LDA topic modeling shows the significant increase in the media topics of the conflict frames implicating schizophrenia patients in crimes regardless of the schizophrenia names
LDA topic modeling was performed on a dataset divided into news articles by period (before/after the revision of disease name) and by disease names (‘Jungshinbunyeolbyung’ for mind split disorder and ‘Johyeonbyung’ for attunement disorder). We tried to utilize the perflexity values to determine the analyzable number of topics, but the values decreased monotonically in all sections. Thus, in this study, we determined the analyzable number of topics as 30, which were found appropriate to interpret due to high similarity between major keywords in the same topic and low similarity between topics, after performing topic modeling by setting the number of topics to 10, 20, 30, and 40. The authors extracted 20 keywords per topic, and annotated each of the topics based on the association between keywords within the topics. The resulting topics were visualized using the pyLDavis library in Python (Figure 1). The annotated topics were further estimated for their individual significance in the dataset from the perspective of media frames, by calculating the proportion of each of the topics within the dataset (Table 1). The weight of the media frame was calculated by totaling the values of topics belonging to the frame, where a value of topic was calculated by dividing the total sum of ratio scores (“theta”) [26] reflecting the significance of topic within an article by the total number of articles [27].
Topics were classified into four different media frames as follows: medical frames (symptoms, research, causes, and treatments), conflict frames (crime), policy frames (policy and welfare), and neutral frames (anecdotes and art). To evaluate the classifications, two independent psychologists were invited to engage, and they qualitatively confirmed the reliability of the suggested terms representing the different media frames [28]. The inter-investigator agreement scores in both loose and strict matches were calculated by dividing the number of consistently agreed topics by all the topics. To further validate the inter-investigator agreement scores, we additionally adopted Krippendorff’s alpha [29] to correct any potential biases arising from the redundancy of media frames and participating number of investigators. The scores were remarkably high, regardless of the metrics (strict match=0.9, loose match=0.99). The Krippendorff’s alpha scores were 0.88, and thus considered reliable [29]. In particular, the loose match scores were close to 1, which shows that disagreement for the classification of media frames rarely occurred between the investigators. To note, the 'politics' topic of dataset exploited for Figure 1b,c was not considered in the analysis because articles irrelevant to schizophrenia such as 'today's popular news' were collected at the side of the body of the articles.
Figure 1 and Table 1 commonly show that medical frames were the most common in news articles of schizophrenia (i.e. ‘Jungshinbunyeolbyung’ as mind split disorder) before the revision of the disease name, followed by policy, conflict, and neutral frames. On the other hand, after the revision of the disease name, the conflict frames were the most frequently reported in the articles of schizophrenia (i.e. ‘Jungshinbunyeolbyung’ as mind split disorder), followed by medical, policy, and neutral frames. Lastly, in the articles of the renamed schizophrenia (i.e. ‘Johyeonbyung’ as attunement disorder), the conflict frames were also overwhelmingly reported, then followed by medical, policy, and neutral frames. These results indicate that there was a significant increase in the conflict frames in the articles of schizophrenia after the revision of the disease name compared to those before the revision. Crime articles of the conflict frames, which accounted for the smallest proportion of all articles before the revision of the schizophrenia name, increased by about five times after the revision of the disease name. This seems to be due to a series of media reports of violent crimes of schizophrenic patients shortly after the revision of the disease name. For instance, a distinct topic frame, which was marked with an asterisk in the Figure 1b,c, was associated with the Gangnam station homicide case committed by a schizophrenia patient in 2016. On the other hand, the proportion of medical frames decreased by more than a half after the revision of the disease name. This suggests that medical knowledge and information such as symptoms, research, and treatment of schizophrenia were no longer regarded as a crucial social issue after the revision of the disease name, nonetheless those used to be a major topic before the revision of the disease name.
