correlated with the number of words of sadness and the comment scores, which indicates that more severe the doctor-patient dispute was, the more serious the casualties were. The number of the disgust emotion words in the comments were relatively lower, the number of sadness emotion words were relatively higher, and the scores of the incident comments were higher. A possible. reason was a higher incidence level was associated with more severe injury. the comments about such incidents had fewer words that were biased remarks of a bad nature and more emotional, such as expressing regret and sadness for the injured side. Another possible reason is that with the continuous development and progress of information networks and social media, fragmentation of information and the emergence of tragedies caused by cyber violence, cultural quality continues to improve, people view issues related to doctor-patient incidents with greater emphasis on factual content, and their comments are more peaceful. For example, a level 4 incident was “Net rumors that a doctor in Chongqing was stabbed by a member of patient’s family while on duty and died.” Comments on this incident included “Doctors cause heartache for others, and doctors with medical ethics and medical style are admired and respected,” “Saved countless lives and died such a tragic death! Sad!” The content of their comments reflected the public’s pity and grief for the casualties in the incident. The results of this study confirmed Hypothesis two. There was a positive correlation between the level of the incident and the number of words of sadness (i.e. the tendency of the public to feel pity and sadness for the victim) in the comments. The incident level was positively correlated with comment scores and the positive variable was positively correlated with a number of words such as negative, disgust, sadness, surprise, good, and happy. The negative variable was positively correlated with a number of words such as anger, disgust, fear, sadness, good, and happy because positive = happy + good + surprise, negative = anger + sadness + fear + disgust. Therefore, as the severity of the incident increased, the severity and influence of the faulty behavior of the incident bearer increased, the heated nature of public comments about doctor-patient disputes intensified, the number of comments published increased, and the proportion of positive and negative emotional tendencies in the comments increased with the increase in the number of comments. Another possible reason is that in the early stage of the media reports, which involves the comments of affected public figures, the severity of the incident, and the public level of education and cultural quality, a more severe doctor-patient dispute may trigger the majority of the public to make overly radical remarks that condemn the behavior of the doctor or patient. In such situations, there is a strong tendency towards negative emotions in the comments. With continuous reporting of the incident, the public will come to view the incident more rationally and participate in the discussion after the outcome of the incident is announced. The negative emotions in the comments then changed to positive emotions and the public expressed more praise for the outcome of the incident and expectations for the future development of the doctor-patient relationship. Thus, the proportion of comments expressing positive emotion was increased and the final score of the incident comments was higher. The study results confirm hypothesis 1 that the comment scores of doctor-patient dispute incidents were related to the level of the incident (i.e. the severity of the damage caused by the behavior of the responsible at-fault side).

Based on the results of multiple stepwise regression analysis, it can be concluded that the number of negative emotion words, such as anger, disgust, fear, and sadness directly influenced the comment score, and the number of emotion words for incident level, positive, surprise, good, and happy may have indirectly influenced the score. The intensity of the negative emotions brought about by negative emotion words and their direct impact on the incident comments can be determined. The results of this study confirm hypothesis 3: The intensity of negative emotions brought about by negative emotion words in comments has a direct effect on incident comment scores and opinion influence. The reasons for this intense negative emotional sentiment may be as follows. First, the doctor-patient relationship is often tense, and with the rise of social media and networking platforms, there is a subtle influence on the content of comments. According to a survey by the China Medical Doctors Association, an average of 66 medical disputes occurs per hospital per year in mainland China, and more than 30% of doctors have experienced violence from patients and their families. Secondly, media reports about doctor-patient disputes at all levels and comments from public figures can influence the orientation of public opinion towards such disputes on online social platforms. This includes emotional expressions that come from various sources such as ordinary public users, official news media, social media, and influential online bloggers. When subjects express their views and facts regarding doctor-patient disputes, they may often have personal, subjective impressions, which could include intense emotional tendencies such as anger, disgust, fear, and sadness. Additionally, they may excessively criticize various parties involved, including the medical side, patient side, hospital side, or other relevant institutions. In a level 2 incident a “woman experienced infusion pain, argued with hospital staff angrily, smashed items in the infusion room, and slapped the security guard immediately after the attack.” The comments included the following text: “Do not know the sky and the earth. I truly think that who what spoiled you” and “Habitually arrogant and domineering, this time she was taught a well-deserved lesson.” The content of the comments was based on one side of the argument and blamed the other side, and thus did not have a rational view of the incident. Therefore, we suggest introducing laws and regulations that follow a top-level design in order to create a favorable social environment for improving tense doctor-patient relationships[26]. Second, the verbal comments of the media should be regulated at all levels, with guidance provided by public figures and the media. Third, enhancing doctors’ education level, boosting their confidence, and earning patients’ trust, alleviating their work pressure, and nurturing their sense of professional identity and social responsibility are crucial objectives[27].

CONCLUSION

There was a significant difference between the comment score and the number of partial emotion words for the comments about different levels of doctor-patient disputes. The comment score was positively correlated with the level of the incident and the number of positive emotion words such as, good, and happy and negatively correlated with the number of negative emotion words such as, anger, disgust, fear, and sadness. As the severity of the incident increased, the number of words with positive emotional tendencies increased. Furthermore, as the number of reports about the incident results increased, the public comments became more rational, the number of words with negative emotional tendencies decreased, the negative tendency of the comments gradually changed to a positive tendency, and the scores of the comments increased.

The number of emotion words for negative, anger, disgust, fear, and sadness directly influenced comment scores, and the number of emotion words for incident, positive, surprise, good, and happy may have indirectly influenced the comment scores. The comment scores of incidents were more strongly influenced by emotional words related to negative emotional tendencies.

By applying text sentiment analysis to specific cases of doctor-patient disputes, we extracted the sentiment of comments made by the public. Understanding the emotional differences between doctors and patients has practical significance for establishing harmonious doctor-patient relationships and for guiding the development of positive public opinion on doctor-patient incidents in society.