Social media and suicide in social movements: a case study in Hong Kong

Abstract

Research has indicated that excessive and sensationalized suicide reporting can lead to copycat suicides, especially when deaths involve well-known people. Little is known, however, about the impact of the reporting of suspected protestor suicide deaths during social unrest, particularly in an age of social media. In June 2019, the most substantial social unrest in Hong Kong since its handover in 1997 was triggered by the proposed Anti-Extradition Law Amendment Bill (Anti-ELAB). The social unrest subsided when Hong Kong and many parts of the world were hit by Covid-19 and very strict quarantine measures were imposed on crowd gatherings in Hong Kong at the end of January 2020. A number of reported suicides and deaths of undetermined cause took place during this 8-month period that received considerable attention. To better understand the possible effects of these highly publicized deaths, we examined media reports of suspected suicide cases before, during and after the protest period, as well as topics of suicide-related threads and their replies in social media forums. We found no clear evidence of increased rates of suicide as a result of these incidents, or during the protest period; however, it is suggested that certain narratives and attention surrounding the suspected suicides and undetermined deaths may have contributed to collective emotions such as sadness and anxiety. Some implications for misinformation (intentionally or un-intentionally) and mitigation of suicide risk during social unrest are discussed.

Appendices

Appendix 1

Meltwater is a software as a service (SaaS) solution and online media monitoring company, which provides both online and printed media data. URL: Meltwater: Media Monitoring & Social Listening Platform.

Appendix 2

Suicide related keywords used in collecting suicide news and discussions from Meltwater and social forums.

Chinese	English	Chinese	English
自殺	Suicide	跳樓	Jump off from a building
割腕	Cut wrist	燒炭	Charcoal burning
溺水	Drowning	吊頸	Hanging
膠袋笠頭	Choking to death with plastic bag	上吊	Hanging
仰藥	Taking drugs	跳橋	Jump off from a bridge
輕生	Commit suicide	墜軌	Jump off from a rail
自插	Self-plug	墜崖	Falling off a cliff
自盡	Suicide	燒煤氣	Burning gas
遺書	Suicide note	服毒	Take poison
尋死	Seek death	自焚	Self-immolation
窒息	Suffocation	割喉	Cut throat
自縊	Hanging	墮海	Jump into the sea

Appendix 3

Martyrdom related keywords used for filtering related posts.

Chinese	English
烈士	Martyr
義士	Righteous
為香港而死	Die for Hong Kong
捨命	Sacrifice oneself
犧牲者	Martyr
犧牲	Sacrifice
英雄	Hero

Appendix 4

Foul play related keywords used for filtering related posts.

Chinese	English
強迫自殺	Forced suicide
被殺	Killed
謀殺	Murdered
暗殺	Assassinated
非自殺	Fake suicide
非自然死亡	Unnatural death
冤死	Unjustly died

Appendix 5

Sentiment analysis method

We used LIWC software to calculate the sentiment score for each post. LIWC was widely used in relevant sentiment tasks, including suicide risk and emotional distress [36], and emotional dynamics in online social movement [37]. The basic logic behind LIWC is they have created a dictionary about different emotions and topics, and it can count the number of keywords of different categories appeared in the posts/articles and calculated a percentage based on the length of the posts/articles. The score can be used in downstream tasks like classification and correlation analysis. There are five categories of emotions predefined in LIWC, including positive emotion, negative emotion, sad, anger and anxiety. We also created an overall sentiment score based on the positive score and negative score to represent the general sentiment of one post. The overall sentiment scores

$${\text{OSS}} = \frac{{ {\text{Positive sentiment score}}}}{{ {\text{Positive sentiment score}} + {\text{Negative sentiment score}}}},$$

whereas 0.5 were used to represent the posts with both 0 positive score and 0 negative score. For anxiety, anger and sad, the range of the score is 0–100, the larger the score, the more tendency towards this emotion whereas the range of overall sentiment score (OSS) is between 0 and 1, and the post is more positive if SSO is closer to 1, and more negative if closer to 0. There are two steps to get the sentiment score. First, all messages were segmented into a set of keywords with the help of jieba.¹ In order to better fit Hong Kong cultural context, colloquial words and local slangs were added in jieba dictionary to optimize the word segmentation. Second, all punctuation, emojis were removed before calculating the emotion score.