Table 1. Proportion of topics for schizophrenia by media frames
Schizophrenia name
|
Media frame
|
Conflict
|
Medical
|
Policy
|
Neutral
|
Miscellaneous (Politics)
|
Total
|
Mind split disorder before renaming
|
315 (11.5%)
|
1709 (62.3%)
|
529 (19.3%)
|
189 (6.9%)
|
-
|
2743 (100%)
|
Mind split disorder after renaming
|
1859 (59.7%)
|
786 (25.25%)
|
221 (7.1%)
|
104 (3.35%)
|
143
(4.6%)
|
3114 (100%)
|
Attunement disorder
|
1611 (52.5%)
|
693 (22.6%)
|
313 (10.2%)
|
252 (8.2%)
|
199
(6.5%)
|
3068 (100%)
|
TF-IDF weight analysis indicates that media frames’ focus has shifted from the medical themes to the conflict themes on schizophrenia regardless of the revision of the disease name
The larger the TF-IDF weight value of a word, the more it means that it contains a key message within the article containing the word. Table 2 shows the words with the top 20 weight values of TF-IDF for each dataset. The media frames were further analyzed for the context of contents in the articles containing the top 20 words (Table 3).
In the analysis of articles on schizophrenia (mind split disorder) before the revision of the disease name, words of medical frames such as "Medical science”, “Research”, "Medication dose”, "Psychology”, and “Psychiatry" were found to present high TF-IDF values among the top 20 words. There was only a single word with a high TF-IDF value related to conflict frames, which is "Electronic anklet”, a forceful legal procedure for sex offenders in Korea. This shows that the public was interested in medical aspects of schizophrenia rather than negative stereotype of the disease before the revision of the disease name. However, in the analysis of articles on schizophrenia (mind split disorder) after the revision of the disease name, the words with high TF-IDF values of medical frames decreased by more than a half compared to those before the revision. In contrast, words with high TF-IDF values of conflict frames accounted for 7 of the top 20 (35%) after the revision, compared to the 1 of the 20 (5%) before the revision. Similarly, in the analysis of words on schizophrenia (attunement disorder) after the revision of the disease name, there were words of conflict frames such as “Police", "Police station", "Murder", and "Crime” that occupied a considerable portion of the top 20 words with high TF-IDF values (7 of the 20, 35%). In contrast, there was a decrease in the number of words related to medical frames after the renaming (5 of the top 20 words, 25%), compared to before the renaming (11 of the top 20 words, 55%). These findings suggest that the public interest on schizophrenia has shifted from the medical aspects to the conflict frames, regardless of the disease renaming.
Table 2. Top 20 words with TF-IDF weight values for schizophrenia
|
Mind split disorder before the revision
|
Mind split disorder after the revision
|
Attunement disorder after the revision
|
1
|
Medical science
|
0.65
|
Symptoms
|
0.8
|
Police
|
0.65*
|
2
|
Negotiations
|
0.57
|
Hospitalization
|
0.59
|
Counseling
|
0.6
|
3
|
Research
|
0.57
|
Police
|
0.58*
|
Police station
|
0.51*
|
4
|
Stress
|
0.53
|
Crime
|
0.56*
|
Psychology
|
0.5
|
5
|
Medication dose
|
0.52
|
Mentality
|
0.53
|
Research
|
0.49
|
6
|
Mentality
|
0.51
|
Woman
|
0.48
|
Schizophrenia
|
0.49
|
7
|
Military service
|
0.5
|
Investment
|
0.46
|
Sentence of imprisonment
|
0.47*
|
8
|
Psychology
|
0.48
|
Trial
|
0.44*
|
Parent
|
0.47
|
9
|
Suicide
|
0.43
|
Parent
|
0.43
|
Mentality
|
0.46
|
10
|
Psychiatry
|
0.43
|
Heredity
|
0.41
|
Crime
|
0.46*
|
11
|
Counseling
|
0.42
|
Psychology
|
0.41
|
Drug
|
0.45
|
12
|
Child
|
0.38
|
Psychopath
|
0.4*
|
Mother
|
0.42
|
13
|
Company
|
0.34
|
Police station
|
0.39*
|
Woman
|
0.4
|
14
|
Clinic
|
0.34
|
Criminal
|
0.35*
|
Psychosis
|
0.39
|
15
|
Mind
|
0.29
|
Allegations
|
0.32*
|
Trial
|
0.37*
|
16
|
Delusion
|
0.28
|
Life
|
0.32
|
Abandonment
|
0.37
|
17
|
Electronic anklet
|
0.27*
|
Dementia
|
0.28
|
Homelessness
|
0.32
|
18
|
Improvement
|
0.27
|
Descendant
|
0.2
|
Murder
|
0.31*
|
19
|
People
|
0.25
|
Null and invalid
|
0.14*
|
Allegations
|
0.28*
|
20
|
Culture
|
0.24
|
Designation
|
0.14
|
Daegu
|
0.19
|
Asterisk (*) indicates the word belonging to the conflict frames (crimes).
Table 3. Media frames including the top 20 words with TF-IDF weight values for schizophrenia
Schizophrenia name
|
Media frame
|
Conflict
|
Medical
|
Policy
|
Neutral
|
Sum
|
Mind split disorder before renaming
|
1
|
11
|
3
|
5
|
20
|
Mind split disorder after renaming
|
7
|
5
|
2
|
6
|
20
|
Attunement disorder
|
7
|
5
|
0
|
8
|
20
|
Association between the media frames containing negative aspects of schizophrenia and the nationwide hospital admission patterns of schizophrenia patients
The estimation results of regression model were presented in Table 4. The results in Table 4 show that when there was more news coverage of crimes committed by an individual with schizophrenia, the number of schizophrenia patients who admitted to psychiatric hospitals increased compared to that in the preceding month (B = 0.32, p = 0.035). Not much difference between male (B = 0.31, p = 0.043) and female patients (B = 0.34, p = 0.041) was found, albeit there was a slightly larger increase among female patients.
Table 4. Estimation results of regression model
Variables
|
Total change
|
Male change
|
Female change
|
B
|
SE
|
B
|
SE
|
B
|
SE
|
constant
|
-0.17
|
0.99
|
-0.05
|
0.98
|
-0.29
|
1.02
|
# homicide_articles_t
|
0.32*
|
0.17
|
0.31*
|
0.16
|
0.34*
|
0.17
|
# homicide_articles_t-1
|
-0.15
|
0.17
|
-0.15
|
0.16
|
-0.15
|
0.17
|
R 2
|
0.78
|
0.78
|
0.77
|
F (21, 85)
|
14.05**
|
14.64**
|
13.25**
|
Note: N = 107, * p < 0.05, ** p < 0.01. The results of the dummy variables regarding month and year effects were omitted due to space constraints.
Negative media coverages before and after the Gangnam station homicide case committed by a schizophrenia patient
May 17 is the date when the Gangnam homicide happened. Accordingly, most of the articles published in May were about the homicide case. But, for the purpose of this study, we excluded articles about the Gangnam homicide published in other months of the year. Therefore, the values for the months besides May indicate the number of news articles covering negative aspects of schizophrenia but not about the Gangnam homicide case. As the Figure 2 shows there tended to be fewer articles covering the negative aspects of schizophrenia before the homicide case when analyzing the term ‘attunement disorder’, while there were more articles about the negative aspects of schizophrenia when analyzing the term ‘mind split disorder’ after the case. But after the homicide case, the tendency has changed: there were more articles covering the negative aspects of the illness after the homicide when analyzing the term ‘attunement disorder’, while there were fewer articles about the negative aspects of the illness when analyzing the term ‘mind split disorder’ after the case (Table 5).
Table 5. Average numbers of negative media coverages for schizophrenia before and after the Gangnam station homicide case
|
Gangnam station homicide case
(Month of 2016)
|
From January to April
(before the homicide case)
|
From June to December
(after the homicide case)
|
P value from Student’s two-sided t-test
|
Attunement disorder
|
0.5
|
28.43
|
0.017
|
Mind split disorder
|
12.5
|
5.43
|
0.134
|